PostgreSQL Source Code  git master
xlog.c File Reference
#include "postgres.h"
#include <ctype.h>
#include <math.h>
#include <time.h>
#include <fcntl.h>
#include <sys/stat.h>
#include <unistd.h>
#include "access/clog.h"
#include "access/commit_ts.h"
#include "access/heaptoast.h"
#include "access/multixact.h"
#include "access/rewriteheap.h"
#include "access/subtrans.h"
#include "access/timeline.h"
#include "access/transam.h"
#include "access/twophase.h"
#include "access/xact.h"
#include "access/xlog_internal.h"
#include "access/xlogarchive.h"
#include "access/xloginsert.h"
#include "access/xlogreader.h"
#include "access/xlogrecovery.h"
#include "access/xlogutils.h"
#include "backup/basebackup.h"
#include "catalog/catversion.h"
#include "catalog/pg_control.h"
#include "catalog/pg_database.h"
#include "common/controldata_utils.h"
#include "common/file_utils.h"
#include "executor/instrument.h"
#include "miscadmin.h"
#include "pg_trace.h"
#include "pgstat.h"
#include "port/atomics.h"
#include "postmaster/bgwriter.h"
#include "postmaster/startup.h"
#include "postmaster/walsummarizer.h"
#include "postmaster/walwriter.h"
#include "replication/origin.h"
#include "replication/slot.h"
#include "replication/snapbuild.h"
#include "replication/walreceiver.h"
#include "replication/walsender.h"
#include "storage/bufmgr.h"
#include "storage/fd.h"
#include "storage/ipc.h"
#include "storage/large_object.h"
#include "storage/latch.h"
#include "storage/predicate.h"
#include "storage/proc.h"
#include "storage/procarray.h"
#include "storage/reinit.h"
#include "storage/spin.h"
#include "storage/sync.h"
#include "utils/guc_hooks.h"
#include "utils/guc_tables.h"
#include "utils/injection_point.h"
#include "utils/ps_status.h"
#include "utils/relmapper.h"
#include "utils/snapmgr.h"
#include "utils/timeout.h"
#include "utils/timestamp.h"
#include "utils/varlena.h"
Include dependency graph for xlog.c:

Go to the source code of this file.

Data Structures

struct  XLogwrtRqst
 
struct  XLogwrtResult
 
struct  WALInsertLock
 
union  WALInsertLockPadded
 
struct  XLogCtlInsert
 
struct  XLogCtlData
 

Macros

#define BootstrapTimeLineID   1
 
#define NUM_XLOGINSERT_LOCKS   8
 
#define INSERT_FREESPACE(endptr)    (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))
 
#define NextBufIdx(idx)    (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))
 
#define XLogRecPtrToBufIdx(recptr)    (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))
 
#define UsableBytesInPage   (XLOG_BLCKSZ - SizeOfXLogShortPHD)
 
#define ConvertToXSegs(x, segsize)   XLogMBVarToSegs((x), (segsize))
 
#define RefreshXLogWriteResult(_target)
 

Typedefs

typedef struct XLogwrtRqst XLogwrtRqst
 
typedef struct XLogwrtResult XLogwrtResult
 
typedef union WALInsertLockPadded WALInsertLockPadded
 
typedef struct XLogCtlInsert XLogCtlInsert
 
typedef struct XLogCtlData XLogCtlData
 

Enumerations

enum  WalInsertClass { WALINSERT_NORMAL , WALINSERT_SPECIAL_SWITCH , WALINSERT_SPECIAL_CHECKPOINT }
 

Functions

static void CleanupAfterArchiveRecovery (TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog, TimeLineID newTLI)
 
static void CheckRequiredParameterValues (void)
 
static void XLogReportParameters (void)
 
static int LocalSetXLogInsertAllowed (void)
 
static void CreateEndOfRecoveryRecord (void)
 
static XLogRecPtr CreateOverwriteContrecordRecord (XLogRecPtr aborted_lsn, XLogRecPtr pagePtr, TimeLineID newTLI)
 
static void CheckPointGuts (XLogRecPtr checkPointRedo, int flags)
 
static void KeepLogSeg (XLogRecPtr recptr, XLogSegNo *logSegNo)
 
static XLogRecPtr XLogGetReplicationSlotMinimumLSN (void)
 
static void AdvanceXLInsertBuffer (XLogRecPtr upto, TimeLineID tli, bool opportunistic)
 
static void XLogWrite (XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
 
static bool InstallXLogFileSegment (XLogSegNo *segno, char *tmppath, bool find_free, XLogSegNo max_segno, TimeLineID tli)
 
static void XLogFileClose (void)
 
static void PreallocXlogFiles (XLogRecPtr endptr, TimeLineID tli)
 
static void RemoveTempXlogFiles (void)
 
static void RemoveOldXlogFiles (XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr, TimeLineID insertTLI)
 
static void RemoveXlogFile (const struct dirent *segment_de, XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo, TimeLineID insertTLI)
 
static void UpdateLastRemovedPtr (char *filename)
 
static void ValidateXLOGDirectoryStructure (void)
 
static void CleanupBackupHistory (void)
 
static void UpdateMinRecoveryPoint (XLogRecPtr lsn, bool force)
 
static bool PerformRecoveryXLogAction (void)
 
static void InitControlFile (uint64 sysidentifier, uint32 data_checksum_version)
 
static void WriteControlFile (void)
 
static void ReadControlFile (void)
 
static void UpdateControlFile (void)
 
static char * str_time (pg_time_t tnow)
 
static int get_sync_bit (int method)
 
static void CopyXLogRecordToWAL (int write_len, bool isLogSwitch, XLogRecData *rdata, XLogRecPtr StartPos, XLogRecPtr EndPos, TimeLineID tli)
 
static void ReserveXLogInsertLocation (int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
 
static bool ReserveXLogSwitch (XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
 
static XLogRecPtr WaitXLogInsertionsToFinish (XLogRecPtr upto)
 
static char * GetXLogBuffer (XLogRecPtr ptr, TimeLineID tli)
 
static XLogRecPtr XLogBytePosToRecPtr (uint64 bytepos)
 
static XLogRecPtr XLogBytePosToEndRecPtr (uint64 bytepos)
 
static uint64 XLogRecPtrToBytePos (XLogRecPtr ptr)
 
static void WALInsertLockAcquire (void)
 
static void WALInsertLockAcquireExclusive (void)
 
static void WALInsertLockRelease (void)
 
static void WALInsertLockUpdateInsertingAt (XLogRecPtr insertingAt)
 
XLogRecPtr XLogInsertRecord (XLogRecData *rdata, XLogRecPtr fpw_lsn, uint8 flags, int num_fpi, bool topxid_included)
 
Size WALReadFromBuffers (char *dstbuf, XLogRecPtr startptr, Size count, TimeLineID tli)
 
static void CalculateCheckpointSegments (void)
 
void assign_max_wal_size (int newval, void *extra)
 
void assign_checkpoint_completion_target (double newval, void *extra)
 
bool check_wal_segment_size (int *newval, void **extra, GucSource source)
 
bool check_max_slot_wal_keep_size (int *newval, void **extra, GucSource source)
 
static XLogSegNo XLOGfileslop (XLogRecPtr lastredoptr)
 
bool XLogCheckpointNeeded (XLogSegNo new_segno)
 
void XLogSetAsyncXactLSN (XLogRecPtr asyncXactLSN)
 
void XLogSetReplicationSlotMinimumLSN (XLogRecPtr lsn)
 
void XLogFlush (XLogRecPtr record)
 
bool XLogBackgroundFlush (void)
 
bool XLogNeedsFlush (XLogRecPtr record)
 
static int XLogFileInitInternal (XLogSegNo logsegno, TimeLineID logtli, bool *added, char *path)
 
int XLogFileInit (XLogSegNo logsegno, TimeLineID logtli)
 
static void XLogFileCopy (TimeLineID destTLI, XLogSegNo destsegno, TimeLineID srcTLI, XLogSegNo srcsegno, int upto)
 
int XLogFileOpen (XLogSegNo segno, TimeLineID tli)
 
void CheckXLogRemoved (XLogSegNo segno, TimeLineID tli)
 
XLogSegNo XLogGetLastRemovedSegno (void)
 
XLogSegNo XLogGetOldestSegno (TimeLineID tli)
 
void RemoveNonParentXlogFiles (XLogRecPtr switchpoint, TimeLineID newTLI)
 
uint64 GetSystemIdentifier (void)
 
char * GetMockAuthenticationNonce (void)
 
bool DataChecksumsEnabled (void)
 
XLogRecPtr GetFakeLSNForUnloggedRel (void)
 
static int XLOGChooseNumBuffers (void)
 
bool check_wal_buffers (int *newval, void **extra, GucSource source)
 
bool check_wal_consistency_checking (char **newval, void **extra, GucSource source)
 
void assign_wal_consistency_checking (const char *newval, void *extra)
 
void InitializeWalConsistencyChecking (void)
 
const char * show_archive_command (void)
 
const char * show_in_hot_standby (void)
 
void LocalProcessControlFile (bool reset)
 
WalLevel GetActiveWalLevelOnStandby (void)
 
Size XLOGShmemSize (void)
 
void XLOGShmemInit (void)
 
void BootStrapXLOG (uint32 data_checksum_version)
 
static void XLogInitNewTimeline (TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
 
void StartupXLOG (void)
 
void SwitchIntoArchiveRecovery (XLogRecPtr EndRecPtr, TimeLineID replayTLI)
 
void ReachedEndOfBackup (XLogRecPtr EndRecPtr, TimeLineID tli)
 
bool RecoveryInProgress (void)
 
RecoveryState GetRecoveryState (void)
 
bool XLogInsertAllowed (void)
 
XLogRecPtr GetRedoRecPtr (void)
 
void GetFullPageWriteInfo (XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
 
XLogRecPtr GetInsertRecPtr (void)
 
XLogRecPtr GetFlushRecPtr (TimeLineID *insertTLI)
 
TimeLineID GetWALInsertionTimeLine (void)
 
TimeLineID GetWALInsertionTimeLineIfSet (void)
 
XLogRecPtr GetLastImportantRecPtr (void)
 
pg_time_t GetLastSegSwitchData (XLogRecPtr *lastSwitchLSN)
 
void ShutdownXLOG (int code, Datum arg)
 
static void LogCheckpointStart (int flags, bool restartpoint)
 
static void LogCheckpointEnd (bool restartpoint)
 
static void UpdateCheckPointDistanceEstimate (uint64 nbytes)
 
static void update_checkpoint_display (int flags, bool restartpoint, bool reset)
 
bool CreateCheckPoint (int flags)
 
static void RecoveryRestartPoint (const CheckPoint *checkPoint, XLogReaderState *record)
 
bool CreateRestartPoint (int flags)
 
WALAvailability GetWALAvailability (XLogRecPtr targetLSN)
 
void XLogPutNextOid (Oid nextOid)
 
XLogRecPtr RequestXLogSwitch (bool mark_unimportant)
 
XLogRecPtr XLogRestorePoint (const char *rpName)
 
void UpdateFullPageWrites (void)
 
void xlog_redo (XLogReaderState *record)
 
void assign_wal_sync_method (int new_wal_sync_method, void *extra)
 
void issue_xlog_fsync (int fd, XLogSegNo segno, TimeLineID tli)
 
void do_pg_backup_start (const char *backupidstr, bool fast, List **tablespaces, BackupState *state, StringInfo tblspcmapfile)
 
SessionBackupState get_backup_status (void)
 
void do_pg_backup_stop (BackupState *state, bool waitforarchive)
 
void do_pg_abort_backup (int code, Datum arg)
 
void register_persistent_abort_backup_handler (void)
 
XLogRecPtr GetXLogInsertRecPtr (void)
 
XLogRecPtr GetXLogWriteRecPtr (void)
 
void GetOldestRestartPoint (XLogRecPtr *oldrecptr, TimeLineID *oldtli)
 
void XLogShutdownWalRcv (void)
 
void SetInstallXLogFileSegmentActive (void)
 
bool IsInstallXLogFileSegmentActive (void)
 
void SetWalWriterSleeping (bool sleeping)
 

Variables

int max_wal_size_mb = 1024
 
int min_wal_size_mb = 80
 
int wal_keep_size_mb = 0
 
int XLOGbuffers = -1
 
int XLogArchiveTimeout = 0
 
int XLogArchiveMode = ARCHIVE_MODE_OFF
 
char * XLogArchiveCommand = NULL
 
bool EnableHotStandby = false
 
bool fullPageWrites = true
 
bool wal_log_hints = false
 
int wal_compression = WAL_COMPRESSION_NONE
 
char * wal_consistency_checking_string = NULL
 
boolwal_consistency_checking = NULL
 
bool wal_init_zero = true
 
bool wal_recycle = true
 
bool log_checkpoints = true
 
int wal_sync_method = DEFAULT_WAL_SYNC_METHOD
 
int wal_level = WAL_LEVEL_REPLICA
 
int CommitDelay = 0
 
int CommitSiblings = 5
 
int wal_retrieve_retry_interval = 5000
 
int max_slot_wal_keep_size_mb = -1
 
int wal_decode_buffer_size = 512 * 1024
 
bool track_wal_io_timing = false
 
int wal_segment_size = DEFAULT_XLOG_SEG_SIZE
 
int CheckPointSegments
 
static double CheckPointDistanceEstimate = 0
 
static double PrevCheckPointDistance = 0
 
static bool check_wal_consistency_checking_deferred = false
 
const struct config_enum_entry wal_sync_method_options []
 
const struct config_enum_entry archive_mode_options []
 
CheckpointStatsData CheckpointStats
 
static bool lastFullPageWrites
 
static bool LocalRecoveryInProgress = true
 
static int LocalXLogInsertAllowed = -1
 
XLogRecPtr ProcLastRecPtr = InvalidXLogRecPtr
 
XLogRecPtr XactLastRecEnd = InvalidXLogRecPtr
 
XLogRecPtr XactLastCommitEnd = InvalidXLogRecPtr
 
static XLogRecPtr RedoRecPtr
 
static bool doPageWrites
 
static SessionBackupState sessionBackupState = SESSION_BACKUP_NONE
 
static XLogCtlDataXLogCtl = NULL
 
static WALInsertLockPaddedWALInsertLocks = NULL
 
static ControlFileDataControlFile = NULL
 
static int UsableBytesInSegment
 
static XLogwrtResult LogwrtResult = {0, 0}
 
static int openLogFile = -1
 
static XLogSegNo openLogSegNo = 0
 
static TimeLineID openLogTLI = 0
 
static XLogRecPtr LocalMinRecoveryPoint
 
static TimeLineID LocalMinRecoveryPointTLI
 
static bool updateMinRecoveryPoint = true
 
static int MyLockNo = 0
 
static bool holdingAllLocks = false
 

Macro Definition Documentation

◆ BootstrapTimeLineID

#define BootstrapTimeLineID   1

Definition at line 111 of file xlog.c.

◆ ConvertToXSegs

#define ConvertToXSegs (   x,
  segsize 
)    XLogMBVarToSegs((x), (segsize))

Definition at line 603 of file xlog.c.

◆ INSERT_FREESPACE

#define INSERT_FREESPACE (   endptr)     (((endptr) % XLOG_BLCKSZ == 0) ? 0 : (XLOG_BLCKSZ - (endptr) % XLOG_BLCKSZ))

Definition at line 580 of file xlog.c.

◆ NextBufIdx

#define NextBufIdx (   idx)     (((idx) == XLogCtl->XLogCacheBlck) ? 0 : ((idx) + 1))

Definition at line 584 of file xlog.c.

◆ NUM_XLOGINSERT_LOCKS

#define NUM_XLOGINSERT_LOCKS   8

Definition at line 150 of file xlog.c.

◆ RefreshXLogWriteResult

#define RefreshXLogWriteResult (   _target)
Value:
do { \
pg_read_barrier(); \
} while (0)
static uint64 pg_atomic_read_u64(volatile pg_atomic_uint64 *ptr)
Definition: atomics.h:467
pg_atomic_uint64 logWriteResult
Definition: xlog.c:472
pg_atomic_uint64 logFlushResult
Definition: xlog.c:473
static XLogCtlData * XLogCtl
Definition: xlog.c:566

Definition at line 620 of file xlog.c.

◆ UsableBytesInPage

#define UsableBytesInPage   (XLOG_BLCKSZ - SizeOfXLogShortPHD)

Definition at line 597 of file xlog.c.

◆ XLogRecPtrToBufIdx

#define XLogRecPtrToBufIdx (   recptr)     (((recptr) / XLOG_BLCKSZ) % (XLogCtl->XLogCacheBlck + 1))

Definition at line 591 of file xlog.c.

Typedef Documentation

◆ WALInsertLockPadded

◆ XLogCtlData

typedef struct XLogCtlData XLogCtlData

◆ XLogCtlInsert

typedef struct XLogCtlInsert XLogCtlInsert

◆ XLogwrtResult

typedef struct XLogwrtResult XLogwrtResult

◆ XLogwrtRqst

typedef struct XLogwrtRqst XLogwrtRqst

Enumeration Type Documentation

◆ WalInsertClass

Enumerator
WALINSERT_NORMAL 
WALINSERT_SPECIAL_SWITCH 
WALINSERT_SPECIAL_CHECKPOINT 

Definition at line 559 of file xlog.c.

560 {
WalInsertClass
Definition: xlog.c:560
@ WALINSERT_SPECIAL_SWITCH
Definition: xlog.c:562
@ WALINSERT_NORMAL
Definition: xlog.c:561
@ WALINSERT_SPECIAL_CHECKPOINT
Definition: xlog.c:563

Function Documentation

◆ AdvanceXLInsertBuffer()

static void AdvanceXLInsertBuffer ( XLogRecPtr  upto,
TimeLineID  tli,
bool  opportunistic 
)
static

Definition at line 1985 of file xlog.c.

1986 {
1988  int nextidx;
1989  XLogRecPtr OldPageRqstPtr;
1990  XLogwrtRqst WriteRqst;
1991  XLogRecPtr NewPageEndPtr = InvalidXLogRecPtr;
1992  XLogRecPtr NewPageBeginPtr;
1993  XLogPageHeader NewPage;
1994  int npages pg_attribute_unused() = 0;
1995 
1996  LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
1997 
1998  /*
1999  * Now that we have the lock, check if someone initialized the page
2000  * already.
2001  */
2002  while (upto >= XLogCtl->InitializedUpTo || opportunistic)
2003  {
2005 
2006  /*
2007  * Get ending-offset of the buffer page we need to replace (this may
2008  * be zero if the buffer hasn't been used yet). Fall through if it's
2009  * already written out.
2010  */
2011  OldPageRqstPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[nextidx]);
2012  if (LogwrtResult.Write < OldPageRqstPtr)
2013  {
2014  /*
2015  * Nope, got work to do. If we just want to pre-initialize as much
2016  * as we can without flushing, give up now.
2017  */
2018  if (opportunistic)
2019  break;
2020 
2021  /* Advance shared memory write request position */
2023  if (XLogCtl->LogwrtRqst.Write < OldPageRqstPtr)
2024  XLogCtl->LogwrtRqst.Write = OldPageRqstPtr;
2026 
2027  /*
2028  * Acquire an up-to-date LogwrtResult value and see if we still
2029  * need to write it or if someone else already did.
2030  */
2032  if (LogwrtResult.Write < OldPageRqstPtr)
2033  {
2034  /*
2035  * Must acquire write lock. Release WALBufMappingLock first,
2036  * to make sure that all insertions that we need to wait for
2037  * can finish (up to this same position). Otherwise we risk
2038  * deadlock.
2039  */
2040  LWLockRelease(WALBufMappingLock);
2041 
2042  WaitXLogInsertionsToFinish(OldPageRqstPtr);
2043 
2044  LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
2045 
2047  if (LogwrtResult.Write >= OldPageRqstPtr)
2048  {
2049  /* OK, someone wrote it already */
2050  LWLockRelease(WALWriteLock);
2051  }
2052  else
2053  {
2054  /* Have to write it ourselves */
2055  TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_START();
2056  WriteRqst.Write = OldPageRqstPtr;
2057  WriteRqst.Flush = 0;
2058  XLogWrite(WriteRqst, tli, false);
2059  LWLockRelease(WALWriteLock);
2061  TRACE_POSTGRESQL_WAL_BUFFER_WRITE_DIRTY_DONE();
2062  }
2063  /* Re-acquire WALBufMappingLock and retry */
2064  LWLockAcquire(WALBufMappingLock, LW_EXCLUSIVE);
2065  continue;
2066  }
2067  }
2068 
2069  /*
2070  * Now the next buffer slot is free and we can set it up to be the
2071  * next output page.
2072  */
2073  NewPageBeginPtr = XLogCtl->InitializedUpTo;
2074  NewPageEndPtr = NewPageBeginPtr + XLOG_BLCKSZ;
2075 
2076  Assert(XLogRecPtrToBufIdx(NewPageBeginPtr) == nextidx);
2077 
2078  NewPage = (XLogPageHeader) (XLogCtl->pages + nextidx * (Size) XLOG_BLCKSZ);
2079 
2080  /*
2081  * Mark the xlblock with InvalidXLogRecPtr and issue a write barrier
2082  * before initializing. Otherwise, the old page may be partially
2083  * zeroed but look valid.
2084  */
2086  pg_write_barrier();
2087 
2088  /*
2089  * Be sure to re-zero the buffer so that bytes beyond what we've
2090  * written will look like zeroes and not valid XLOG records...
2091  */
2092  MemSet((char *) NewPage, 0, XLOG_BLCKSZ);
2093 
2094  /*
2095  * Fill the new page's header
2096  */
2097  NewPage->xlp_magic = XLOG_PAGE_MAGIC;
2098 
2099  /* NewPage->xlp_info = 0; */ /* done by memset */
2100  NewPage->xlp_tli = tli;
2101  NewPage->xlp_pageaddr = NewPageBeginPtr;
2102 
2103  /* NewPage->xlp_rem_len = 0; */ /* done by memset */
2104 
2105  /*
2106  * If online backup is not in progress, mark the header to indicate
2107  * that WAL records beginning in this page have removable backup
2108  * blocks. This allows the WAL archiver to know whether it is safe to
2109  * compress archived WAL data by transforming full-block records into
2110  * the non-full-block format. It is sufficient to record this at the
2111  * page level because we force a page switch (in fact a segment
2112  * switch) when starting a backup, so the flag will be off before any
2113  * records can be written during the backup. At the end of a backup,
2114  * the last page will be marked as all unsafe when perhaps only part
2115  * is unsafe, but at worst the archiver would miss the opportunity to
2116  * compress a few records.
2117  */
2118  if (Insert->runningBackups == 0)
2119  NewPage->xlp_info |= XLP_BKP_REMOVABLE;
2120 
2121  /*
2122  * If first page of an XLOG segment file, make it a long header.
2123  */
2124  if ((XLogSegmentOffset(NewPage->xlp_pageaddr, wal_segment_size)) == 0)
2125  {
2126  XLogLongPageHeader NewLongPage = (XLogLongPageHeader) NewPage;
2127 
2128  NewLongPage->xlp_sysid = ControlFile->system_identifier;
2129  NewLongPage->xlp_seg_size = wal_segment_size;
2130  NewLongPage->xlp_xlog_blcksz = XLOG_BLCKSZ;
2131  NewPage->xlp_info |= XLP_LONG_HEADER;
2132  }
2133 
2134  /*
2135  * Make sure the initialization of the page becomes visible to others
2136  * before the xlblocks update. GetXLogBuffer() reads xlblocks without
2137  * holding a lock.
2138  */
2139  pg_write_barrier();
2140 
2141  pg_atomic_write_u64(&XLogCtl->xlblocks[nextidx], NewPageEndPtr);
2142  XLogCtl->InitializedUpTo = NewPageEndPtr;
2143 
2144  npages++;
2145  }
2146  LWLockRelease(WALBufMappingLock);
2147 
2148 #ifdef WAL_DEBUG
2149  if (XLOG_DEBUG && npages > 0)
2150  {
2151  elog(DEBUG1, "initialized %d pages, up to %X/%X",
2152  npages, LSN_FORMAT_ARGS(NewPageEndPtr));
2153  }
2154 #endif
2155 }
static void pg_atomic_write_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:485
#define pg_write_barrier()
Definition: atomics.h:157
#define pg_attribute_unused()
Definition: c.h:126
#define Assert(condition)
Definition: c.h:861
#define MemSet(start, val, len)
Definition: c.h:1023
size_t Size
Definition: c.h:608
#define DEBUG1
Definition: elog.h:30
#define elog(elevel,...)
Definition: elog.h:225
static void Insert(File file)
Definition: fd.c:1312
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1168
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1781
@ LW_EXCLUSIVE
Definition: lwlock.h:114
PgStat_PendingWalStats PendingWalStats
Definition: pgstat_wal.c:24
#define SpinLockRelease(lock)
Definition: spin.h:61
#define SpinLockAcquire(lock)
Definition: spin.h:59
uint64 system_identifier
Definition: pg_control.h:110
PgStat_Counter wal_buffers_full
Definition: pgstat.h:489
XLogwrtRqst LogwrtRqst
Definition: xlog.c:455
slock_t info_lck
Definition: xlog.c:553
XLogRecPtr InitializedUpTo
Definition: xlog.c:485
char * pages
Definition: xlog.c:492
pg_atomic_uint64 * xlblocks
Definition: xlog.c:493
XLogCtlInsert Insert
Definition: xlog.c:452
TimeLineID xlp_tli
Definition: xlog_internal.h:40
XLogRecPtr xlp_pageaddr
Definition: xlog_internal.h:41
XLogRecPtr Write
Definition: xlog.c:327
XLogRecPtr Flush
Definition: xlog.c:322
XLogRecPtr Write
Definition: xlog.c:321
static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto)
Definition: xlog.c:1504
#define RefreshXLogWriteResult(_target)
Definition: xlog.c:620
int wal_segment_size
Definition: xlog.c:143
static XLogwrtResult LogwrtResult
Definition: xlog.c:612
#define XLogRecPtrToBufIdx(recptr)
Definition: xlog.c:591
static void XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
Definition: xlog.c:2314
static ControlFileData * ControlFile
Definition: xlog.c:574
XLogLongPageHeaderData * XLogLongPageHeader
Definition: xlog_internal.h:71
#define XLogSegmentOffset(xlogptr, wal_segsz_bytes)
XLogPageHeaderData * XLogPageHeader
Definition: xlog_internal.h:54
#define XLP_LONG_HEADER
Definition: xlog_internal.h:76
#define XLP_BKP_REMOVABLE
Definition: xlog_internal.h:78
#define XLOG_PAGE_MAGIC
Definition: xlog_internal.h:34
#define LSN_FORMAT_ARGS(lsn)
Definition: xlogdefs.h:43
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28

References Assert, ControlFile, DEBUG1, elog, XLogwrtRqst::Flush, XLogCtlData::info_lck, XLogCtlData::InitializedUpTo, XLogCtlData::Insert, Insert(), InvalidXLogRecPtr, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MemSet, XLogCtlData::pages, PendingWalStats, pg_atomic_read_u64(), pg_atomic_write_u64(), pg_attribute_unused, pg_write_barrier, RefreshXLogWriteResult, SpinLockAcquire, SpinLockRelease, ControlFileData::system_identifier, WaitXLogInsertionsToFinish(), PgStat_PendingWalStats::wal_buffers_full, wal_segment_size, XLogwrtRqst::Write, XLogwrtResult::Write, XLogCtlData::xlblocks, XLOG_PAGE_MAGIC, XLogCtl, XLogRecPtrToBufIdx, XLogSegmentOffset, XLogWrite(), XLP_BKP_REMOVABLE, XLP_LONG_HEADER, XLogLongPageHeaderData::xlp_seg_size, XLogLongPageHeaderData::xlp_sysid, and XLogLongPageHeaderData::xlp_xlog_blcksz.

Referenced by GetXLogBuffer(), and XLogBackgroundFlush().

◆ assign_checkpoint_completion_target()

void assign_checkpoint_completion_target ( double  newval,
void *  extra 
)

Definition at line 2198 of file xlog.c.

2199 {
2202 }
double CheckPointCompletionTarget
Definition: checkpointer.c:138
#define newval
static void CalculateCheckpointSegments(void)
Definition: xlog.c:2162

References CalculateCheckpointSegments(), CheckPointCompletionTarget, and newval.

◆ assign_max_wal_size()

void assign_max_wal_size ( int  newval,
void *  extra 
)

Definition at line 2191 of file xlog.c.

2192 {
2195 }
int max_wal_size_mb
Definition: xlog.c:114

References CalculateCheckpointSegments(), max_wal_size_mb, and newval.

◆ assign_wal_consistency_checking()

void assign_wal_consistency_checking ( const char *  newval,
void *  extra 
)

Definition at line 4757 of file xlog.c.

4758 {
4759  /*
4760  * If some checks were deferred, it's possible that the checks will fail
4761  * later during InitializeWalConsistencyChecking(). But in that case, the
4762  * postmaster will exit anyway, so it's safe to proceed with the
4763  * assignment.
4764  *
4765  * Any built-in resource managers specified are assigned immediately,
4766  * which affects WAL created before shared_preload_libraries are
4767  * processed. Any custom resource managers specified won't be assigned
4768  * until after shared_preload_libraries are processed, but that's OK
4769  * because WAL for a custom resource manager can't be written before the
4770  * module is loaded anyway.
4771  */
4772  wal_consistency_checking = extra;
4773 }
bool * wal_consistency_checking
Definition: xlog.c:126

References wal_consistency_checking.

◆ assign_wal_sync_method()

void assign_wal_sync_method ( int  new_wal_sync_method,
void *  extra 
)

Definition at line 8641 of file xlog.c.

8642 {
8643  if (wal_sync_method != new_wal_sync_method)
8644  {
8645  /*
8646  * To ensure that no blocks escape unsynced, force an fsync on the
8647  * currently open log segment (if any). Also, if the open flag is
8648  * changing, close the log file so it will be reopened (with new flag
8649  * bit) at next use.
8650  */
8651  if (openLogFile >= 0)
8652  {
8653  pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC_METHOD_ASSIGN);
8654  if (pg_fsync(openLogFile) != 0)
8655  {
8656  char xlogfname[MAXFNAMELEN];
8657  int save_errno;
8658 
8659  save_errno = errno;
8660  XLogFileName(xlogfname, openLogTLI, openLogSegNo,
8662  errno = save_errno;
8663  ereport(PANIC,
8665  errmsg("could not fsync file \"%s\": %m", xlogfname)));
8666  }
8667 
8669  if (get_sync_bit(wal_sync_method) != get_sync_bit(new_wal_sync_method))
8670  XLogFileClose();
8671  }
8672  }
8673 }
int errcode_for_file_access(void)
Definition: elog.c:876
int errmsg(const char *fmt,...)
Definition: elog.c:1070
#define PANIC
Definition: elog.h:42
#define ereport(elevel,...)
Definition: elog.h:149
int pg_fsync(int fd)
Definition: fd.c:385
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: wait_event.h:85
static void pgstat_report_wait_end(void)
Definition: wait_event.h:101
static int openLogFile
Definition: xlog.c:635
static int get_sync_bit(int method)
Definition: xlog.c:8593
int wal_sync_method
Definition: xlog.c:130
static TimeLineID openLogTLI
Definition: xlog.c:637
static void XLogFileClose(void)
Definition: xlog.c:3639
static XLogSegNo openLogSegNo
Definition: xlog.c:636
#define MAXFNAMELEN
static void XLogFileName(char *fname, TimeLineID tli, XLogSegNo logSegNo, int wal_segsz_bytes)

References ereport, errcode_for_file_access(), errmsg(), get_sync_bit(), MAXFNAMELEN, openLogFile, openLogSegNo, openLogTLI, PANIC, pg_fsync(), pgstat_report_wait_end(), pgstat_report_wait_start(), wal_segment_size, wal_sync_method, XLogFileClose(), and XLogFileName().

◆ BootStrapXLOG()

void BootStrapXLOG ( uint32  data_checksum_version)

Definition at line 5033 of file xlog.c.

5034 {
5035  CheckPoint checkPoint;
5036  char *buffer;
5037  XLogPageHeader page;
5038  XLogLongPageHeader longpage;
5039  XLogRecord *record;
5040  char *recptr;
5041  uint64 sysidentifier;
5042  struct timeval tv;
5043  pg_crc32c crc;
5044 
5045  /* allow ordinary WAL segment creation, like StartupXLOG() would */
5047 
5048  /*
5049  * Select a hopefully-unique system identifier code for this installation.
5050  * We use the result of gettimeofday(), including the fractional seconds
5051  * field, as being about as unique as we can easily get. (Think not to
5052  * use random(), since it hasn't been seeded and there's no portable way
5053  * to seed it other than the system clock value...) The upper half of the
5054  * uint64 value is just the tv_sec part, while the lower half contains the
5055  * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
5056  * PID for a little extra uniqueness. A person knowing this encoding can
5057  * determine the initialization time of the installation, which could
5058  * perhaps be useful sometimes.
5059  */
5060  gettimeofday(&tv, NULL);
5061  sysidentifier = ((uint64) tv.tv_sec) << 32;
5062  sysidentifier |= ((uint64) tv.tv_usec) << 12;
5063  sysidentifier |= getpid() & 0xFFF;
5064 
5065  /* page buffer must be aligned suitably for O_DIRECT */
5066  buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
5067  page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
5068  memset(page, 0, XLOG_BLCKSZ);
5069 
5070  /*
5071  * Set up information for the initial checkpoint record
5072  *
5073  * The initial checkpoint record is written to the beginning of the WAL
5074  * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
5075  * used, so that we can use 0/0 to mean "before any valid WAL segment".
5076  */
5077  checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
5078  checkPoint.ThisTimeLineID = BootstrapTimeLineID;
5079  checkPoint.PrevTimeLineID = BootstrapTimeLineID;
5080  checkPoint.fullPageWrites = fullPageWrites;
5081  checkPoint.wal_level = wal_level;
5082  checkPoint.nextXid =
5084  checkPoint.nextOid = FirstGenbkiObjectId;
5085  checkPoint.nextMulti = FirstMultiXactId;
5086  checkPoint.nextMultiOffset = 0;
5087  checkPoint.oldestXid = FirstNormalTransactionId;
5088  checkPoint.oldestXidDB = Template1DbOid;
5089  checkPoint.oldestMulti = FirstMultiXactId;
5090  checkPoint.oldestMultiDB = Template1DbOid;
5093  checkPoint.time = (pg_time_t) time(NULL);
5095 
5096  TransamVariables->nextXid = checkPoint.nextXid;
5097  TransamVariables->nextOid = checkPoint.nextOid;
5099  MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5100  AdvanceOldestClogXid(checkPoint.oldestXid);
5101  SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5102  SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5104 
5105  /* Set up the XLOG page header */
5106  page->xlp_magic = XLOG_PAGE_MAGIC;
5107  page->xlp_info = XLP_LONG_HEADER;
5108  page->xlp_tli = BootstrapTimeLineID;
5110  longpage = (XLogLongPageHeader) page;
5111  longpage->xlp_sysid = sysidentifier;
5112  longpage->xlp_seg_size = wal_segment_size;
5113  longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5114 
5115  /* Insert the initial checkpoint record */
5116  recptr = ((char *) page + SizeOfXLogLongPHD);
5117  record = (XLogRecord *) recptr;
5118  record->xl_prev = 0;
5119  record->xl_xid = InvalidTransactionId;
5120  record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
5122  record->xl_rmid = RM_XLOG_ID;
5123  recptr += SizeOfXLogRecord;
5124  /* fill the XLogRecordDataHeaderShort struct */
5125  *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
5126  *(recptr++) = sizeof(checkPoint);
5127  memcpy(recptr, &checkPoint, sizeof(checkPoint));
5128  recptr += sizeof(checkPoint);
5129  Assert(recptr - (char *) record == record->xl_tot_len);
5130 
5131  INIT_CRC32C(crc);
5132  COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
5133  COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5134  FIN_CRC32C(crc);
5135  record->xl_crc = crc;
5136 
5137  /* Create first XLOG segment file */
5140 
5141  /*
5142  * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5143  * close the file again in a moment.
5144  */
5145 
5146  /* Write the first page with the initial record */
5147  errno = 0;
5148  pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
5149  if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5150  {
5151  /* if write didn't set errno, assume problem is no disk space */
5152  if (errno == 0)
5153  errno = ENOSPC;
5154  ereport(PANIC,
5156  errmsg("could not write bootstrap write-ahead log file: %m")));
5157  }
5159 
5160  pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
5161  if (pg_fsync(openLogFile) != 0)
5162  ereport(PANIC,
5164  errmsg("could not fsync bootstrap write-ahead log file: %m")));
5166 
5167  if (close(openLogFile) != 0)
5168  ereport(PANIC,
5170  errmsg("could not close bootstrap write-ahead log file: %m")));
5171 
5172  openLogFile = -1;
5173 
5174  /* Now create pg_control */
5175  InitControlFile(sysidentifier, data_checksum_version);
5176  ControlFile->time = checkPoint.time;
5177  ControlFile->checkPoint = checkPoint.redo;
5178  ControlFile->checkPointCopy = checkPoint;
5179 
5180  /* some additional ControlFile fields are set in WriteControlFile() */
5181  WriteControlFile();
5182 
5183  /* Bootstrap the commit log, too */
5184  BootStrapCLOG();
5188 
5189  pfree(buffer);
5190 
5191  /*
5192  * Force control file to be read - in contrast to normal processing we'd
5193  * otherwise never run the checks and GUC related initializations therein.
5194  */
5195  ReadControlFile();
5196 }
#define TYPEALIGN(ALIGNVAL, LEN)
Definition: c.h:807
void BootStrapCLOG(void)
Definition: clog.c:833
void BootStrapCommitTs(void)
Definition: commit_ts.c:596
void SetCommitTsLimit(TransactionId oldestXact, TransactionId newestXact)
Definition: commit_ts.c:909
#define close(a)
Definition: win32.h:12
#define write(a, b, c)
Definition: win32.h:14
void pfree(void *pointer)
Definition: mcxt.c:1521
void * palloc(Size size)
Definition: mcxt.c:1317
void MultiXactSetNextMXact(MultiXactId nextMulti, MultiXactOffset nextMultiOffset)
Definition: multixact.c:2328
void SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid, bool is_startup)
Definition: multixact.c:2362
void BootStrapMultiXact(void)
Definition: multixact.c:2034
#define FirstMultiXactId
Definition: multixact.h:25
#define XLOG_CHECKPOINT_SHUTDOWN
Definition: pg_control.h:68
uint32 pg_crc32c
Definition: pg_crc32c.h:38
#define COMP_CRC32C(crc, data, len)
Definition: pg_crc32c.h:98
#define INIT_CRC32C(crc)
Definition: pg_crc32c.h:41
#define FIN_CRC32C(crc)
Definition: pg_crc32c.h:103
return crc
int64 pg_time_t
Definition: pgtime.h:23
Oid oldestMultiDB
Definition: pg_control.h:51
MultiXactId oldestMulti
Definition: pg_control.h:50
MultiXactOffset nextMultiOffset
Definition: pg_control.h:47
TransactionId newestCommitTsXid
Definition: pg_control.h:55
TransactionId oldestXid
Definition: pg_control.h:48
TimeLineID PrevTimeLineID
Definition: pg_control.h:40
TimeLineID ThisTimeLineID
Definition: pg_control.h:39
Oid nextOid
Definition: pg_control.h:45
TransactionId oldestActiveXid
Definition: pg_control.h:64
bool fullPageWrites
Definition: pg_control.h:42
MultiXactId nextMulti
Definition: pg_control.h:46
FullTransactionId nextXid
Definition: pg_control.h:44
TransactionId oldestCommitTsXid
Definition: pg_control.h:53
pg_time_t time
Definition: pg_control.h:52
int wal_level
Definition: pg_control.h:43
XLogRecPtr redo
Definition: pg_control.h:37
Oid oldestXidDB
Definition: pg_control.h:49
CheckPoint checkPointCopy
Definition: pg_control.h:135
pg_time_t time
Definition: pg_control.h:132
XLogRecPtr checkPoint
Definition: pg_control.h:133
FullTransactionId nextXid
Definition: transam.h:220
XLogRecPtr xl_prev
Definition: xlogrecord.h:45
uint8 xl_info
Definition: xlogrecord.h:46
uint32 xl_tot_len
Definition: xlogrecord.h:43
TransactionId xl_xid
Definition: xlogrecord.h:44
RmgrId xl_rmid
Definition: xlogrecord.h:47
void BootStrapSUBTRANS(void)
Definition: subtrans.c:270
#define InvalidTransactionId
Definition: transam.h:31
#define FirstGenbkiObjectId
Definition: transam.h:195
#define FirstNormalTransactionId
Definition: transam.h:34
static FullTransactionId FullTransactionIdFromEpochAndXid(uint32 epoch, TransactionId xid)
Definition: transam.h:71
void SetTransactionIdLimit(TransactionId oldest_datfrozenxid, Oid oldest_datoid)
Definition: varsup.c:372
void AdvanceOldestClogXid(TransactionId oldest_datfrozenxid)
Definition: varsup.c:355
TransamVariablesData * TransamVariables
Definition: varsup.c:34
int gettimeofday(struct timeval *tp, void *tzp)
int XLogFileInit(XLogSegNo logsegno, TimeLineID logtli)
Definition: xlog.c:3380
bool fullPageWrites
Definition: xlog.c:122
static void InitControlFile(uint64 sysidentifier, uint32 data_checksum_version)
Definition: xlog.c:4204
void SetInstallXLogFileSegmentActive(void)
Definition: xlog.c:9484
int wal_level
Definition: xlog.c:131
static void WriteControlFile(void)
Definition: xlog.c:4239
#define BootstrapTimeLineID
Definition: xlog.c:111
static void ReadControlFile(void)
Definition: xlog.c:4321
#define SizeOfXLogLongPHD
Definition: xlog_internal.h:69
#define SizeOfXLogRecordDataHeaderShort
Definition: xlogrecord.h:217
#define XLR_BLOCK_ID_DATA_SHORT
Definition: xlogrecord.h:241
#define SizeOfXLogRecord
Definition: xlogrecord.h:55

References AdvanceOldestClogXid(), Assert, BootStrapCLOG(), BootStrapCommitTs(), BootStrapMultiXact(), BootStrapSUBTRANS(), BootstrapTimeLineID, ControlFileData::checkPoint, ControlFileData::checkPointCopy, close, COMP_CRC32C, ControlFile, crc, ereport, errcode_for_file_access(), errmsg(), FIN_CRC32C, FirstGenbkiObjectId, FirstMultiXactId, FirstNormalTransactionId, fullPageWrites, CheckPoint::fullPageWrites, FullTransactionIdFromEpochAndXid(), gettimeofday(), INIT_CRC32C, InitControlFile(), InvalidTransactionId, MultiXactSetNextMXact(), CheckPoint::newestCommitTsXid, CheckPoint::nextMulti, CheckPoint::nextMultiOffset, TransamVariablesData::nextOid, CheckPoint::nextOid, TransamVariablesData::nextXid, CheckPoint::nextXid, TransamVariablesData::oidCount, CheckPoint::oldestActiveXid, CheckPoint::oldestCommitTsXid, CheckPoint::oldestMulti, CheckPoint::oldestMultiDB, CheckPoint::oldestXid, CheckPoint::oldestXidDB, openLogFile, openLogTLI, palloc(), PANIC, pfree(), pg_fsync(), pgstat_report_wait_end(), pgstat_report_wait_start(), CheckPoint::PrevTimeLineID, ReadControlFile(), CheckPoint::redo, SetCommitTsLimit(), SetInstallXLogFileSegmentActive(), SetMultiXactIdLimit(), SetTransactionIdLimit(), SizeOfXLogLongPHD, SizeOfXLogRecord, SizeOfXLogRecordDataHeaderShort, CheckPoint::ThisTimeLineID, CheckPoint::time, ControlFileData::time, TransamVariables, TYPEALIGN, wal_level, CheckPoint::wal_level, wal_segment_size, write, WriteControlFile(), XLogRecord::xl_crc, XLogRecord::xl_info, XLogRecord::xl_prev, XLogRecord::xl_rmid, XLogRecord::xl_tot_len, XLogRecord::xl_xid, XLOG_CHECKPOINT_SHUTDOWN, XLOG_PAGE_MAGIC, XLogFileInit(), XLogPageHeaderData::xlp_info, XLP_LONG_HEADER, XLogPageHeaderData::xlp_magic, XLogPageHeaderData::xlp_pageaddr, XLogLongPageHeaderData::xlp_seg_size, XLogLongPageHeaderData::xlp_sysid, XLogPageHeaderData::xlp_tli, XLogLongPageHeaderData::xlp_xlog_blcksz, and XLR_BLOCK_ID_DATA_SHORT.

Referenced by BootstrapModeMain().

◆ CalculateCheckpointSegments()

static void CalculateCheckpointSegments ( void  )
static

Definition at line 2162 of file xlog.c.

2163 {
2164  double target;
2165 
2166  /*-------
2167  * Calculate the distance at which to trigger a checkpoint, to avoid
2168  * exceeding max_wal_size_mb. This is based on two assumptions:
2169  *
2170  * a) we keep WAL for only one checkpoint cycle (prior to PG11 we kept
2171  * WAL for two checkpoint cycles to allow us to recover from the
2172  * secondary checkpoint if the first checkpoint failed, though we
2173  * only did this on the primary anyway, not on standby. Keeping just
2174  * one checkpoint simplifies processing and reduces disk space in
2175  * many smaller databases.)
2176  * b) during checkpoint, we consume checkpoint_completion_target *
2177  * number of segments consumed between checkpoints.
2178  *-------
2179  */
2180  target = (double) ConvertToXSegs(max_wal_size_mb, wal_segment_size) /
2182 
2183  /* round down */
2184  CheckPointSegments = (int) target;
2185 
2186  if (CheckPointSegments < 1)
2187  CheckPointSegments = 1;
2188 }
#define ConvertToXSegs(x, segsize)
Definition: xlog.c:603
int CheckPointSegments
Definition: xlog.c:156

References CheckPointCompletionTarget, CheckPointSegments, ConvertToXSegs, max_wal_size_mb, and wal_segment_size.

Referenced by assign_checkpoint_completion_target(), assign_max_wal_size(), and ReadControlFile().

◆ check_max_slot_wal_keep_size()

bool check_max_slot_wal_keep_size ( int *  newval,
void **  extra,
GucSource  source 
)

Definition at line 2223 of file xlog.c.

2224 {
2225  if (IsBinaryUpgrade && *newval != -1)
2226  {
2227  GUC_check_errdetail("\"%s\" must be set to -1 during binary upgrade mode.",
2228  "max_slot_wal_keep_size");
2229  return false;
2230  }
2231 
2232  return true;
2233 }
bool IsBinaryUpgrade
Definition: globals.c:120
#define GUC_check_errdetail
Definition: guc.h:476

References GUC_check_errdetail, IsBinaryUpgrade, and newval.

◆ check_wal_buffers()

bool check_wal_buffers ( int *  newval,
void **  extra,
GucSource  source 
)

Definition at line 4637 of file xlog.c.

4638 {
4639  /*
4640  * -1 indicates a request for auto-tune.
4641  */
4642  if (*newval == -1)
4643  {
4644  /*
4645  * If we haven't yet changed the boot_val default of -1, just let it
4646  * be. We'll fix it when XLOGShmemSize is called.
4647  */
4648  if (XLOGbuffers == -1)
4649  return true;
4650 
4651  /* Otherwise, substitute the auto-tune value */
4653  }
4654 
4655  /*
4656  * We clamp manually-set values to at least 4 blocks. Prior to PostgreSQL
4657  * 9.1, a minimum of 4 was enforced by guc.c, but since that is no longer
4658  * the case, we just silently treat such values as a request for the
4659  * minimum. (We could throw an error instead, but that doesn't seem very
4660  * helpful.)
4661  */
4662  if (*newval < 4)
4663  *newval = 4;
4664 
4665  return true;
4666 }
static int XLOGChooseNumBuffers(void)
Definition: xlog.c:4621
int XLOGbuffers
Definition: xlog.c:117

References newval, XLOGbuffers, and XLOGChooseNumBuffers().

◆ check_wal_consistency_checking()

bool check_wal_consistency_checking ( char **  newval,
void **  extra,
GucSource  source 
)

Definition at line 4672 of file xlog.c.

4673 {
4674  char *rawstring;
4675  List *elemlist;
4676  ListCell *l;
4677  bool newwalconsistency[RM_MAX_ID + 1];
4678 
4679  /* Initialize the array */
4680  MemSet(newwalconsistency, 0, (RM_MAX_ID + 1) * sizeof(bool));
4681 
4682  /* Need a modifiable copy of string */
4683  rawstring = pstrdup(*newval);
4684 
4685  /* Parse string into list of identifiers */
4686  if (!SplitIdentifierString(rawstring, ',', &elemlist))
4687  {
4688  /* syntax error in list */
4689  GUC_check_errdetail("List syntax is invalid.");
4690  pfree(rawstring);
4691  list_free(elemlist);
4692  return false;
4693  }
4694 
4695  foreach(l, elemlist)
4696  {
4697  char *tok = (char *) lfirst(l);
4698  int rmid;
4699 
4700  /* Check for 'all'. */
4701  if (pg_strcasecmp(tok, "all") == 0)
4702  {
4703  for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4704  if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL)
4705  newwalconsistency[rmid] = true;
4706  }
4707  else
4708  {
4709  /* Check if the token matches any known resource manager. */
4710  bool found = false;
4711 
4712  for (rmid = 0; rmid <= RM_MAX_ID; rmid++)
4713  {
4714  if (RmgrIdExists(rmid) && GetRmgr(rmid).rm_mask != NULL &&
4715  pg_strcasecmp(tok, GetRmgr(rmid).rm_name) == 0)
4716  {
4717  newwalconsistency[rmid] = true;
4718  found = true;
4719  break;
4720  }
4721  }
4722  if (!found)
4723  {
4724  /*
4725  * During startup, it might be a not-yet-loaded custom
4726  * resource manager. Defer checking until
4727  * InitializeWalConsistencyChecking().
4728  */
4730  {
4732  }
4733  else
4734  {
4735  GUC_check_errdetail("Unrecognized key word: \"%s\".", tok);
4736  pfree(rawstring);
4737  list_free(elemlist);
4738  return false;
4739  }
4740  }
4741  }
4742  }
4743 
4744  pfree(rawstring);
4745  list_free(elemlist);
4746 
4747  /* assign new value */
4748  *extra = guc_malloc(ERROR, (RM_MAX_ID + 1) * sizeof(bool));
4749  memcpy(*extra, newwalconsistency, (RM_MAX_ID + 1) * sizeof(bool));
4750  return true;
4751 }
#define ERROR
Definition: elog.h:39
void * guc_malloc(int elevel, size_t size)
Definition: guc.c:638
void list_free(List *list)
Definition: list.c:1546
char * pstrdup(const char *in)
Definition: mcxt.c:1696
bool process_shared_preload_libraries_done
Definition: miscinit.c:1840
#define lfirst(lc)
Definition: pg_list.h:172
int pg_strcasecmp(const char *s1, const char *s2)
Definition: pgstrcasecmp.c:36
#define RM_MAX_ID
Definition: rmgr.h:33
Definition: pg_list.h:54
void(* rm_mask)(char *pagedata, BlockNumber blkno)
bool SplitIdentifierString(char *rawstring, char separator, List **namelist)
Definition: varlena.c:3432
static bool check_wal_consistency_checking_deferred
Definition: xlog.c:166
static RmgrData GetRmgr(RmgrId rmid)
static bool RmgrIdExists(RmgrId rmid)

References check_wal_consistency_checking_deferred, ERROR, GetRmgr(), GUC_check_errdetail, guc_malloc(), lfirst, list_free(), MemSet, newval, pfree(), pg_strcasecmp(), process_shared_preload_libraries_done, pstrdup(), RmgrData::rm_mask, RM_MAX_ID, RmgrIdExists(), and SplitIdentifierString().

◆ check_wal_segment_size()

bool check_wal_segment_size ( int *  newval,
void **  extra,
GucSource  source 
)

Definition at line 2205 of file xlog.c.

2206 {
2207  if (!IsValidWalSegSize(*newval))
2208  {
2209  GUC_check_errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.");
2210  return false;
2211  }
2212 
2213  return true;
2214 }
#define IsValidWalSegSize(size)
Definition: xlog_internal.h:96

References GUC_check_errdetail, IsValidWalSegSize, and newval.

◆ CheckPointGuts()

static void CheckPointGuts ( XLogRecPtr  checkPointRedo,
int  flags 
)
static

Definition at line 7510 of file xlog.c.

7511 {
7517 
7518  /* Write out all dirty data in SLRUs and the main buffer pool */
7519  TRACE_POSTGRESQL_BUFFER_CHECKPOINT_START(flags);
7521  CheckPointCLOG();
7526  CheckPointBuffers(flags);
7527 
7528  /* Perform all queued up fsyncs */
7529  TRACE_POSTGRESQL_BUFFER_CHECKPOINT_SYNC_START();
7533  TRACE_POSTGRESQL_BUFFER_CHECKPOINT_DONE();
7534 
7535  /* We deliberately delay 2PC checkpointing as long as possible */
7536  CheckPointTwoPhase(checkPointRedo);
7537 }
TimestampTz GetCurrentTimestamp(void)
Definition: timestamp.c:1644
void CheckPointBuffers(int flags)
Definition: bufmgr.c:3710
void CheckPointCLOG(void)
Definition: clog.c:937
void CheckPointCommitTs(void)
Definition: commit_ts.c:820
void CheckPointMultiXact(void)
Definition: multixact.c:2304
void CheckPointReplicationOrigin(void)
Definition: origin.c:577
void CheckPointPredicate(void)
Definition: predicate.c:1041
void CheckPointRelationMap(void)
Definition: relmapper.c:611
void CheckPointLogicalRewriteHeap(void)
Definition: rewriteheap.c:1155
void CheckPointReplicationSlots(bool is_shutdown)
Definition: slot.c:1867
void CheckPointSnapBuild(void)
Definition: snapbuild.c:1922
TimestampTz ckpt_write_t
Definition: xlog.h:162
TimestampTz ckpt_sync_end_t
Definition: xlog.h:164
TimestampTz ckpt_sync_t
Definition: xlog.h:163
void CheckPointSUBTRANS(void)
Definition: subtrans.c:355
void ProcessSyncRequests(void)
Definition: sync.c:286
void CheckPointTwoPhase(XLogRecPtr redo_horizon)
Definition: twophase.c:1823
CheckpointStatsData CheckpointStats
Definition: xlog.c:209
#define CHECKPOINT_IS_SHUTDOWN
Definition: xlog.h:139

References CHECKPOINT_IS_SHUTDOWN, CheckPointBuffers(), CheckPointCLOG(), CheckPointCommitTs(), CheckPointLogicalRewriteHeap(), CheckPointMultiXact(), CheckPointPredicate(), CheckPointRelationMap(), CheckPointReplicationOrigin(), CheckPointReplicationSlots(), CheckPointSnapBuild(), CheckpointStats, CheckPointSUBTRANS(), CheckPointTwoPhase(), CheckpointStatsData::ckpt_sync_end_t, CheckpointStatsData::ckpt_sync_t, CheckpointStatsData::ckpt_write_t, GetCurrentTimestamp(), and ProcessSyncRequests().

Referenced by CreateCheckPoint(), and CreateRestartPoint().

◆ CheckRequiredParameterValues()

static void CheckRequiredParameterValues ( void  )
static

Definition at line 5385 of file xlog.c.

5386 {
5387  /*
5388  * For archive recovery, the WAL must be generated with at least 'replica'
5389  * wal_level.
5390  */
5392  {
5393  ereport(FATAL,
5394  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5395  errmsg("WAL was generated with \"wal_level=minimal\", cannot continue recovering"),
5396  errdetail("This happens if you temporarily set \"wal_level=minimal\" on the server."),
5397  errhint("Use a backup taken after setting \"wal_level\" to higher than \"minimal\".")));
5398  }
5399 
5400  /*
5401  * For Hot Standby, the WAL must be generated with 'replica' mode, and we
5402  * must have at least as many backend slots as the primary.
5403  */
5405  {
5406  /* We ignore autovacuum_max_workers when we make this test. */
5407  RecoveryRequiresIntParameter("max_connections",
5410  RecoveryRequiresIntParameter("max_worker_processes",
5413  RecoveryRequiresIntParameter("max_wal_senders",
5416  RecoveryRequiresIntParameter("max_prepared_transactions",
5419  RecoveryRequiresIntParameter("max_locks_per_transaction",
5422  }
5423 }
int errdetail(const char *fmt,...)
Definition: elog.c:1203
int errhint(const char *fmt,...)
Definition: elog.c:1317
int errcode(int sqlerrcode)
Definition: elog.c:853
#define FATAL
Definition: elog.h:41
int MaxConnections
Definition: globals.c:142
int max_worker_processes
Definition: globals.c:143
int max_locks_per_xact
Definition: lock.c:52
int max_worker_processes
Definition: pg_control.h:181
int max_locks_per_xact
Definition: pg_control.h:184
int max_prepared_xacts
Definition: pg_control.h:183
int max_prepared_xacts
Definition: twophase.c:115
int max_wal_senders
Definition: walsender.c:121
bool EnableHotStandby
Definition: xlog.c:121
@ WAL_LEVEL_MINIMAL
Definition: xlog.h:74
bool ArchiveRecoveryRequested
Definition: xlogrecovery.c:137
void RecoveryRequiresIntParameter(const char *param_name, int currValue, int minValue)

References ArchiveRecoveryRequested, ControlFile, EnableHotStandby, ereport, errcode(), errdetail(), errhint(), errmsg(), FATAL, max_locks_per_xact, ControlFileData::max_locks_per_xact, max_prepared_xacts, ControlFileData::max_prepared_xacts, max_wal_senders, ControlFileData::max_wal_senders, max_worker_processes, ControlFileData::max_worker_processes, MaxConnections, ControlFileData::MaxConnections, RecoveryRequiresIntParameter(), ControlFileData::wal_level, and WAL_LEVEL_MINIMAL.

Referenced by StartupXLOG(), and xlog_redo().

◆ CheckXLogRemoved()

void CheckXLogRemoved ( XLogSegNo  segno,
TimeLineID  tli 
)

Definition at line 3727 of file xlog.c.

3728 {
3729  int save_errno = errno;
3730  XLogSegNo lastRemovedSegNo;
3731 
3733  lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3735 
3736  if (segno <= lastRemovedSegNo)
3737  {
3738  char filename[MAXFNAMELEN];
3739 
3740  XLogFileName(filename, tli, segno, wal_segment_size);
3741  errno = save_errno;
3742  ereport(ERROR,
3744  errmsg("requested WAL segment %s has already been removed",
3745  filename)));
3746  }
3747  errno = save_errno;
3748 }
static char * filename
Definition: pg_dumpall.c:119
XLogSegNo lastRemovedSegNo
Definition: xlog.c:461
uint64 XLogSegNo
Definition: xlogdefs.h:48

References ereport, errcode_for_file_access(), errmsg(), ERROR, filename, XLogCtlData::info_lck, XLogCtlData::lastRemovedSegNo, MAXFNAMELEN, SpinLockAcquire, SpinLockRelease, wal_segment_size, XLogCtl, and XLogFileName().

Referenced by logical_read_xlog_page(), perform_base_backup(), and XLogSendPhysical().

◆ CleanupAfterArchiveRecovery()

static void CleanupAfterArchiveRecovery ( TimeLineID  EndOfLogTLI,
XLogRecPtr  EndOfLog,
TimeLineID  newTLI 
)
static

Definition at line 5289 of file xlog.c.

5291 {
5292  /*
5293  * Execute the recovery_end_command, if any.
5294  */
5295  if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
5297  "recovery_end_command",
5298  true,
5299  WAIT_EVENT_RECOVERY_END_COMMAND);
5300 
5301  /*
5302  * We switched to a new timeline. Clean up segments on the old timeline.
5303  *
5304  * If there are any higher-numbered segments on the old timeline, remove
5305  * them. They might contain valid WAL, but they might also be
5306  * pre-allocated files containing garbage. In any case, they are not part
5307  * of the new timeline's history so we don't need them.
5308  */
5309  RemoveNonParentXlogFiles(EndOfLog, newTLI);
5310 
5311  /*
5312  * If the switch happened in the middle of a segment, what to do with the
5313  * last, partial segment on the old timeline? If we don't archive it, and
5314  * the server that created the WAL never archives it either (e.g. because
5315  * it was hit by a meteor), it will never make it to the archive. That's
5316  * OK from our point of view, because the new segment that we created with
5317  * the new TLI contains all the WAL from the old timeline up to the switch
5318  * point. But if you later try to do PITR to the "missing" WAL on the old
5319  * timeline, recovery won't find it in the archive. It's physically
5320  * present in the new file with new TLI, but recovery won't look there
5321  * when it's recovering to the older timeline. On the other hand, if we
5322  * archive the partial segment, and the original server on that timeline
5323  * is still running and archives the completed version of the same segment
5324  * later, it will fail. (We used to do that in 9.4 and below, and it
5325  * caused such problems).
5326  *
5327  * As a compromise, we rename the last segment with the .partial suffix,
5328  * and archive it. Archive recovery will never try to read .partial
5329  * segments, so they will normally go unused. But in the odd PITR case,
5330  * the administrator can copy them manually to the pg_wal directory
5331  * (removing the suffix). They can be useful in debugging, too.
5332  *
5333  * If a .done or .ready file already exists for the old timeline, however,
5334  * we had already determined that the segment is complete, so we can let
5335  * it be archived normally. (In particular, if it was restored from the
5336  * archive to begin with, it's expected to have a .done file).
5337  */
5338  if (XLogSegmentOffset(EndOfLog, wal_segment_size) != 0 &&
5340  {
5341  char origfname[MAXFNAMELEN];
5342  XLogSegNo endLogSegNo;
5343 
5344  XLByteToPrevSeg(EndOfLog, endLogSegNo, wal_segment_size);
5345  XLogFileName(origfname, EndOfLogTLI, endLogSegNo, wal_segment_size);
5346 
5347  if (!XLogArchiveIsReadyOrDone(origfname))
5348  {
5349  char origpath[MAXPGPATH];
5350  char partialfname[MAXFNAMELEN];
5351  char partialpath[MAXPGPATH];
5352 
5353  /*
5354  * If we're summarizing WAL, we can't rename the partial file
5355  * until the summarizer finishes with it, else it will fail.
5356  */
5357  if (summarize_wal)
5358  WaitForWalSummarization(EndOfLog);
5359 
5360  XLogFilePath(origpath, EndOfLogTLI, endLogSegNo, wal_segment_size);
5361  snprintf(partialfname, MAXFNAMELEN, "%s.partial", origfname);
5362  snprintf(partialpath, MAXPGPATH, "%s.partial", origpath);
5363 
5364  /*
5365  * Make sure there's no .done or .ready file for the .partial
5366  * file.
5367  */
5368  XLogArchiveCleanup(partialfname);
5369 
5370  durable_rename(origpath, partialpath, ERROR);
5371  XLogArchiveNotify(partialfname);
5372  }
5373  }
5374 }
int durable_rename(const char *oldfile, const char *newfile, int elevel)
Definition: fd.c:781
#define MAXPGPATH
#define snprintf
Definition: port.h:238
bool summarize_wal
void WaitForWalSummarization(XLogRecPtr lsn)
void RemoveNonParentXlogFiles(XLogRecPtr switchpoint, TimeLineID newTLI)
Definition: xlog.c:3940
#define XLogArchivingActive()
Definition: xlog.h:99
#define XLByteToPrevSeg(xlrp, logSegNo, wal_segsz_bytes)
static void XLogFilePath(char *path, TimeLineID tli, XLogSegNo logSegNo, int wal_segsz_bytes)
bool XLogArchiveIsReadyOrDone(const char *xlog)
Definition: xlogarchive.c:664
void ExecuteRecoveryCommand(const char *command, const char *commandName, bool failOnSignal, uint32 wait_event_info)
Definition: xlogarchive.c:295
void XLogArchiveNotify(const char *xlog)
Definition: xlogarchive.c:444
void XLogArchiveCleanup(const char *xlog)
Definition: xlogarchive.c:712
char * recoveryEndCommand
Definition: xlogrecovery.c:83

References durable_rename(), ERROR, ExecuteRecoveryCommand(), MAXFNAMELEN, MAXPGPATH, recoveryEndCommand, RemoveNonParentXlogFiles(), snprintf, summarize_wal, WaitForWalSummarization(), wal_segment_size, XLByteToPrevSeg, XLogArchiveCleanup(), XLogArchiveIsReadyOrDone(), XLogArchiveNotify(), XLogArchivingActive, XLogFileName(), XLogFilePath(), and XLogSegmentOffset.

Referenced by StartupXLOG().

◆ CleanupBackupHistory()

static void CleanupBackupHistory ( void  )
static

Definition at line 4161 of file xlog.c.

4162 {
4163  DIR *xldir;
4164  struct dirent *xlde;
4165  char path[MAXPGPATH + sizeof(XLOGDIR)];
4166 
4167  xldir = AllocateDir(XLOGDIR);
4168 
4169  while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
4170  {
4171  if (IsBackupHistoryFileName(xlde->d_name))
4172  {
4173  if (XLogArchiveCheckDone(xlde->d_name))
4174  {
4175  elog(DEBUG2, "removing WAL backup history file \"%s\"",
4176  xlde->d_name);
4177  snprintf(path, sizeof(path), XLOGDIR "/%s", xlde->d_name);
4178  unlink(path);
4179  XLogArchiveCleanup(xlde->d_name);
4180  }
4181  }
4182  }
4183 
4184  FreeDir(xldir);
4185 }
#define DEBUG2
Definition: elog.h:29
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2931
int FreeDir(DIR *dir)
Definition: fd.c:2983
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2865
Definition: dirent.c:26
Definition: dirent.h:10
char d_name[MAX_PATH]
Definition: dirent.h:15
#define XLOGDIR
static bool IsBackupHistoryFileName(const char *fname)
bool XLogArchiveCheckDone(const char *xlog)
Definition: xlogarchive.c:565

References AllocateDir(), dirent::d_name, DEBUG2, elog, FreeDir(), IsBackupHistoryFileName(), MAXPGPATH, ReadDir(), snprintf, XLogArchiveCheckDone(), XLogArchiveCleanup(), and XLOGDIR.

Referenced by do_pg_backup_stop().

◆ CopyXLogRecordToWAL()

static void CopyXLogRecordToWAL ( int  write_len,
bool  isLogSwitch,
XLogRecData rdata,
XLogRecPtr  StartPos,
XLogRecPtr  EndPos,
TimeLineID  tli 
)
static

Definition at line 1225 of file xlog.c.

1227 {
1228  char *currpos;
1229  int freespace;
1230  int written;
1231  XLogRecPtr CurrPos;
1232  XLogPageHeader pagehdr;
1233 
1234  /*
1235  * Get a pointer to the right place in the right WAL buffer to start
1236  * inserting to.
1237  */
1238  CurrPos = StartPos;
1239  currpos = GetXLogBuffer(CurrPos, tli);
1240  freespace = INSERT_FREESPACE(CurrPos);
1241 
1242  /*
1243  * there should be enough space for at least the first field (xl_tot_len)
1244  * on this page.
1245  */
1246  Assert(freespace >= sizeof(uint32));
1247 
1248  /* Copy record data */
1249  written = 0;
1250  while (rdata != NULL)
1251  {
1252  const char *rdata_data = rdata->data;
1253  int rdata_len = rdata->len;
1254 
1255  while (rdata_len > freespace)
1256  {
1257  /*
1258  * Write what fits on this page, and continue on the next page.
1259  */
1260  Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || freespace == 0);
1261  memcpy(currpos, rdata_data, freespace);
1262  rdata_data += freespace;
1263  rdata_len -= freespace;
1264  written += freespace;
1265  CurrPos += freespace;
1266 
1267  /*
1268  * Get pointer to beginning of next page, and set the xlp_rem_len
1269  * in the page header. Set XLP_FIRST_IS_CONTRECORD.
1270  *
1271  * It's safe to set the contrecord flag and xlp_rem_len without a
1272  * lock on the page. All the other flags were already set when the
1273  * page was initialized, in AdvanceXLInsertBuffer, and we're the
1274  * only backend that needs to set the contrecord flag.
1275  */
1276  currpos = GetXLogBuffer(CurrPos, tli);
1277  pagehdr = (XLogPageHeader) currpos;
1278  pagehdr->xlp_rem_len = write_len - written;
1279  pagehdr->xlp_info |= XLP_FIRST_IS_CONTRECORD;
1280 
1281  /* skip over the page header */
1282  if (XLogSegmentOffset(CurrPos, wal_segment_size) == 0)
1283  {
1284  CurrPos += SizeOfXLogLongPHD;
1285  currpos += SizeOfXLogLongPHD;
1286  }
1287  else
1288  {
1289  CurrPos += SizeOfXLogShortPHD;
1290  currpos += SizeOfXLogShortPHD;
1291  }
1292  freespace = INSERT_FREESPACE(CurrPos);
1293  }
1294 
1295  Assert(CurrPos % XLOG_BLCKSZ >= SizeOfXLogShortPHD || rdata_len == 0);
1296  memcpy(currpos, rdata_data, rdata_len);
1297  currpos += rdata_len;
1298  CurrPos += rdata_len;
1299  freespace -= rdata_len;
1300  written += rdata_len;
1301 
1302  rdata = rdata->next;
1303  }
1304  Assert(written == write_len);
1305 
1306  /*
1307  * If this was an xlog-switch, it's not enough to write the switch record,
1308  * we also have to consume all the remaining space in the WAL segment. We
1309  * have already reserved that space, but we need to actually fill it.
1310  */
1311  if (isLogSwitch && XLogSegmentOffset(CurrPos, wal_segment_size) != 0)
1312  {
1313  /* An xlog-switch record doesn't contain any data besides the header */
1314  Assert(write_len == SizeOfXLogRecord);
1315 
1316  /* Assert that we did reserve the right amount of space */
1317  Assert(XLogSegmentOffset(EndPos, wal_segment_size) == 0);
1318 
1319  /* Use up all the remaining space on the current page */
1320  CurrPos += freespace;
1321 
1322  /*
1323  * Cause all remaining pages in the segment to be flushed, leaving the
1324  * XLog position where it should be, at the start of the next segment.
1325  * We do this one page at a time, to make sure we don't deadlock
1326  * against ourselves if wal_buffers < wal_segment_size.
1327  */
1328  while (CurrPos < EndPos)
1329  {
1330  /*
1331  * The minimal action to flush the page would be to call
1332  * WALInsertLockUpdateInsertingAt(CurrPos) followed by
1333  * AdvanceXLInsertBuffer(...). The page would be left initialized
1334  * mostly to zeros, except for the page header (always the short
1335  * variant, as this is never a segment's first page).
1336  *
1337  * The large vistas of zeros are good for compressibility, but the
1338  * headers interrupting them every XLOG_BLCKSZ (with values that
1339  * differ from page to page) are not. The effect varies with
1340  * compression tool, but bzip2 for instance compresses about an
1341  * order of magnitude worse if those headers are left in place.
1342  *
1343  * Rather than complicating AdvanceXLInsertBuffer itself (which is
1344  * called in heavily-loaded circumstances as well as this lightly-
1345  * loaded one) with variant behavior, we just use GetXLogBuffer
1346  * (which itself calls the two methods we need) to get the pointer
1347  * and zero most of the page. Then we just zero the page header.
1348  */
1349  currpos = GetXLogBuffer(CurrPos, tli);
1350  MemSet(currpos, 0, SizeOfXLogShortPHD);
1351 
1352  CurrPos += XLOG_BLCKSZ;
1353  }
1354  }
1355  else
1356  {
1357  /* Align the end position, so that the next record starts aligned */
1358  CurrPos = MAXALIGN64(CurrPos);
1359  }
1360 
1361  if (CurrPos != EndPos)
1362  ereport(PANIC,
1364  errmsg_internal("space reserved for WAL record does not match what was written"));
1365 }
unsigned int uint32
Definition: c.h:518
#define MAXALIGN64(LEN)
Definition: c.h:839
int errmsg_internal(const char *fmt,...)
Definition: elog.c:1157
#define ERRCODE_DATA_CORRUPTED
Definition: pg_basebackup.c:41
const char * data
struct XLogRecData * next
#define INSERT_FREESPACE(endptr)
Definition: xlog.c:580
static char * GetXLogBuffer(XLogRecPtr ptr, TimeLineID tli)
Definition: xlog.c:1632
#define XLP_FIRST_IS_CONTRECORD
Definition: xlog_internal.h:74
#define SizeOfXLogShortPHD
Definition: xlog_internal.h:52

References Assert, XLogRecData::data, ereport, errcode(), ERRCODE_DATA_CORRUPTED, errmsg_internal(), GetXLogBuffer(), INSERT_FREESPACE, XLogRecData::len, MAXALIGN64, MemSet, XLogRecData::next, PANIC, SizeOfXLogLongPHD, SizeOfXLogRecord, SizeOfXLogShortPHD, wal_segment_size, XLogSegmentOffset, XLP_FIRST_IS_CONTRECORD, XLogPageHeaderData::xlp_info, and XLogPageHeaderData::xlp_rem_len.

Referenced by XLogInsertRecord().

◆ CreateCheckPoint()

bool CreateCheckPoint ( int  flags)

Definition at line 6889 of file xlog.c.

6890 {
6891  bool shutdown;
6892  CheckPoint checkPoint;
6893  XLogRecPtr recptr;
6894  XLogSegNo _logSegNo;
6896  uint32 freespace;
6897  XLogRecPtr PriorRedoPtr;
6898  XLogRecPtr last_important_lsn;
6899  VirtualTransactionId *vxids;
6900  int nvxids;
6901  int oldXLogAllowed = 0;
6902 
6903  /*
6904  * An end-of-recovery checkpoint is really a shutdown checkpoint, just
6905  * issued at a different time.
6906  */
6908  shutdown = true;
6909  else
6910  shutdown = false;
6911 
6912  /* sanity check */
6913  if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
6914  elog(ERROR, "can't create a checkpoint during recovery");
6915 
6916  /*
6917  * Prepare to accumulate statistics.
6918  *
6919  * Note: because it is possible for log_checkpoints to change while a
6920  * checkpoint proceeds, we always accumulate stats, even if
6921  * log_checkpoints is currently off.
6922  */
6923  MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
6925 
6926  /*
6927  * Let smgr prepare for checkpoint; this has to happen outside the
6928  * critical section and before we determine the REDO pointer. Note that
6929  * smgr must not do anything that'd have to be undone if we decide no
6930  * checkpoint is needed.
6931  */
6933 
6934  /*
6935  * Use a critical section to force system panic if we have trouble.
6936  */
6938 
6939  if (shutdown)
6940  {
6941  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6944  LWLockRelease(ControlFileLock);
6945  }
6946 
6947  /* Begin filling in the checkpoint WAL record */
6948  MemSet(&checkPoint, 0, sizeof(checkPoint));
6949  checkPoint.time = (pg_time_t) time(NULL);
6950 
6951  /*
6952  * For Hot Standby, derive the oldestActiveXid before we fix the redo
6953  * pointer. This allows us to begin accumulating changes to assemble our
6954  * starting snapshot of locks and transactions.
6955  */
6956  if (!shutdown && XLogStandbyInfoActive())
6958  else
6960 
6961  /*
6962  * Get location of last important record before acquiring insert locks (as
6963  * GetLastImportantRecPtr() also locks WAL locks).
6964  */
6965  last_important_lsn = GetLastImportantRecPtr();
6966 
6967  /*
6968  * If this isn't a shutdown or forced checkpoint, and if there has been no
6969  * WAL activity requiring a checkpoint, skip it. The idea here is to
6970  * avoid inserting duplicate checkpoints when the system is idle.
6971  */
6973  CHECKPOINT_FORCE)) == 0)
6974  {
6975  if (last_important_lsn == ControlFile->checkPoint)
6976  {
6977  END_CRIT_SECTION();
6978  ereport(DEBUG1,
6979  (errmsg_internal("checkpoint skipped because system is idle")));
6980  return false;
6981  }
6982  }
6983 
6984  /*
6985  * An end-of-recovery checkpoint is created before anyone is allowed to
6986  * write WAL. To allow us to write the checkpoint record, temporarily
6987  * enable XLogInsertAllowed.
6988  */
6989  if (flags & CHECKPOINT_END_OF_RECOVERY)
6990  oldXLogAllowed = LocalSetXLogInsertAllowed();
6991 
6992  checkPoint.ThisTimeLineID = XLogCtl->InsertTimeLineID;
6993  if (flags & CHECKPOINT_END_OF_RECOVERY)
6994  checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
6995  else
6996  checkPoint.PrevTimeLineID = checkPoint.ThisTimeLineID;
6997 
6998  /*
6999  * We must block concurrent insertions while examining insert state.
7000  */
7002 
7003  checkPoint.fullPageWrites = Insert->fullPageWrites;
7004  checkPoint.wal_level = wal_level;
7005 
7006  if (shutdown)
7007  {
7008  XLogRecPtr curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
7009 
7010  /*
7011  * Compute new REDO record ptr = location of next XLOG record.
7012  *
7013  * Since this is a shutdown checkpoint, there can't be any concurrent
7014  * WAL insertion.
7015  */
7016  freespace = INSERT_FREESPACE(curInsert);
7017  if (freespace == 0)
7018  {
7019  if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
7020  curInsert += SizeOfXLogLongPHD;
7021  else
7022  curInsert += SizeOfXLogShortPHD;
7023  }
7024  checkPoint.redo = curInsert;
7025 
7026  /*
7027  * Here we update the shared RedoRecPtr for future XLogInsert calls;
7028  * this must be done while holding all the insertion locks.
7029  *
7030  * Note: if we fail to complete the checkpoint, RedoRecPtr will be
7031  * left pointing past where it really needs to point. This is okay;
7032  * the only consequence is that XLogInsert might back up whole buffers
7033  * that it didn't really need to. We can't postpone advancing
7034  * RedoRecPtr because XLogInserts that happen while we are dumping
7035  * buffers must assume that their buffer changes are not included in
7036  * the checkpoint.
7037  */
7038  RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
7039  }
7040 
7041  /*
7042  * Now we can release the WAL insertion locks, allowing other xacts to
7043  * proceed while we are flushing disk buffers.
7044  */
7046 
7047  /*
7048  * If this is an online checkpoint, we have not yet determined the redo
7049  * point. We do so now by inserting the special XLOG_CHECKPOINT_REDO
7050  * record; the LSN at which it starts becomes the new redo pointer. We
7051  * don't do this for a shutdown checkpoint, because in that case no WAL
7052  * can be written between the redo point and the insertion of the
7053  * checkpoint record itself, so the checkpoint record itself serves to
7054  * mark the redo point.
7055  */
7056  if (!shutdown)
7057  {
7058  /* Include WAL level in record for WAL summarizer's benefit. */
7059  XLogBeginInsert();
7060  XLogRegisterData((char *) &wal_level, sizeof(wal_level));
7061  (void) XLogInsert(RM_XLOG_ID, XLOG_CHECKPOINT_REDO);
7062 
7063  /*
7064  * XLogInsertRecord will have updated XLogCtl->Insert.RedoRecPtr in
7065  * shared memory and RedoRecPtr in backend-local memory, but we need
7066  * to copy that into the record that will be inserted when the
7067  * checkpoint is complete.
7068  */
7069  checkPoint.redo = RedoRecPtr;
7070  }
7071 
7072  /* Update the info_lck-protected copy of RedoRecPtr as well */
7074  XLogCtl->RedoRecPtr = checkPoint.redo;
7076 
7077  /*
7078  * If enabled, log checkpoint start. We postpone this until now so as not
7079  * to log anything if we decided to skip the checkpoint.
7080  */
7081  if (log_checkpoints)
7082  LogCheckpointStart(flags, false);
7083 
7084  /* Update the process title */
7085  update_checkpoint_display(flags, false, false);
7086 
7087  TRACE_POSTGRESQL_CHECKPOINT_START(flags);
7088 
7089  /*
7090  * Get the other info we need for the checkpoint record.
7091  *
7092  * We don't need to save oldestClogXid in the checkpoint, it only matters
7093  * for the short period in which clog is being truncated, and if we crash
7094  * during that we'll redo the clog truncation and fix up oldestClogXid
7095  * there.
7096  */
7097  LWLockAcquire(XidGenLock, LW_SHARED);
7098  checkPoint.nextXid = TransamVariables->nextXid;
7099  checkPoint.oldestXid = TransamVariables->oldestXid;
7101  LWLockRelease(XidGenLock);
7102 
7103  LWLockAcquire(CommitTsLock, LW_SHARED);
7106  LWLockRelease(CommitTsLock);
7107 
7108  LWLockAcquire(OidGenLock, LW_SHARED);
7109  checkPoint.nextOid = TransamVariables->nextOid;
7110  if (!shutdown)
7111  checkPoint.nextOid += TransamVariables->oidCount;
7112  LWLockRelease(OidGenLock);
7113 
7114  MultiXactGetCheckptMulti(shutdown,
7115  &checkPoint.nextMulti,
7116  &checkPoint.nextMultiOffset,
7117  &checkPoint.oldestMulti,
7118  &checkPoint.oldestMultiDB);
7119 
7120  /*
7121  * Having constructed the checkpoint record, ensure all shmem disk buffers
7122  * and commit-log buffers are flushed to disk.
7123  *
7124  * This I/O could fail for various reasons. If so, we will fail to
7125  * complete the checkpoint, but there is no reason to force a system
7126  * panic. Accordingly, exit critical section while doing it.
7127  */
7128  END_CRIT_SECTION();
7129 
7130  /*
7131  * In some cases there are groups of actions that must all occur on one
7132  * side or the other of a checkpoint record. Before flushing the
7133  * checkpoint record we must explicitly wait for any backend currently
7134  * performing those groups of actions.
7135  *
7136  * One example is end of transaction, so we must wait for any transactions
7137  * that are currently in commit critical sections. If an xact inserted
7138  * its commit record into XLOG just before the REDO point, then a crash
7139  * restart from the REDO point would not replay that record, which means
7140  * that our flushing had better include the xact's update of pg_xact. So
7141  * we wait till he's out of his commit critical section before proceeding.
7142  * See notes in RecordTransactionCommit().
7143  *
7144  * Because we've already released the insertion locks, this test is a bit
7145  * fuzzy: it is possible that we will wait for xacts we didn't really need
7146  * to wait for. But the delay should be short and it seems better to make
7147  * checkpoint take a bit longer than to hold off insertions longer than
7148  * necessary. (In fact, the whole reason we have this issue is that xact.c
7149  * does commit record XLOG insertion and clog update as two separate steps
7150  * protected by different locks, but again that seems best on grounds of
7151  * minimizing lock contention.)
7152  *
7153  * A transaction that has not yet set delayChkptFlags when we look cannot
7154  * be at risk, since it has not inserted its commit record yet; and one
7155  * that's already cleared it is not at risk either, since it's done fixing
7156  * clog and we will correctly flush the update below. So we cannot miss
7157  * any xacts we need to wait for.
7158  */
7160  if (nvxids > 0)
7161  {
7162  do
7163  {
7164  /*
7165  * Keep absorbing fsync requests while we wait. There could even
7166  * be a deadlock if we don't, if the process that prevents the
7167  * checkpoint is trying to add a request to the queue.
7168  */
7170 
7171  pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_START);
7172  pg_usleep(10000L); /* wait for 10 msec */
7174  } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7176  }
7177  pfree(vxids);
7178 
7179  CheckPointGuts(checkPoint.redo, flags);
7180 
7182  if (nvxids > 0)
7183  {
7184  do
7185  {
7187 
7188  pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_COMPLETE);
7189  pg_usleep(10000L); /* wait for 10 msec */
7191  } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7193  }
7194  pfree(vxids);
7195 
7196  /*
7197  * Take a snapshot of running transactions and write this to WAL. This
7198  * allows us to reconstruct the state of running transactions during
7199  * archive recovery, if required. Skip, if this info disabled.
7200  *
7201  * If we are shutting down, or Startup process is completing crash
7202  * recovery we don't need to write running xact data.
7203  */
7204  if (!shutdown && XLogStandbyInfoActive())
7206 
7208 
7209  /*
7210  * Now insert the checkpoint record into XLOG.
7211  */
7212  XLogBeginInsert();
7213  XLogRegisterData((char *) (&checkPoint), sizeof(checkPoint));
7214  recptr = XLogInsert(RM_XLOG_ID,
7215  shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
7217 
7218  XLogFlush(recptr);
7219 
7220  /*
7221  * We mustn't write any new WAL after a shutdown checkpoint, or it will be
7222  * overwritten at next startup. No-one should even try, this just allows
7223  * sanity-checking. In the case of an end-of-recovery checkpoint, we want
7224  * to just temporarily disable writing until the system has exited
7225  * recovery.
7226  */
7227  if (shutdown)
7228  {
7229  if (flags & CHECKPOINT_END_OF_RECOVERY)
7230  LocalXLogInsertAllowed = oldXLogAllowed;
7231  else
7232  LocalXLogInsertAllowed = 0; /* never again write WAL */
7233  }
7234 
7235  /*
7236  * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
7237  * = end of actual checkpoint record.
7238  */
7239  if (shutdown && checkPoint.redo != ProcLastRecPtr)
7240  ereport(PANIC,
7241  (errmsg("concurrent write-ahead log activity while database system is shutting down")));
7242 
7243  /*
7244  * Remember the prior checkpoint's redo ptr for
7245  * UpdateCheckPointDistanceEstimate()
7246  */
7247  PriorRedoPtr = ControlFile->checkPointCopy.redo;
7248 
7249  /*
7250  * Update the control file.
7251  */
7252  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7253  if (shutdown)
7256  ControlFile->checkPointCopy = checkPoint;
7257  /* crash recovery should always recover to the end of WAL */
7260 
7261  /*
7262  * Persist unloggedLSN value. It's reset on crash recovery, so this goes
7263  * unused on non-shutdown checkpoints, but seems useful to store it always
7264  * for debugging purposes.
7265  */
7267 
7269  LWLockRelease(ControlFileLock);
7270 
7271  /* Update shared-memory copy of checkpoint XID/epoch */
7273  XLogCtl->ckptFullXid = checkPoint.nextXid;
7275 
7276  /*
7277  * We are now done with critical updates; no need for system panic if we
7278  * have trouble while fooling with old log segments.
7279  */
7280  END_CRIT_SECTION();
7281 
7282  /*
7283  * WAL summaries end when the next XLOG_CHECKPOINT_REDO or
7284  * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point
7285  * where (a) we're not inside of a critical section and (b) we can be
7286  * certain that the relevant record has been flushed to disk, which must
7287  * happen before it can be summarized.
7288  *
7289  * If this is a shutdown checkpoint, then this happens reasonably
7290  * promptly: we've only just inserted and flushed the
7291  * XLOG_CHECKPOINT_SHUTDOWN record. If this is not a shutdown checkpoint,
7292  * then this might not be very prompt at all: the XLOG_CHECKPOINT_REDO
7293  * record was written before we began flushing data to disk, and that
7294  * could be many minutes ago at this point. However, we don't XLogFlush()
7295  * after inserting that record, so we're not guaranteed that it's on disk
7296  * until after the above call that flushes the XLOG_CHECKPOINT_ONLINE
7297  * record.
7298  */
7300 
7301  /*
7302  * Let smgr do post-checkpoint cleanup (eg, deleting old files).
7303  */
7305 
7306  /*
7307  * Update the average distance between checkpoints if the prior checkpoint
7308  * exists.
7309  */
7310  if (PriorRedoPtr != InvalidXLogRecPtr)
7312 
7313  /*
7314  * Delete old log files, those no longer needed for last checkpoint to
7315  * prevent the disk holding the xlog from growing full.
7316  */
7318  KeepLogSeg(recptr, &_logSegNo);
7320  _logSegNo, InvalidOid,
7322  {
7323  /*
7324  * Some slots have been invalidated; recalculate the old-segment
7325  * horizon, starting again from RedoRecPtr.
7326  */
7328  KeepLogSeg(recptr, &_logSegNo);
7329  }
7330  _logSegNo--;
7331  RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr,
7332  checkPoint.ThisTimeLineID);
7333 
7334  /*
7335  * Make more log segments if needed. (Do this after recycling old log
7336  * segments, since that may supply some of the needed files.)
7337  */
7338  if (!shutdown)
7339  PreallocXlogFiles(recptr, checkPoint.ThisTimeLineID);
7340 
7341  /*
7342  * Truncate pg_subtrans if possible. We can throw away all data before
7343  * the oldest XMIN of any running transaction. No future transaction will
7344  * attempt to reference any pg_subtrans entry older than that (see Asserts
7345  * in subtrans.c). During recovery, though, we mustn't do this because
7346  * StartupSUBTRANS hasn't been called yet.
7347  */
7348  if (!RecoveryInProgress())
7350 
7351  /* Real work is done; log and update stats. */
7352  LogCheckpointEnd(false);
7353 
7354  /* Reset the process title */
7355  update_checkpoint_display(flags, false, true);
7356 
7357  TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
7358  NBuffers,
7362 
7363  return true;
7364 }
static uint64 pg_atomic_read_membarrier_u64(volatile pg_atomic_uint64 *ptr)
Definition: atomics.h:476
void AbsorbSyncRequests(void)
int NBuffers
Definition: globals.c:141
@ LW_SHARED
Definition: lwlock.h:115
#define START_CRIT_SECTION()
Definition: miscadmin.h:149
#define END_CRIT_SECTION()
Definition: miscadmin.h:151
void MultiXactGetCheckptMulti(bool is_shutdown, MultiXactId *nextMulti, MultiXactOffset *nextMultiOffset, MultiXactId *oldestMulti, Oid *oldestMultiDB)
Definition: multixact.c:2282
#define XLOG_CHECKPOINT_REDO
Definition: pg_control.h:82
@ DB_SHUTDOWNING
Definition: pg_control.h:94
@ DB_SHUTDOWNED
Definition: pg_control.h:92
#define XLOG_CHECKPOINT_ONLINE
Definition: pg_control.h:69
#define InvalidOid
Definition: postgres_ext.h:36
#define DELAY_CHKPT_START
Definition: proc.h:119
#define DELAY_CHKPT_COMPLETE
Definition: proc.h:120
VirtualTransactionId * GetVirtualXIDsDelayingChkpt(int *nvxids, int type)
Definition: procarray.c:3047
TransactionId GetOldestActiveTransactionId(void)
Definition: procarray.c:2884
TransactionId GetOldestTransactionIdConsideredRunning(void)
Definition: procarray.c:2034
bool HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids, int type)
Definition: procarray.c:3093
void pg_usleep(long microsec)
Definition: signal.c:53
bool InvalidateObsoleteReplicationSlots(ReplicationSlotInvalidationCause cause, XLogSegNo oldestSegno, Oid dboid, TransactionId snapshotConflictHorizon)
Definition: slot.c:1811
@ RS_INVAL_WAL_REMOVED
Definition: slot.h:54
XLogRecPtr LogStandbySnapshot(void)
Definition: standby.c:1281
TimestampTz ckpt_start_t
Definition: xlog.h:161
int ckpt_segs_removed
Definition: xlog.h:171
int ckpt_segs_added
Definition: xlog.h:170
int ckpt_bufs_written
Definition: xlog.h:167
int ckpt_segs_recycled
Definition: xlog.h:172
XLogRecPtr minRecoveryPoint
Definition: pg_control.h:168
XLogRecPtr unloggedLSN
Definition: pg_control.h:137
TimeLineID minRecoveryPointTLI
Definition: pg_control.h:169
TransactionId oldestCommitTsXid
Definition: transam.h:232
TransactionId newestCommitTsXid
Definition: transam.h:233
TransactionId oldestXid
Definition: transam.h:222
FullTransactionId ckptFullXid
Definition: xlog.c:457
TimeLineID InsertTimeLineID
Definition: xlog.c:509
XLogRecPtr RedoRecPtr
Definition: xlog.c:456
TimeLineID PrevTimeLineID
Definition: xlog.c:510
pg_atomic_uint64 unloggedLSN
Definition: xlog.c:464
XLogRecPtr RedoRecPtr
Definition: xlog.c:430
void TruncateSUBTRANS(TransactionId oldestXact)
Definition: subtrans.c:411
void SyncPreCheckpoint(void)
Definition: sync.c:177
void SyncPostCheckpoint(void)
Definition: sync.c:202
void WakeupWalSummarizer(void)
XLogRecPtr ProcLastRecPtr
Definition: xlog.c:253
bool RecoveryInProgress(void)
Definition: xlog.c:6334
static void WALInsertLockRelease(void)
Definition: xlog.c:1445
static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos)
Definition: xlog.c:1858
static void WALInsertLockAcquireExclusive(void)
Definition: xlog.c:1416
static void UpdateControlFile(void)
Definition: xlog.c:4559
static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr, TimeLineID insertTLI)
Definition: xlog.c:3865
static void LogCheckpointStart(int flags, bool restartpoint)
Definition: xlog.c:6649
static XLogRecPtr RedoRecPtr
Definition: xlog.c:273
static void LogCheckpointEnd(bool restartpoint)
Definition: xlog.c:6681
static void PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli)
Definition: xlog.c:3690
bool log_checkpoints
Definition: xlog.c:129
static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
Definition: xlog.c:7953
static int LocalSetXLogInsertAllowed(void)
Definition: xlog.c:6422
XLogRecPtr GetLastImportantRecPtr(void)
Definition: xlog.c:6556
static void UpdateCheckPointDistanceEstimate(uint64 nbytes)
Definition: xlog.c:6786
static int LocalXLogInsertAllowed
Definition: xlog.c:236
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2802
static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
Definition: xlog.c:7510
static void update_checkpoint_display(int flags, bool restartpoint, bool reset)
Definition: xlog.c:6824
#define CHECKPOINT_END_OF_RECOVERY
Definition: xlog.h:140
#define CHECKPOINT_FORCE
Definition: xlog.h:142
#define XLogStandbyInfoActive()
Definition: xlog.h:123
#define XLByteToSeg(xlrp, logSegNo, wal_segsz_bytes)
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:474
void XLogRegisterData(const char *data, uint32 len)
Definition: xloginsert.c:364
void XLogBeginInsert(void)
Definition: xloginsert.c:149

References AbsorbSyncRequests(), ControlFileData::checkPoint, CHECKPOINT_END_OF_RECOVERY, CHECKPOINT_FORCE, CHECKPOINT_IS_SHUTDOWN, ControlFileData::checkPointCopy, CheckPointGuts(), CheckpointStats, CheckpointStatsData::ckpt_bufs_written, CheckpointStatsData::ckpt_segs_added, CheckpointStatsData::ckpt_segs_recycled, CheckpointStatsData::ckpt_segs_removed, CheckpointStatsData::ckpt_start_t, XLogCtlData::ckptFullXid, ControlFile, DB_SHUTDOWNED, DB_SHUTDOWNING, DEBUG1, DELAY_CHKPT_COMPLETE, DELAY_CHKPT_START, elog, END_CRIT_SECTION, ereport, errmsg(), errmsg_internal(), ERROR, CheckPoint::fullPageWrites, GetCurrentTimestamp(), GetLastImportantRecPtr(), GetOldestActiveTransactionId(), GetOldestTransactionIdConsideredRunning(), GetVirtualXIDsDelayingChkpt(), HaveVirtualXIDsDelayingChkpt(), XLogCtlData::info_lck, XLogCtlData::Insert, Insert(), INSERT_FREESPACE, XLogCtlData::InsertTimeLineID, InvalidateObsoleteReplicationSlots(), InvalidOid, InvalidTransactionId, InvalidXLogRecPtr, KeepLogSeg(), LocalSetXLogInsertAllowed(), LocalXLogInsertAllowed, log_checkpoints, LogCheckpointEnd(), LogCheckpointStart(), LogStandbySnapshot(), LW_EXCLUSIVE, LW_SHARED, LWLockAcquire(), LWLockRelease(), MemSet, ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, MultiXactGetCheckptMulti(), NBuffers, TransamVariablesData::newestCommitTsXid, CheckPoint::newestCommitTsXid, CheckPoint::nextMulti, CheckPoint::nextMultiOffset, TransamVariablesData::nextOid, CheckPoint::nextOid, TransamVariablesData::nextXid, CheckPoint::nextXid, TransamVariablesData::oidCount, CheckPoint::oldestActiveXid, TransamVariablesData::oldestCommitTsXid, CheckPoint::oldestCommitTsXid, CheckPoint::oldestMulti, CheckPoint::oldestMultiDB, TransamVariablesData::oldestXid, CheckPoint::oldestXid, TransamVariablesData::oldestXidDB, CheckPoint::oldestXidDB, PANIC, pfree(), pg_atomic_read_membarrier_u64(), pg_usleep(), pgstat_report_wait_end(), pgstat_report_wait_start(), PreallocXlogFiles(), XLogCtlData::PrevTimeLineID, CheckPoint::PrevTimeLineID, ProcLastRecPtr, RecoveryInProgress(), CheckPoint::redo, RedoRecPtr, XLogCtlInsert::RedoRecPtr, XLogCtlData::RedoRecPtr, RemoveOldXlogFiles(), RS_INVAL_WAL_REMOVED, SizeOfXLogLongPHD, SizeOfXLogShortPHD, SpinLockAcquire, SpinLockRelease, START_CRIT_SECTION, ControlFileData::state, SyncPostCheckpoint(), SyncPreCheckpoint(), CheckPoint::ThisTimeLineID, CheckPoint::time, TransamVariables, TruncateSUBTRANS(), XLogCtlData::unloggedLSN, ControlFileData::unloggedLSN, update_checkpoint_display(), UpdateCheckPointDistanceEstimate(), UpdateControlFile(), WakeupWalSummarizer(), wal_level, CheckPoint::wal_level, wal_segment_size, WALInsertLockAcquireExclusive(), WALInsertLockRelease(), XLByteToSeg, XLOG_CHECKPOINT_ONLINE, XLOG_CHECKPOINT_REDO, XLOG_CHECKPOINT_SHUTDOWN, XLogBeginInsert(), XLogBytePosToRecPtr(), XLogCtl, XLogFlush(), XLogInsert(), XLogRegisterData(), XLogSegmentOffset, and XLogStandbyInfoActive.

Referenced by CheckpointerMain(), RequestCheckpoint(), and ShutdownXLOG().

◆ CreateEndOfRecoveryRecord()

static void CreateEndOfRecoveryRecord ( void  )
static

Definition at line 7375 of file xlog.c.

7376 {
7377  xl_end_of_recovery xlrec;
7378  XLogRecPtr recptr;
7379 
7380  /* sanity check */
7381  if (!RecoveryInProgress())
7382  elog(ERROR, "can only be used to end recovery");
7383 
7384  xlrec.end_time = GetCurrentTimestamp();
7385  xlrec.wal_level = wal_level;
7386 
7391 
7393 
7394  XLogBeginInsert();
7395  XLogRegisterData((char *) &xlrec, sizeof(xl_end_of_recovery));
7396  recptr = XLogInsert(RM_XLOG_ID, XLOG_END_OF_RECOVERY);
7397 
7398  XLogFlush(recptr);
7399 
7400  /*
7401  * Update the control file so that crash recovery can follow the timeline
7402  * changes to this point.
7403  */
7404  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7405  ControlFile->minRecoveryPoint = recptr;
7406  ControlFile->minRecoveryPointTLI = xlrec.ThisTimeLineID;
7408  LWLockRelease(ControlFileLock);
7409 
7410  END_CRIT_SECTION();
7411 }
#define XLOG_END_OF_RECOVERY
Definition: pg_control.h:77
TimeLineID PrevTimeLineID
TimestampTz end_time
TimeLineID ThisTimeLineID

References ControlFile, elog, END_CRIT_SECTION, xl_end_of_recovery::end_time, ERROR, GetCurrentTimestamp(), XLogCtlData::InsertTimeLineID, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, XLogCtlData::PrevTimeLineID, xl_end_of_recovery::PrevTimeLineID, RecoveryInProgress(), START_CRIT_SECTION, xl_end_of_recovery::ThisTimeLineID, UpdateControlFile(), wal_level, xl_end_of_recovery::wal_level, WALInsertLockAcquireExclusive(), WALInsertLockRelease(), XLOG_END_OF_RECOVERY, XLogBeginInsert(), XLogCtl, XLogFlush(), XLogInsert(), and XLogRegisterData().

Referenced by PerformRecoveryXLogAction().

◆ CreateOverwriteContrecordRecord()

static XLogRecPtr CreateOverwriteContrecordRecord ( XLogRecPtr  aborted_lsn,
XLogRecPtr  pagePtr,
TimeLineID  newTLI 
)
static

Definition at line 7440 of file xlog.c.

7442 {
7444  XLogRecPtr recptr;
7445  XLogPageHeader pagehdr;
7446  XLogRecPtr startPos;
7447 
7448  /* sanity checks */
7449  if (!RecoveryInProgress())
7450  elog(ERROR, "can only be used at end of recovery");
7451  if (pagePtr % XLOG_BLCKSZ != 0)
7452  elog(ERROR, "invalid position for missing continuation record %X/%X",
7453  LSN_FORMAT_ARGS(pagePtr));
7454 
7455  /* The current WAL insert position should be right after the page header */
7456  startPos = pagePtr;
7457  if (XLogSegmentOffset(startPos, wal_segment_size) == 0)
7458  startPos += SizeOfXLogLongPHD;
7459  else
7460  startPos += SizeOfXLogShortPHD;
7461  recptr = GetXLogInsertRecPtr();
7462  if (recptr != startPos)
7463  elog(ERROR, "invalid WAL insert position %X/%X for OVERWRITE_CONTRECORD",
7464  LSN_FORMAT_ARGS(recptr));
7465 
7467 
7468  /*
7469  * Initialize the XLOG page header (by GetXLogBuffer), and set the
7470  * XLP_FIRST_IS_OVERWRITE_CONTRECORD flag.
7471  *
7472  * No other backend is allowed to write WAL yet, so acquiring the WAL
7473  * insertion lock is just pro forma.
7474  */
7476  pagehdr = (XLogPageHeader) GetXLogBuffer(pagePtr, newTLI);
7479 
7480  /*
7481  * Insert the XLOG_OVERWRITE_CONTRECORD record as the first record on the
7482  * page. We know it becomes the first record, because no other backend is
7483  * allowed to write WAL yet.
7484  */
7485  XLogBeginInsert();
7486  xlrec.overwritten_lsn = aborted_lsn;
7488  XLogRegisterData((char *) &xlrec, sizeof(xl_overwrite_contrecord));
7489  recptr = XLogInsert(RM_XLOG_ID, XLOG_OVERWRITE_CONTRECORD);
7490 
7491  /* check that the record was inserted to the right place */
7492  if (ProcLastRecPtr != startPos)
7493  elog(ERROR, "OVERWRITE_CONTRECORD was inserted to unexpected position %X/%X",
7495 
7496  XLogFlush(recptr);
7497 
7498  END_CRIT_SECTION();
7499 
7500  return recptr;
7501 }
#define XLOG_OVERWRITE_CONTRECORD
Definition: pg_control.h:81
static void WALInsertLockAcquire(void)
Definition: xlog.c:1371
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:9435
#define XLP_FIRST_IS_OVERWRITE_CONTRECORD
Definition: xlog_internal.h:80

References elog, END_CRIT_SECTION, ERROR, GetCurrentTimestamp(), GetXLogBuffer(), GetXLogInsertRecPtr(), LSN_FORMAT_ARGS, xl_overwrite_contrecord::overwrite_time, xl_overwrite_contrecord::overwritten_lsn, ProcLastRecPtr, RecoveryInProgress(), SizeOfXLogLongPHD, SizeOfXLogShortPHD, START_CRIT_SECTION, wal_segment_size, WALInsertLockAcquire(), WALInsertLockRelease(), XLOG_OVERWRITE_CONTRECORD, XLogBeginInsert(), XLogFlush(), XLogInsert(), XLogRegisterData(), XLogSegmentOffset, XLP_FIRST_IS_OVERWRITE_CONTRECORD, and XLogPageHeaderData::xlp_info.

Referenced by StartupXLOG().

◆ CreateRestartPoint()

bool CreateRestartPoint ( int  flags)

Definition at line 7591 of file xlog.c.

7592 {
7593  XLogRecPtr lastCheckPointRecPtr;
7594  XLogRecPtr lastCheckPointEndPtr;
7595  CheckPoint lastCheckPoint;
7596  XLogRecPtr PriorRedoPtr;
7597  XLogRecPtr receivePtr;
7598  XLogRecPtr replayPtr;
7599  TimeLineID replayTLI;
7600  XLogRecPtr endptr;
7601  XLogSegNo _logSegNo;
7602  TimestampTz xtime;
7603 
7604  /* Concurrent checkpoint/restartpoint cannot happen */
7606 
7607  /* Get a local copy of the last safe checkpoint record. */
7609  lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
7610  lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
7611  lastCheckPoint = XLogCtl->lastCheckPoint;
7613 
7614  /*
7615  * Check that we're still in recovery mode. It's ok if we exit recovery
7616  * mode after this check, the restart point is valid anyway.
7617  */
7618  if (!RecoveryInProgress())
7619  {
7620  ereport(DEBUG2,
7621  (errmsg_internal("skipping restartpoint, recovery has already ended")));
7622  return false;
7623  }
7624 
7625  /*
7626  * If the last checkpoint record we've replayed is already our last
7627  * restartpoint, we can't perform a new restart point. We still update
7628  * minRecoveryPoint in that case, so that if this is a shutdown restart
7629  * point, we won't start up earlier than before. That's not strictly
7630  * necessary, but when hot standby is enabled, it would be rather weird if
7631  * the database opened up for read-only connections at a point-in-time
7632  * before the last shutdown. Such time travel is still possible in case of
7633  * immediate shutdown, though.
7634  *
7635  * We don't explicitly advance minRecoveryPoint when we do create a
7636  * restartpoint. It's assumed that flushing the buffers will do that as a
7637  * side-effect.
7638  */
7639  if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
7640  lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
7641  {
7642  ereport(DEBUG2,
7643  (errmsg_internal("skipping restartpoint, already performed at %X/%X",
7644  LSN_FORMAT_ARGS(lastCheckPoint.redo))));
7645 
7647  if (flags & CHECKPOINT_IS_SHUTDOWN)
7648  {
7649  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7652  LWLockRelease(ControlFileLock);
7653  }
7654  return false;
7655  }
7656 
7657  /*
7658  * Update the shared RedoRecPtr so that the startup process can calculate
7659  * the number of segments replayed since last restartpoint, and request a
7660  * restartpoint if it exceeds CheckPointSegments.
7661  *
7662  * Like in CreateCheckPoint(), hold off insertions to update it, although
7663  * during recovery this is just pro forma, because no WAL insertions are
7664  * happening.
7665  */
7667  RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
7669 
7670  /* Also update the info_lck-protected copy */
7672  XLogCtl->RedoRecPtr = lastCheckPoint.redo;
7674 
7675  /*
7676  * Prepare to accumulate statistics.
7677  *
7678  * Note: because it is possible for log_checkpoints to change while a
7679  * checkpoint proceeds, we always accumulate stats, even if
7680  * log_checkpoints is currently off.
7681  */
7682  MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
7684 
7685  if (log_checkpoints)
7686  LogCheckpointStart(flags, true);
7687 
7688  /* Update the process title */
7689  update_checkpoint_display(flags, true, false);
7690 
7691  CheckPointGuts(lastCheckPoint.redo, flags);
7692 
7693  /*
7694  * This location needs to be after CheckPointGuts() to ensure that some
7695  * work has already happened during this checkpoint.
7696  */
7697  INJECTION_POINT("create-restart-point");
7698 
7699  /*
7700  * Remember the prior checkpoint's redo ptr for
7701  * UpdateCheckPointDistanceEstimate()
7702  */
7703  PriorRedoPtr = ControlFile->checkPointCopy.redo;
7704 
7705  /*
7706  * Update pg_control, using current time. Check that it still shows an
7707  * older checkpoint, else do nothing; this is a quick hack to make sure
7708  * nothing really bad happens if somehow we get here after the
7709  * end-of-recovery checkpoint.
7710  */
7711  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7712  if (ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
7713  {
7714  /*
7715  * Update the checkpoint information. We do this even if the cluster
7716  * does not show DB_IN_ARCHIVE_RECOVERY to match with the set of WAL
7717  * segments recycled below.
7718  */
7719  ControlFile->checkPoint = lastCheckPointRecPtr;
7720  ControlFile->checkPointCopy = lastCheckPoint;
7721 
7722  /*
7723  * Ensure minRecoveryPoint is past the checkpoint record and update it
7724  * if the control file still shows DB_IN_ARCHIVE_RECOVERY. Normally,
7725  * this will have happened already while writing out dirty buffers,
7726  * but not necessarily - e.g. because no buffers were dirtied. We do
7727  * this because a backup performed in recovery uses minRecoveryPoint
7728  * to determine which WAL files must be included in the backup, and
7729  * the file (or files) containing the checkpoint record must be
7730  * included, at a minimum. Note that for an ordinary restart of
7731  * recovery there's no value in having the minimum recovery point any
7732  * earlier than this anyway, because redo will begin just after the
7733  * checkpoint record.
7734  */
7736  {
7737  if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
7738  {
7739  ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
7741 
7742  /* update local copy */
7745  }
7746  if (flags & CHECKPOINT_IS_SHUTDOWN)
7748  }
7750  }
7751  LWLockRelease(ControlFileLock);
7752 
7753  /*
7754  * Update the average distance between checkpoints/restartpoints if the
7755  * prior checkpoint exists.
7756  */
7757  if (PriorRedoPtr != InvalidXLogRecPtr)
7759 
7760  /*
7761  * Delete old log files, those no longer needed for last restartpoint to
7762  * prevent the disk holding the xlog from growing full.
7763  */
7765 
7766  /*
7767  * Retreat _logSegNo using the current end of xlog replayed or received,
7768  * whichever is later.
7769  */
7770  receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
7771  replayPtr = GetXLogReplayRecPtr(&replayTLI);
7772  endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
7773  KeepLogSeg(endptr, &_logSegNo);
7775  _logSegNo, InvalidOid,
7777  {
7778  /*
7779  * Some slots have been invalidated; recalculate the old-segment
7780  * horizon, starting again from RedoRecPtr.
7781  */
7783  KeepLogSeg(endptr, &_logSegNo);
7784  }
7785  _logSegNo--;
7786 
7787  /*
7788  * Try to recycle segments on a useful timeline. If we've been promoted
7789  * since the beginning of this restartpoint, use the new timeline chosen
7790  * at end of recovery. If we're still in recovery, use the timeline we're
7791  * currently replaying.
7792  *
7793  * There is no guarantee that the WAL segments will be useful on the
7794  * current timeline; if recovery proceeds to a new timeline right after
7795  * this, the pre-allocated WAL segments on this timeline will not be used,
7796  * and will go wasted until recycled on the next restartpoint. We'll live
7797  * with that.
7798  */
7799  if (!RecoveryInProgress())
7800  replayTLI = XLogCtl->InsertTimeLineID;
7801 
7802  RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr, replayTLI);
7803 
7804  /*
7805  * Make more log segments if needed. (Do this after recycling old log
7806  * segments, since that may supply some of the needed files.)
7807  */
7808  PreallocXlogFiles(endptr, replayTLI);
7809 
7810  /*
7811  * Truncate pg_subtrans if possible. We can throw away all data before
7812  * the oldest XMIN of any running transaction. No future transaction will
7813  * attempt to reference any pg_subtrans entry older than that (see Asserts
7814  * in subtrans.c). When hot standby is disabled, though, we mustn't do
7815  * this because StartupSUBTRANS hasn't been called yet.
7816  */
7817  if (EnableHotStandby)
7819 
7820  /* Real work is done; log and update stats. */
7821  LogCheckpointEnd(true);
7822 
7823  /* Reset the process title */
7824  update_checkpoint_display(flags, true, true);
7825 
7826  xtime = GetLatestXTime();
7828  (errmsg("recovery restart point at %X/%X",
7829  LSN_FORMAT_ARGS(lastCheckPoint.redo)),
7830  xtime ? errdetail("Last completed transaction was at log time %s.",
7831  timestamptz_to_str(xtime)) : 0));
7832 
7833  /*
7834  * Finally, execute archive_cleanup_command, if any.
7835  */
7836  if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
7838  "archive_cleanup_command",
7839  false,
7840  WAIT_EVENT_ARCHIVE_CLEANUP_COMMAND);
7841 
7842  return true;
7843 }
const char * timestamptz_to_str(TimestampTz t)
Definition: timestamp.c:1843
int64 TimestampTz
Definition: timestamp.h:39
#define LOG
Definition: elog.h:31
bool IsUnderPostmaster
Definition: globals.c:119
#define INJECTION_POINT(name)
@ B_CHECKPOINTER
Definition: miscadmin.h:355
BackendType MyBackendType
Definition: miscinit.c:64
@ DB_IN_ARCHIVE_RECOVERY
Definition: pg_control.h:96
@ DB_SHUTDOWNED_IN_RECOVERY
Definition: pg_control.h:93
CheckPoint lastCheckPoint
Definition: xlog.c:545
XLogRecPtr lastCheckPointRecPtr
Definition: xlog.c:543
XLogRecPtr lastCheckPointEndPtr
Definition: xlog.c:544
XLogRecPtr GetWalRcvFlushRecPtr(XLogRecPtr *latestChunkStart, TimeLineID *receiveTLI)
static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
Definition: xlog.c:2722
static XLogRecPtr LocalMinRecoveryPoint
Definition: xlog.c:646
static TimeLineID LocalMinRecoveryPointTLI
Definition: xlog.c:647
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
uint32 TimeLineID
Definition: xlogdefs.h:59
char * archiveCleanupCommand
Definition: xlogrecovery.c:84
XLogRecPtr GetXLogReplayRecPtr(TimeLineID *replayTLI)
TimestampTz GetLatestXTime(void)

References archiveCleanupCommand, Assert, B_CHECKPOINTER, ControlFileData::checkPoint, CHECKPOINT_IS_SHUTDOWN, ControlFileData::checkPointCopy, CheckPointGuts(), CheckpointStats, CheckpointStatsData::ckpt_start_t, ControlFile, DB_IN_ARCHIVE_RECOVERY, DB_SHUTDOWNED_IN_RECOVERY, DEBUG2, EnableHotStandby, ereport, errdetail(), errmsg(), errmsg_internal(), ExecuteRecoveryCommand(), GetCurrentTimestamp(), GetLatestXTime(), GetOldestTransactionIdConsideredRunning(), GetWalRcvFlushRecPtr(), GetXLogReplayRecPtr(), XLogCtlData::info_lck, INJECTION_POINT, XLogCtlData::Insert, XLogCtlData::InsertTimeLineID, InvalidateObsoleteReplicationSlots(), InvalidOid, InvalidTransactionId, InvalidXLogRecPtr, IsUnderPostmaster, KeepLogSeg(), XLogCtlData::lastCheckPoint, XLogCtlData::lastCheckPointEndPtr, XLogCtlData::lastCheckPointRecPtr, LocalMinRecoveryPoint, LocalMinRecoveryPointTLI, LOG, log_checkpoints, LogCheckpointEnd(), LogCheckpointStart(), LSN_FORMAT_ARGS, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MemSet, ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, MyBackendType, PreallocXlogFiles(), RecoveryInProgress(), CheckPoint::redo, RedoRecPtr, XLogCtlInsert::RedoRecPtr, XLogCtlData::RedoRecPtr, RemoveOldXlogFiles(), RS_INVAL_WAL_REMOVED, SpinLockAcquire, SpinLockRelease, ControlFileData::state, CheckPoint::ThisTimeLineID, timestamptz_to_str(), TruncateSUBTRANS(), update_checkpoint_display(), UpdateCheckPointDistanceEstimate(), UpdateControlFile(), UpdateMinRecoveryPoint(), wal_segment_size, WALInsertLockAcquireExclusive(), WALInsertLockRelease(), XLByteToSeg, XLogCtl, and XLogRecPtrIsInvalid.

Referenced by CheckpointerMain(), and ShutdownXLOG().

◆ DataChecksumsEnabled()

◆ do_pg_abort_backup()

void do_pg_abort_backup ( int  code,
Datum  arg 
)

Definition at line 9394 of file xlog.c.

9395 {
9396  bool during_backup_start = DatumGetBool(arg);
9397 
9398  /* If called during backup start, there shouldn't be one already running */
9399  Assert(!during_backup_start || sessionBackupState == SESSION_BACKUP_NONE);
9400 
9401  if (during_backup_start || sessionBackupState != SESSION_BACKUP_NONE)
9402  {
9406 
9409 
9410  if (!during_backup_start)
9411  ereport(WARNING,
9412  errmsg("aborting backup due to backend exiting before pg_backup_stop was called"));
9413  }
9414 }
#define WARNING
Definition: elog.h:36
void * arg
static bool DatumGetBool(Datum X)
Definition: postgres.h:90
int runningBackups
Definition: xlog.c:438
static SessionBackupState sessionBackupState
Definition: xlog.c:391
@ SESSION_BACKUP_NONE
Definition: xlog.h:287

References arg, Assert, DatumGetBool(), ereport, errmsg(), XLogCtlData::Insert, XLogCtlInsert::runningBackups, SESSION_BACKUP_NONE, sessionBackupState, WALInsertLockAcquireExclusive(), WALInsertLockRelease(), WARNING, and XLogCtl.

Referenced by do_pg_backup_start(), perform_base_backup(), and register_persistent_abort_backup_handler().

◆ do_pg_backup_start()

void do_pg_backup_start ( const char *  backupidstr,
bool  fast,
List **  tablespaces,
BackupState state,
StringInfo  tblspcmapfile 
)

Definition at line 8792 of file xlog.c.

8794 {
8796 
8797  Assert(state != NULL);
8799 
8800  /*
8801  * During recovery, we don't need to check WAL level. Because, if WAL
8802  * level is not sufficient, it's impossible to get here during recovery.
8803  */
8805  ereport(ERROR,
8806  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8807  errmsg("WAL level not sufficient for making an online backup"),
8808  errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
8809 
8810  if (strlen(backupidstr) > MAXPGPATH)
8811  ereport(ERROR,
8812  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8813  errmsg("backup label too long (max %d bytes)",
8814  MAXPGPATH)));
8815 
8816  strlcpy(state->name, backupidstr, sizeof(state->name));
8817 
8818  /*
8819  * Mark backup active in shared memory. We must do full-page WAL writes
8820  * during an on-line backup even if not doing so at other times, because
8821  * it's quite possible for the backup dump to obtain a "torn" (partially
8822  * written) copy of a database page if it reads the page concurrently with
8823  * our write to the same page. This can be fixed as long as the first
8824  * write to the page in the WAL sequence is a full-page write. Hence, we
8825  * increment runningBackups then force a CHECKPOINT, to ensure there are
8826  * no dirty pages in shared memory that might get dumped while the backup
8827  * is in progress without having a corresponding WAL record. (Once the
8828  * backup is complete, we need not force full-page writes anymore, since
8829  * we expect that any pages not modified during the backup interval must
8830  * have been correctly captured by the backup.)
8831  *
8832  * Note that forcing full-page writes has no effect during an online
8833  * backup from the standby.
8834  *
8835  * We must hold all the insertion locks to change the value of
8836  * runningBackups, to ensure adequate interlocking against
8837  * XLogInsertRecord().
8838  */
8842 
8843  /*
8844  * Ensure we decrement runningBackups if we fail below. NB -- for this to
8845  * work correctly, it is critical that sessionBackupState is only updated
8846  * after this block is over.
8847  */
8849  {
8850  bool gotUniqueStartpoint = false;
8851  DIR *tblspcdir;
8852  struct dirent *de;
8853  tablespaceinfo *ti;
8854  int datadirpathlen;
8855 
8856  /*
8857  * Force an XLOG file switch before the checkpoint, to ensure that the
8858  * WAL segment the checkpoint is written to doesn't contain pages with
8859  * old timeline IDs. That would otherwise happen if you called
8860  * pg_backup_start() right after restoring from a PITR archive: the
8861  * first WAL segment containing the startup checkpoint has pages in
8862  * the beginning with the old timeline ID. That can cause trouble at
8863  * recovery: we won't have a history file covering the old timeline if
8864  * pg_wal directory was not included in the base backup and the WAL
8865  * archive was cleared too before starting the backup.
8866  *
8867  * This also ensures that we have emitted a WAL page header that has
8868  * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
8869  * Therefore, if a WAL archiver (such as pglesslog) is trying to
8870  * compress out removable backup blocks, it won't remove any that
8871  * occur after this point.
8872  *
8873  * During recovery, we skip forcing XLOG file switch, which means that
8874  * the backup taken during recovery is not available for the special
8875  * recovery case described above.
8876  */
8878  RequestXLogSwitch(false);
8879 
8880  do
8881  {
8882  bool checkpointfpw;
8883 
8884  /*
8885  * Force a CHECKPOINT. Aside from being necessary to prevent torn
8886  * page problems, this guarantees that two successive backup runs
8887  * will have different checkpoint positions and hence different
8888  * history file names, even if nothing happened in between.
8889  *
8890  * During recovery, establish a restartpoint if possible. We use
8891  * the last restartpoint as the backup starting checkpoint. This
8892  * means that two successive backup runs can have same checkpoint
8893  * positions.
8894  *
8895  * Since the fact that we are executing do_pg_backup_start()
8896  * during recovery means that checkpointer is running, we can use
8897  * RequestCheckpoint() to establish a restartpoint.
8898  *
8899  * We use CHECKPOINT_IMMEDIATE only if requested by user (via
8900  * passing fast = true). Otherwise this can take awhile.
8901  */
8903  (fast ? CHECKPOINT_IMMEDIATE : 0));
8904 
8905  /*
8906  * Now we need to fetch the checkpoint record location, and also
8907  * its REDO pointer. The oldest point in WAL that would be needed
8908  * to restore starting from the checkpoint is precisely the REDO
8909  * pointer.
8910  */
8911  LWLockAcquire(ControlFileLock, LW_SHARED);
8912  state->checkpointloc = ControlFile->checkPoint;
8913  state->startpoint = ControlFile->checkPointCopy.redo;
8915  checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
8916  LWLockRelease(ControlFileLock);
8917 
8919  {
8920  XLogRecPtr recptr;
8921 
8922  /*
8923  * Check to see if all WAL replayed during online backup
8924  * (i.e., since last restartpoint used as backup starting
8925  * checkpoint) contain full-page writes.
8926  */
8928  recptr = XLogCtl->lastFpwDisableRecPtr;
8930 
8931  if (!checkpointfpw || state->startpoint <= recptr)
8932  ereport(ERROR,
8933  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8934  errmsg("WAL generated with \"full_page_writes=off\" was replayed "
8935  "since last restartpoint"),
8936  errhint("This means that the backup being taken on the standby "
8937  "is corrupt and should not be used. "
8938  "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
8939  "and then try an online backup again.")));
8940 
8941  /*
8942  * During recovery, since we don't use the end-of-backup WAL
8943  * record and don't write the backup history file, the
8944  * starting WAL location doesn't need to be unique. This means
8945  * that two base backups started at the same time might use
8946  * the same checkpoint as starting locations.
8947  */
8948  gotUniqueStartpoint = true;
8949  }
8950 
8951  /*
8952  * If two base backups are started at the same time (in WAL sender
8953  * processes), we need to make sure that they use different
8954  * checkpoints as starting locations, because we use the starting
8955  * WAL location as a unique identifier for the base backup in the
8956  * end-of-backup WAL record and when we write the backup history
8957  * file. Perhaps it would be better generate a separate unique ID
8958  * for each backup instead of forcing another checkpoint, but
8959  * taking a checkpoint right after another is not that expensive
8960  * either because only few buffers have been dirtied yet.
8961  */
8963  if (XLogCtl->Insert.lastBackupStart < state->startpoint)
8964  {
8965  XLogCtl->Insert.lastBackupStart = state->startpoint;
8966  gotUniqueStartpoint = true;
8967  }
8969  } while (!gotUniqueStartpoint);
8970 
8971  /*
8972  * Construct tablespace_map file.
8973  */
8974  datadirpathlen = strlen(DataDir);
8975 
8976  /* Collect information about all tablespaces */
8977  tblspcdir = AllocateDir(PG_TBLSPC_DIR);
8978  while ((de = ReadDir(tblspcdir, PG_TBLSPC_DIR)) != NULL)
8979  {
8980  char fullpath[MAXPGPATH + sizeof(PG_TBLSPC_DIR)];
8981  char linkpath[MAXPGPATH];
8982  char *relpath = NULL;
8983  char *s;
8984  PGFileType de_type;
8985  char *badp;
8986  Oid tsoid;
8987 
8988  /*
8989  * Try to parse the directory name as an unsigned integer.
8990  *
8991  * Tablespace directories should be positive integers that can be
8992  * represented in 32 bits, with no leading zeroes or trailing
8993  * garbage. If we come across a name that doesn't meet those
8994  * criteria, skip it.
8995  */
8996  if (de->d_name[0] < '1' || de->d_name[1] > '9')
8997  continue;
8998  errno = 0;
8999  tsoid = strtoul(de->d_name, &badp, 10);
9000  if (*badp != '\0' || errno == EINVAL || errno == ERANGE)
9001  continue;
9002 
9003  snprintf(fullpath, sizeof(fullpath), "%s/%s", PG_TBLSPC_DIR, de->d_name);
9004 
9005  de_type = get_dirent_type(fullpath, de, false, ERROR);
9006 
9007  if (de_type == PGFILETYPE_LNK)
9008  {
9009  StringInfoData escapedpath;
9010  int rllen;
9011 
9012  rllen = readlink(fullpath, linkpath, sizeof(linkpath));
9013  if (rllen < 0)
9014  {
9015  ereport(WARNING,
9016  (errmsg("could not read symbolic link \"%s\": %m",
9017  fullpath)));
9018  continue;
9019  }
9020  else if (rllen >= sizeof(linkpath))
9021  {
9022  ereport(WARNING,
9023  (errmsg("symbolic link \"%s\" target is too long",
9024  fullpath)));
9025  continue;
9026  }
9027  linkpath[rllen] = '\0';
9028 
9029  /*
9030  * Relpath holds the relative path of the tablespace directory
9031  * when it's located within PGDATA, or NULL if it's located
9032  * elsewhere.
9033  */
9034  if (rllen > datadirpathlen &&
9035  strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
9036  IS_DIR_SEP(linkpath[datadirpathlen]))
9037  relpath = pstrdup(linkpath + datadirpathlen + 1);
9038 
9039  /*
9040  * Add a backslash-escaped version of the link path to the
9041  * tablespace map file.
9042  */
9043  initStringInfo(&escapedpath);
9044  for (s = linkpath; *s; s++)
9045  {
9046  if (*s == '\n' || *s == '\r' || *s == '\\')
9047  appendStringInfoChar(&escapedpath, '\\');
9048  appendStringInfoChar(&escapedpath, *s);
9049  }
9050  appendStringInfo(tblspcmapfile, "%s %s\n",
9051  de->d_name, escapedpath.data);
9052  pfree(escapedpath.data);
9053  }
9054  else if (de_type == PGFILETYPE_DIR)
9055  {
9056  /*
9057  * It's possible to use allow_in_place_tablespaces to create
9058  * directories directly under pg_tblspc, for testing purposes
9059  * only.
9060  *
9061  * In this case, we store a relative path rather than an
9062  * absolute path into the tablespaceinfo.
9063  */
9064  snprintf(linkpath, sizeof(linkpath), "%s/%s",
9065  PG_TBLSPC_DIR, de->d_name);
9066  relpath = pstrdup(linkpath);
9067  }
9068  else
9069  {
9070  /* Skip any other file type that appears here. */
9071  continue;
9072  }
9073 
9074  ti = palloc(sizeof(tablespaceinfo));
9075  ti->oid = tsoid;
9076  ti->path = pstrdup(linkpath);
9077  ti->rpath = relpath;
9078  ti->size = -1;
9079 
9080  if (tablespaces)
9081  *tablespaces = lappend(*tablespaces, ti);
9082  }
9083  FreeDir(tblspcdir);
9084 
9085  state->starttime = (pg_time_t) time(NULL);
9086  }
9088 
9089  state->started_in_recovery = backup_started_in_recovery;
9090 
9091  /*
9092  * Mark that the start phase has correctly finished for the backup.
9093  */
9095 }
static bool backup_started_in_recovery
Definition: basebackup.c:123
void RequestCheckpoint(int flags)
Definition: checkpointer.c:952
PGFileType get_dirent_type(const char *path, const struct dirent *de, bool look_through_symlinks, int elevel)
Definition: file_utils.c:526
PGFileType
Definition: file_utils.h:19
@ PGFILETYPE_LNK
Definition: file_utils.h:24
@ PGFILETYPE_DIR
Definition: file_utils.h:23
char * DataDir
Definition: globals.c:70
#define PG_ENSURE_ERROR_CLEANUP(cleanup_function, arg)
Definition: ipc.h:47
#define PG_END_ENSURE_ERROR_CLEANUP(cleanup_function, arg)
Definition: ipc.h:52
List * lappend(List *list, void *datum)
Definition: list.c:339
#define IS_DIR_SEP(ch)
Definition: port.h:102
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45
unsigned int Oid
Definition: postgres_ext.h:31
#define relpath(rlocator, forknum)
Definition: relpath.h:102
#define PG_TBLSPC_DIR
Definition: relpath.h:41
void appendStringInfo(StringInfo str, const char *fmt,...)
Definition: stringinfo.c:94
void appendStringInfoChar(StringInfo str, char ch)
Definition: stringinfo.c:191
void initStringInfo(StringInfo str)
Definition: stringinfo.c:56
XLogRecPtr lastFpwDisableRecPtr
Definition: xlog.c:551
XLogRecPtr lastBackupStart
Definition: xlog.c:439
Definition: regguts.h:323
char * rpath
Definition: basebackup.h:32
#define readlink(path, buf, size)
Definition: win32_port.h:236
XLogRecPtr RequestXLogSwitch(bool mark_unimportant)
Definition: xlog.c:8060
void do_pg_abort_backup(int code, Datum arg)
Definition: xlog.c:9394
@ SESSION_BACKUP_RUNNING
Definition: xlog.h:288
#define CHECKPOINT_WAIT
Definition: xlog.h:145
#define CHECKPOINT_IMMEDIATE
Definition: xlog.h:141
#define XLogIsNeeded()
Definition: xlog.h:109

References AllocateDir(), appendStringInfo(), appendStringInfoChar(), Assert, backup_started_in_recovery, ControlFileData::checkPoint, CHECKPOINT_FORCE, CHECKPOINT_IMMEDIATE, CHECKPOINT_WAIT, ControlFileData::checkPointCopy, ControlFile, dirent::d_name, StringInfoData::data, DataDir, DatumGetBool(), do_pg_abort_backup(), ereport, errcode(), errhint(), errmsg(), ERROR, FreeDir(), CheckPoint::fullPageWrites, get_dirent_type(), XLogCtlData::info_lck, initStringInfo(), XLogCtlData::Insert, IS_DIR_SEP, lappend(), XLogCtlInsert::lastBackupStart, XLogCtlData::lastFpwDisableRecPtr, LW_SHARED, LWLockAcquire(), LWLockRelease(), MAXPGPATH, tablespaceinfo::oid, palloc(), tablespaceinfo::path, pfree(), PG_END_ENSURE_ERROR_CLEANUP, PG_ENSURE_ERROR_CLEANUP, PG_TBLSPC_DIR, PGFILETYPE_DIR, PGFILETYPE_LNK, pstrdup(), ReadDir(), readlink, RecoveryInProgress(), CheckPoint::redo, relpath, RequestCheckpoint(), RequestXLogSwitch(), tablespaceinfo::rpath, XLogCtlInsert::runningBackups, SESSION_BACKUP_RUNNING, sessionBackupState, tablespaceinfo::size, snprintf, SpinLockAcquire, SpinLockRelease, strlcpy(), CheckPoint::ThisTimeLineID, WALInsertLockAcquireExclusive(), WALInsertLockRelease(), WARNING, XLogCtl, and XLogIsNeeded.

Referenced by perform_base_backup(), and pg_backup_start().

◆ do_pg_backup_stop()

void do_pg_backup_stop ( BackupState state,
bool  waitforarchive 
)

Definition at line 9120 of file xlog.c.

9121 {
9122  bool backup_stopped_in_recovery = false;
9123  char histfilepath[MAXPGPATH];
9124  char lastxlogfilename[MAXFNAMELEN];
9125  char histfilename[MAXFNAMELEN];
9126  XLogSegNo _logSegNo;
9127  FILE *fp;
9128  int seconds_before_warning;
9129  int waits = 0;
9130  bool reported_waiting = false;
9131 
9132  Assert(state != NULL);
9133 
9134  backup_stopped_in_recovery = RecoveryInProgress();
9135 
9136  /*
9137  * During recovery, we don't need to check WAL level. Because, if WAL
9138  * level is not sufficient, it's impossible to get here during recovery.
9139  */
9140  if (!backup_stopped_in_recovery && !XLogIsNeeded())
9141  ereport(ERROR,
9142  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9143  errmsg("WAL level not sufficient for making an online backup"),
9144  errhint("\"wal_level\" must be set to \"replica\" or \"logical\" at server start.")));
9145 
9146  /*
9147  * OK to update backup counter and session-level lock.
9148  *
9149  * Note that CHECK_FOR_INTERRUPTS() must not occur while updating them,
9150  * otherwise they can be updated inconsistently, which might cause
9151  * do_pg_abort_backup() to fail.
9152  */
9154 
9155  /*
9156  * It is expected that each do_pg_backup_start() call is matched by
9157  * exactly one do_pg_backup_stop() call.
9158  */
9161 
9162  /*
9163  * Clean up session-level lock.
9164  *
9165  * You might think that WALInsertLockRelease() can be called before
9166  * cleaning up session-level lock because session-level lock doesn't need
9167  * to be protected with WAL insertion lock. But since
9168  * CHECK_FOR_INTERRUPTS() can occur in it, session-level lock must be
9169  * cleaned up before it.
9170  */
9172 
9174 
9175  /*
9176  * If we are taking an online backup from the standby, we confirm that the
9177  * standby has not been promoted during the backup.
9178  */
9179  if (state->started_in_recovery && !backup_stopped_in_recovery)
9180  ereport(ERROR,
9181  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9182  errmsg("the standby was promoted during online backup"),
9183  errhint("This means that the backup being taken is corrupt "
9184  "and should not be used. "
9185  "Try taking another online backup.")));
9186 
9187  /*
9188  * During recovery, we don't write an end-of-backup record. We assume that
9189  * pg_control was backed up last and its minimum recovery point can be
9190  * available as the backup end location. Since we don't have an
9191  * end-of-backup record, we use the pg_control value to check whether
9192  * we've reached the end of backup when starting recovery from this
9193  * backup. We have no way of checking if pg_control wasn't backed up last
9194  * however.
9195  *
9196  * We don't force a switch to new WAL file but it is still possible to
9197  * wait for all the required files to be archived if waitforarchive is
9198  * true. This is okay if we use the backup to start a standby and fetch
9199  * the missing WAL using streaming replication. But in the case of an
9200  * archive recovery, a user should set waitforarchive to true and wait for
9201  * them to be archived to ensure that all the required files are
9202  * available.
9203  *
9204  * We return the current minimum recovery point as the backup end
9205  * location. Note that it can be greater than the exact backup end
9206  * location if the minimum recovery point is updated after the backup of
9207  * pg_control. This is harmless for current uses.
9208  *
9209  * XXX currently a backup history file is for informational and debug
9210  * purposes only. It's not essential for an online backup. Furthermore,
9211  * even if it's created, it will not be archived during recovery because
9212  * an archiver is not invoked. So it doesn't seem worthwhile to write a
9213  * backup history file during recovery.
9214  */
9215  if (backup_stopped_in_recovery)
9216  {
9217  XLogRecPtr recptr;
9218 
9219  /*
9220  * Check to see if all WAL replayed during online backup contain
9221  * full-page writes.
9222  */
9224  recptr = XLogCtl->lastFpwDisableRecPtr;
9226 
9227  if (state->startpoint <= recptr)
9228  ereport(ERROR,
9229  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9230  errmsg("WAL generated with \"full_page_writes=off\" was replayed "
9231  "during online backup"),
9232  errhint("This means that the backup being taken on the standby "
9233  "is corrupt and should not be used. "
9234  "Enable \"full_page_writes\" and run CHECKPOINT on the primary, "
9235  "and then try an online backup again.")));
9236 
9237 
9238  LWLockAcquire(ControlFileLock, LW_SHARED);
9239  state->stoppoint = ControlFile->minRecoveryPoint;
9241  LWLockRelease(ControlFileLock);
9242  }
9243  else
9244  {
9245  char *history_file;
9246 
9247  /*
9248  * Write the backup-end xlog record
9249  */
9250  XLogBeginInsert();
9251  XLogRegisterData((char *) (&state->startpoint),
9252  sizeof(state->startpoint));
9253  state->stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
9254 
9255  /*
9256  * Given that we're not in recovery, InsertTimeLineID is set and can't
9257  * change, so we can read it without a lock.
9258  */
9259  state->stoptli = XLogCtl->InsertTimeLineID;
9260 
9261  /*
9262  * Force a switch to a new xlog segment file, so that the backup is
9263  * valid as soon as archiver moves out the current segment file.
9264  */
9265  RequestXLogSwitch(false);
9266 
9267  state->stoptime = (pg_time_t) time(NULL);
9268 
9269  /*
9270  * Write the backup history file
9271  */
9272  XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9273  BackupHistoryFilePath(histfilepath, state->stoptli, _logSegNo,
9274  state->startpoint, wal_segment_size);
9275  fp = AllocateFile(histfilepath, "w");
9276  if (!fp)
9277  ereport(ERROR,
9279  errmsg("could not create file \"%s\": %m",
9280  histfilepath)));
9281 
9282  /* Build and save the contents of the backup history file */
9283  history_file = build_backup_content(state, true);
9284  fprintf(fp, "%s", history_file);
9285  pfree(history_file);
9286 
9287  if (fflush(fp) || ferror(fp) || FreeFile(fp))
9288  ereport(ERROR,
9290  errmsg("could not write file \"%s\": %m",
9291  histfilepath)));
9292 
9293  /*
9294  * Clean out any no-longer-needed history files. As a side effect,
9295  * this will post a .ready file for the newly created history file,
9296  * notifying the archiver that history file may be archived
9297  * immediately.
9298  */
9300  }
9301 
9302  /*
9303  * If archiving is enabled, wait for all the required WAL files to be
9304  * archived before returning. If archiving isn't enabled, the required WAL
9305  * needs to be transported via streaming replication (hopefully with
9306  * wal_keep_size set high enough), or some more exotic mechanism like
9307  * polling and copying files from pg_wal with script. We have no knowledge
9308  * of those mechanisms, so it's up to the user to ensure that he gets all
9309  * the required WAL.
9310  *
9311  * We wait until both the last WAL file filled during backup and the
9312  * history file have been archived, and assume that the alphabetic sorting
9313  * property of the WAL files ensures any earlier WAL files are safely
9314  * archived as well.
9315  *
9316  * We wait forever, since archive_command is supposed to work and we
9317  * assume the admin wanted his backup to work completely. If you don't
9318  * wish to wait, then either waitforarchive should be passed in as false,
9319  * or you can set statement_timeout. Also, some notices are issued to
9320  * clue in anyone who might be doing this interactively.
9321  */
9322 
9323  if (waitforarchive &&
9324  ((!backup_stopped_in_recovery && XLogArchivingActive()) ||
9325  (backup_stopped_in_recovery && XLogArchivingAlways())))
9326  {
9327  XLByteToPrevSeg(state->stoppoint, _logSegNo, wal_segment_size);
9328  XLogFileName(lastxlogfilename, state->stoptli, _logSegNo,
9330 
9331  XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9332  BackupHistoryFileName(histfilename, state->stoptli, _logSegNo,
9333  state->startpoint, wal_segment_size);
9334 
9335  seconds_before_warning = 60;
9336  waits = 0;
9337 
9338  while (XLogArchiveIsBusy(lastxlogfilename) ||
9339  XLogArchiveIsBusy(histfilename))
9340  {
9342 
9343  if (!reported_waiting && waits > 5)
9344  {
9345  ereport(NOTICE,
9346  (errmsg("base backup done, waiting for required WAL segments to be archived")));
9347  reported_waiting = true;
9348  }
9349 
9350  (void) WaitLatch(MyLatch,
9352  1000L,
9353  WAIT_EVENT_BACKUP_WAIT_WAL_ARCHIVE);
9355 
9356  if (++waits >= seconds_before_warning)
9357  {
9358  seconds_before_warning *= 2; /* This wraps in >10 years... */
9359  ereport(WARNING,
9360  (errmsg("still waiting for all required WAL segments to be archived (%d seconds elapsed)",
9361  waits),
9362  errhint("Check that your \"archive_command\" is executing properly. "
9363  "You can safely cancel this backup, "
9364  "but the database backup will not be usable without all the WAL segments.")));
9365  }
9366  }
9367 
9368  ereport(NOTICE,
9369  (errmsg("all required WAL segments have been archived")));
9370  }
9371  else if (waitforarchive)
9372  ereport(NOTICE,
9373  (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
9374 }
#define NOTICE
Definition: elog.h:35
FILE * AllocateFile(const char *name, const char *mode)
Definition: fd.c:2605
int FreeFile(FILE *file)
Definition: fd.c:2803
struct Latch * MyLatch
Definition: globals.c:62
void ResetLatch(Latch *latch)
Definition: latch.c:724
int WaitLatch(Latch *latch, int wakeEvents, long timeout, uint32 wait_event_info)
Definition: latch.c:517
#define WL_TIMEOUT
Definition: latch.h:130
#define WL_EXIT_ON_PM_DEATH
Definition: latch.h:132
#define WL_LATCH_SET
Definition: latch.h:127
static void const char fflush(stdout)
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:122
#define XLOG_BACKUP_END
Definition: pg_control.h:73
#define fprintf
Definition: port.h:242
static void CleanupBackupHistory(void)
Definition: xlog.c:4161
#define XLogArchivingAlways()
Definition: xlog.h:102
static void BackupHistoryFileName(char *fname, TimeLineID tli, XLogSegNo logSegNo, XLogRecPtr startpoint, int wal_segsz_bytes)
static void BackupHistoryFilePath(char *path, TimeLineID tli, XLogSegNo logSegNo, XLogRecPtr startpoint, int wal_segsz_bytes)
bool XLogArchiveIsBusy(const char *xlog)
Definition: xlogarchive.c:619
char * build_backup_content(BackupState *state, bool ishistoryfile)
Definition: xlogbackup.c:29

References AllocateFile(), Assert, BackupHistoryFileName(), BackupHistoryFilePath(), build_backup_content(), CHECK_FOR_INTERRUPTS, CleanupBackupHistory(), ControlFile, ereport, errcode(), errcode_for_file_access(), errhint(), errmsg(), ERROR, fflush(), fprintf, FreeFile(), XLogCtlData::info_lck, XLogCtlData::Insert, XLogCtlData::InsertTimeLineID, XLogCtlData::lastFpwDisableRecPtr, LW_SHARED, LWLockAcquire(), LWLockRelease(), MAXFNAMELEN, MAXPGPATH, ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, MyLatch, NOTICE, pfree(), RecoveryInProgress(), RequestXLogSwitch(), ResetLatch(), XLogCtlInsert::runningBackups, SESSION_BACKUP_NONE, sessionBackupState, SpinLockAcquire, SpinLockRelease, WaitLatch(), wal_segment_size, WALInsertLockAcquireExclusive(), WALInsertLockRelease(), WARNING, WL_EXIT_ON_PM_DEATH, WL_LATCH_SET, WL_TIMEOUT, XLByteToPrevSeg, XLByteToSeg, XLOG_BACKUP_END, XLogArchiveIsBusy(), XLogArchivingActive, XLogArchivingAlways, XLogBeginInsert(), XLogCtl, XLogFileName(), XLogInsert(), XLogIsNeeded, and XLogRegisterData().

Referenced by perform_base_backup(), and pg_backup_stop().

◆ get_backup_status()

SessionBackupState get_backup_status ( void  )

Definition at line 9101 of file xlog.c.

9102 {
9103  return sessionBackupState;
9104 }

References sessionBackupState.

Referenced by pg_backup_start(), pg_backup_stop(), and SendBaseBackup().

◆ get_sync_bit()

static int get_sync_bit ( int  method)
static

Definition at line 8593 of file xlog.c.

8594 {
8595  int o_direct_flag = 0;
8596 
8597  /*
8598  * Use O_DIRECT if requested, except in walreceiver process. The WAL
8599  * written by walreceiver is normally read by the startup process soon
8600  * after it's written. Also, walreceiver performs unaligned writes, which
8601  * don't work with O_DIRECT, so it is required for correctness too.
8602  */
8604  o_direct_flag = PG_O_DIRECT;
8605 
8606  /* If fsync is disabled, never open in sync mode */
8607  if (!enableFsync)
8608  return o_direct_flag;
8609 
8610  switch (method)
8611  {
8612  /*
8613  * enum values for all sync options are defined even if they are
8614  * not supported on the current platform. But if not, they are
8615  * not included in the enum option array, and therefore will never
8616  * be seen here.
8617  */
8618  case WAL_SYNC_METHOD_FSYNC:
8621  return o_direct_flag;
8622 #ifdef O_SYNC
8623  case WAL_SYNC_METHOD_OPEN:
8624  return O_SYNC | o_direct_flag;
8625 #endif
8626 #ifdef O_DSYNC
8628  return O_DSYNC | o_direct_flag;
8629 #endif
8630  default:
8631  /* can't happen (unless we are out of sync with option array) */
8632  elog(ERROR, "unrecognized \"wal_sync_method\": %d", method);
8633  return 0; /* silence warning */
8634  }
8635 }
int io_direct_flags
Definition: fd.c:167
#define IO_DIRECT_WAL
Definition: fd.h:55
#define PG_O_DIRECT
Definition: fd.h:97
bool enableFsync
Definition: globals.c:128
#define AmWalReceiverProcess()
Definition: miscadmin.h:381
#define O_DSYNC
Definition: win32_port.h:352
@ WAL_SYNC_METHOD_OPEN
Definition: xlog.h:26
@ WAL_SYNC_METHOD_FDATASYNC
Definition: xlog.h:25
@ WAL_SYNC_METHOD_FSYNC_WRITETHROUGH
Definition: xlog.h:27
@ WAL_SYNC_METHOD_OPEN_DSYNC
Definition: xlog.h:28
@ WAL_SYNC_METHOD_FSYNC
Definition: xlog.h:24

References AmWalReceiverProcess, elog, enableFsync, ERROR, io_direct_flags, IO_DIRECT_WAL, O_DSYNC, PG_O_DIRECT, WAL_SYNC_METHOD_FDATASYNC, WAL_SYNC_METHOD_FSYNC, WAL_SYNC_METHOD_FSYNC_WRITETHROUGH, WAL_SYNC_METHOD_OPEN, and WAL_SYNC_METHOD_OPEN_DSYNC.

Referenced by assign_wal_sync_method(), XLogFileInit(), XLogFileInitInternal(), and XLogFileOpen().

◆ GetActiveWalLevelOnStandby()

WalLevel GetActiveWalLevelOnStandby ( void  )

Definition at line 4859 of file xlog.c.

4860 {
4861  return ControlFile->wal_level;
4862 }

References ControlFile, and ControlFileData::wal_level.

Referenced by CheckLogicalDecodingRequirements().

◆ GetFakeLSNForUnloggedRel()

XLogRecPtr GetFakeLSNForUnloggedRel ( void  )

Definition at line 4604 of file xlog.c.

4605 {
4607 }
static uint64 pg_atomic_fetch_add_u64(volatile pg_atomic_uint64 *ptr, int64 add_)
Definition: atomics.h:522

References pg_atomic_fetch_add_u64(), XLogCtlData::unloggedLSN, and XLogCtl.

Referenced by gistGetFakeLSN().

◆ GetFlushRecPtr()

XLogRecPtr GetFlushRecPtr ( TimeLineID insertTLI)

Definition at line 6499 of file xlog.c.

6500 {
6502 
6504 
6505  /*
6506  * If we're writing and flushing WAL, the time line can't be changing, so
6507  * no lock is required.
6508  */
6509  if (insertTLI)
6510  *insertTLI = XLogCtl->InsertTimeLineID;
6511 
6512  return LogwrtResult.Flush;
6513 }
RecoveryState SharedRecoveryState
Definition: xlog.c:516
XLogRecPtr Flush
Definition: xlog.c:328
@ RECOVERY_STATE_DONE
Definition: xlog.h:93

References Assert, XLogwrtResult::Flush, XLogCtlData::InsertTimeLineID, LogwrtResult, RECOVERY_STATE_DONE, RefreshXLogWriteResult, XLogCtlData::SharedRecoveryState, and XLogCtl.

Referenced by binary_upgrade_logical_slot_has_caught_up(), get_flush_position(), GetCurrentLSN(), GetLatestLSN(), IdentifySystem(), pg_current_wal_flush_lsn(), pg_logical_slot_get_changes_guts(), pg_replication_slot_advance(), read_local_xlog_page_guts(), StartReplication(), WalSndWaitForWal(), XLogSendLogical(), and XLogSendPhysical().

◆ GetFullPageWriteInfo()

void GetFullPageWriteInfo ( XLogRecPtr RedoRecPtr_p,
bool doPageWrites_p 
)

Definition at line 6467 of file xlog.c.

6468 {
6469  *RedoRecPtr_p = RedoRecPtr;
6470  *doPageWrites_p = doPageWrites;
6471 }
static bool doPageWrites
Definition: xlog.c:286

References doPageWrites, and RedoRecPtr.

Referenced by XLogCheckBufferNeedsBackup(), and XLogInsert().

◆ GetInsertRecPtr()

XLogRecPtr GetInsertRecPtr ( void  )

Definition at line 6482 of file xlog.c.

6483 {
6484  XLogRecPtr recptr;
6485 
6487  recptr = XLogCtl->LogwrtRqst.Write;
6489 
6490  return recptr;
6491 }

References XLogCtlData::info_lck, XLogCtlData::LogwrtRqst, SpinLockAcquire, SpinLockRelease, XLogwrtRqst::Write, and XLogCtl.

Referenced by CheckpointerMain(), gistvacuumscan(), and IsCheckpointOnSchedule().

◆ GetLastImportantRecPtr()

XLogRecPtr GetLastImportantRecPtr ( void  )

Definition at line 6556 of file xlog.c.

6557 {
6559  int i;
6560 
6561  for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
6562  {
6563  XLogRecPtr last_important;
6564 
6565  /*
6566  * Need to take a lock to prevent torn reads of the LSN, which are
6567  * possible on some of the supported platforms. WAL insert locks only
6568  * support exclusive mode, so we have to use that.
6569  */
6571  last_important = WALInsertLocks[i].l.lastImportantAt;
6572  LWLockRelease(&WALInsertLocks[i].l.lock);
6573 
6574  if (res < last_important)
6575  res = last_important;
6576  }
6577 
6578  return res;
6579 }
int i
Definition: isn.c:72
XLogRecPtr lastImportantAt
Definition: xlog.c:371
WALInsertLock l
Definition: xlog.c:383
static WALInsertLockPadded * WALInsertLocks
Definition: xlog.c:569
#define NUM_XLOGINSERT_LOCKS
Definition: xlog.c:150

References i, InvalidXLogRecPtr, WALInsertLockPadded::l, WALInsertLock::lastImportantAt, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), NUM_XLOGINSERT_LOCKS, res, and WALInsertLocks.

Referenced by BackgroundWriterMain(), CheckArchiveTimeout(), and CreateCheckPoint().

◆ GetLastSegSwitchData()

pg_time_t GetLastSegSwitchData ( XLogRecPtr lastSwitchLSN)

Definition at line 6585 of file xlog.c.

6586 {
6587  pg_time_t result;
6588 
6589  /* Need WALWriteLock, but shared lock is sufficient */
6590  LWLockAcquire(WALWriteLock, LW_SHARED);
6591  result = XLogCtl->lastSegSwitchTime;
6592  *lastSwitchLSN = XLogCtl->lastSegSwitchLSN;
6593  LWLockRelease(WALWriteLock);
6594 
6595  return result;
6596 }
pg_time_t lastSegSwitchTime
Definition: xlog.c:467
XLogRecPtr lastSegSwitchLSN
Definition: xlog.c:468

References XLogCtlData::lastSegSwitchLSN, XLogCtlData::lastSegSwitchTime, LW_SHARED, LWLockAcquire(), LWLockRelease(), and XLogCtl.

Referenced by CheckArchiveTimeout().

◆ GetMockAuthenticationNonce()

char* GetMockAuthenticationNonce ( void  )

Definition at line 4578 of file xlog.c.

4579 {
4580  Assert(ControlFile != NULL);
4582 }
char mock_authentication_nonce[MOCK_AUTH_NONCE_LEN]
Definition: pg_control.h:229

References Assert, ControlFile, and ControlFileData::mock_authentication_nonce.

Referenced by scram_mock_salt().

◆ GetOldestRestartPoint()

void GetOldestRestartPoint ( XLogRecPtr oldrecptr,
TimeLineID oldtli 
)

Definition at line 9463 of file xlog.c.

9464 {
9465  LWLockAcquire(ControlFileLock, LW_SHARED);
9466  *oldrecptr = ControlFile->checkPointCopy.redo;
9468  LWLockRelease(ControlFileLock);
9469 }

References ControlFileData::checkPointCopy, ControlFile, LW_SHARED, LWLockAcquire(), LWLockRelease(), CheckPoint::redo, and CheckPoint::ThisTimeLineID.

Referenced by ExecuteRecoveryCommand(), and RestoreArchivedFile().

◆ GetRecoveryState()

RecoveryState GetRecoveryState ( void  )

Definition at line 6370 of file xlog.c.

6371 {
6372  RecoveryState retval;
6373 
6375  retval = XLogCtl->SharedRecoveryState;
6377 
6378  return retval;
6379 }
RecoveryState
Definition: xlog.h:90

References XLogCtlData::info_lck, XLogCtlData::SharedRecoveryState, SpinLockAcquire, SpinLockRelease, and XLogCtl.

Referenced by XLogArchiveCheckDone().

◆ GetRedoRecPtr()

XLogRecPtr GetRedoRecPtr ( void  )

Definition at line 6437 of file xlog.c.

6438 {
6439  XLogRecPtr ptr;
6440 
6441  /*
6442  * The possibly not up-to-date copy in XlogCtl is enough. Even if we
6443  * grabbed a WAL insertion lock to read the authoritative value in
6444  * Insert->RedoRecPtr, someone might update it just after we've released
6445  * the lock.
6446  */
6448  ptr = XLogCtl->RedoRecPtr;
6450 
6451  if (RedoRecPtr < ptr)
6452  RedoRecPtr = ptr;
6453 
6454  return RedoRecPtr;
6455 }

References XLogCtlData::info_lck, RedoRecPtr, XLogCtlData::RedoRecPtr, SpinLockAcquire, SpinLockRelease, and XLogCtl.

Referenced by CheckPointLogicalRewriteHeap(), CheckPointSnapBuild(), MaybeRemoveOldWalSummaries(), nextval_internal(), pgstat_before_server_shutdown(), ReplicationSlotReserveWal(), smgr_bulk_finish(), smgr_bulk_start_smgr(), XLogPageRead(), XLogSaveBufferForHint(), and XLogWrite().

◆ GetSystemIdentifier()

uint64 GetSystemIdentifier ( void  )

◆ GetWALAvailability()

WALAvailability GetWALAvailability ( XLogRecPtr  targetLSN)

Definition at line 7869 of file xlog.c.

7870 {
7871  XLogRecPtr currpos; /* current write LSN */
7872  XLogSegNo currSeg; /* segid of currpos */
7873  XLogSegNo targetSeg; /* segid of targetLSN */
7874  XLogSegNo oldestSeg; /* actual oldest segid */
7875  XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
7876  XLogSegNo oldestSlotSeg; /* oldest segid kept by slot */
7877  uint64 keepSegs;
7878 
7879  /*
7880  * slot does not reserve WAL. Either deactivated, or has never been active
7881  */
7882  if (XLogRecPtrIsInvalid(targetLSN))
7883  return WALAVAIL_INVALID_LSN;
7884 
7885  /*
7886  * Calculate the oldest segment currently reserved by all slots,
7887  * considering wal_keep_size and max_slot_wal_keep_size. Initialize
7888  * oldestSlotSeg to the current segment.
7889  */
7890  currpos = GetXLogWriteRecPtr();
7891  XLByteToSeg(currpos, oldestSlotSeg, wal_segment_size);
7892  KeepLogSeg(currpos, &oldestSlotSeg);
7893 
7894  /*
7895  * Find the oldest extant segment file. We get 1 until checkpoint removes
7896  * the first WAL segment file since startup, which causes the status being
7897  * wrong under certain abnormal conditions but that doesn't actually harm.
7898  */
7899  oldestSeg = XLogGetLastRemovedSegno() + 1;
7900 
7901  /* calculate oldest segment by max_wal_size */
7902  XLByteToSeg(currpos, currSeg, wal_segment_size);
7904 
7905  if (currSeg > keepSegs)
7906  oldestSegMaxWalSize = currSeg - keepSegs;
7907  else
7908  oldestSegMaxWalSize = 1;
7909 
7910  /* the segment we care about */
7911  XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
7912 
7913  /*
7914  * No point in returning reserved or extended status values if the
7915  * targetSeg is known to be lost.
7916  */
7917  if (targetSeg >= oldestSlotSeg)
7918  {
7919  /* show "reserved" when targetSeg is within max_wal_size */
7920  if (targetSeg >= oldestSegMaxWalSize)
7921  return WALAVAIL_RESERVED;
7922 
7923  /* being retained by slots exceeding max_wal_size */
7924  return WALAVAIL_EXTENDED;
7925  }
7926 
7927  /* WAL segments are no longer retained but haven't been removed yet */
7928  if (targetSeg >= oldestSeg)
7929  return WALAVAIL_UNRESERVED;
7930 
7931  /* Definitely lost */
7932  return WALAVAIL_REMOVED;
7933 }
XLogSegNo XLogGetLastRemovedSegno(void)
Definition: xlog.c:3758
XLogRecPtr GetXLogWriteRecPtr(void)
Definition: xlog.c:9451
@ WALAVAIL_REMOVED
Definition: xlog.h:194
@ WALAVAIL_RESERVED
Definition: xlog.h:190
@ WALAVAIL_UNRESERVED
Definition: xlog.h:193
@ WALAVAIL_EXTENDED
Definition: xlog.h:191
@ WALAVAIL_INVALID_LSN
Definition: xlog.h:189

References ConvertToXSegs, GetXLogWriteRecPtr(), KeepLogSeg(), max_wal_size_mb, wal_segment_size, WALAVAIL_EXTENDED, WALAVAIL_INVALID_LSN, WALAVAIL_REMOVED, WALAVAIL_RESERVED, WALAVAIL_UNRESERVED, XLByteToSeg, XLogGetLastRemovedSegno(), and XLogRecPtrIsInvalid.

Referenced by pg_get_replication_slots().

◆ GetWALInsertionTimeLine()

TimeLineID GetWALInsertionTimeLine ( void  )

Definition at line 6520 of file xlog.c.

6521 {
6523 
6524  /* Since the value can't be changing, no lock is required. */
6525  return XLogCtl->InsertTimeLineID;
6526 }

References Assert, XLogCtlData::InsertTimeLineID, RECOVERY_STATE_DONE, XLogCtlData::SharedRecoveryState, and XLogCtl.

Referenced by logical_read_xlog_page(), pg_walfile_name(), pg_walfile_name_offset(), ReadReplicationSlot(), WALReadFromBuffers(), and XLogSendPhysical().

◆ GetWALInsertionTimeLineIfSet()

TimeLineID GetWALInsertionTimeLineIfSet ( void  )

Definition at line 6536 of file xlog.c.

6537 {
6538  TimeLineID insertTLI;
6539 
6541  insertTLI = XLogCtl->InsertTimeLineID;
6543 
6544  return insertTLI;
6545 }

References XLogCtlData::info_lck, XLogCtlData::InsertTimeLineID, SpinLockAcquire, SpinLockRelease, and XLogCtl.

Referenced by GetLatestLSN().

◆ GetXLogBuffer()

static char * GetXLogBuffer ( XLogRecPtr  ptr,
TimeLineID  tli 
)
static

Definition at line 1632 of file xlog.c.

1633 {
1634  int idx;
1635  XLogRecPtr endptr;
1636  static uint64 cachedPage = 0;
1637  static char *cachedPos = NULL;
1638  XLogRecPtr expectedEndPtr;
1639 
1640  /*
1641  * Fast path for the common case that we need to access again the same
1642  * page as last time.
1643  */
1644  if (ptr / XLOG_BLCKSZ == cachedPage)
1645  {
1646  Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1647  Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1648  return cachedPos + ptr % XLOG_BLCKSZ;
1649  }
1650 
1651  /*
1652  * The XLog buffer cache is organized so that a page is always loaded to a
1653  * particular buffer. That way we can easily calculate the buffer a given
1654  * page must be loaded into, from the XLogRecPtr alone.
1655  */
1656  idx = XLogRecPtrToBufIdx(ptr);
1657 
1658  /*
1659  * See what page is loaded in the buffer at the moment. It could be the
1660  * page we're looking for, or something older. It can't be anything newer
1661  * - that would imply the page we're looking for has already been written
1662  * out to disk and evicted, and the caller is responsible for making sure
1663  * that doesn't happen.
1664  *
1665  * We don't hold a lock while we read the value. If someone is just about
1666  * to initialize or has just initialized the page, it's possible that we
1667  * get InvalidXLogRecPtr. That's ok, we'll grab the mapping lock (in
1668  * AdvanceXLInsertBuffer) and retry if we see anything other than the page
1669  * we're looking for.
1670  */
1671  expectedEndPtr = ptr;
1672  expectedEndPtr += XLOG_BLCKSZ - ptr % XLOG_BLCKSZ;
1673 
1674  endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1675  if (expectedEndPtr != endptr)
1676  {
1677  XLogRecPtr initializedUpto;
1678 
1679  /*
1680  * Before calling AdvanceXLInsertBuffer(), which can block, let others
1681  * know how far we're finished with inserting the record.
1682  *
1683  * NB: If 'ptr' points to just after the page header, advertise a
1684  * position at the beginning of the page rather than 'ptr' itself. If
1685  * there are no other insertions running, someone might try to flush
1686  * up to our advertised location. If we advertised a position after
1687  * the page header, someone might try to flush the page header, even
1688  * though page might actually not be initialized yet. As the first
1689  * inserter on the page, we are effectively responsible for making
1690  * sure that it's initialized, before we let insertingAt to move past
1691  * the page header.
1692  */
1693  if (ptr % XLOG_BLCKSZ == SizeOfXLogShortPHD &&
1694  XLogSegmentOffset(ptr, wal_segment_size) > XLOG_BLCKSZ)
1695  initializedUpto = ptr - SizeOfXLogShortPHD;
1696  else if (ptr % XLOG_BLCKSZ == SizeOfXLogLongPHD &&
1697  XLogSegmentOffset(ptr, wal_segment_size) < XLOG_BLCKSZ)
1698  initializedUpto = ptr - SizeOfXLogLongPHD;
1699  else
1700  initializedUpto = ptr;
1701 
1702  WALInsertLockUpdateInsertingAt(initializedUpto);
1703 
1704  AdvanceXLInsertBuffer(ptr, tli, false);
1705  endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1706 
1707  if (expectedEndPtr != endptr)
1708  elog(PANIC, "could not find WAL buffer for %X/%X",
1709  LSN_FORMAT_ARGS(ptr));
1710  }
1711  else
1712  {
1713  /*
1714  * Make sure the initialization of the page is visible to us, and
1715  * won't arrive later to overwrite the WAL data we write on the page.
1716  */
1718  }
1719 
1720  /*
1721  * Found the buffer holding this page. Return a pointer to the right
1722  * offset within the page.
1723  */
1724  cachedPage = ptr / XLOG_BLCKSZ;
1725  cachedPos = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1726 
1727  Assert(((XLogPageHeader) cachedPos)->xlp_magic == XLOG_PAGE_MAGIC);
1728  Assert(((XLogPageHeader) cachedPos)->xlp_pageaddr == ptr - (ptr % XLOG_BLCKSZ));
1729 
1730  return cachedPos + ptr % XLOG_BLCKSZ;
1731 }
Datum idx(PG_FUNCTION_ARGS)
Definition: _int_op.c:259
#define pg_memory_barrier()
Definition: atomics.h:143
static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt)
Definition: xlog.c:1471
static void AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
Definition: xlog.c:1985

References AdvanceXLInsertBuffer(), Assert, elog, idx(), LSN_FORMAT_ARGS, XLogCtlData::pages, PANIC, pg_atomic_read_u64(), pg_memory_barrier, SizeOfXLogLongPHD, SizeOfXLogShortPHD, wal_segment_size, WALInsertLockUpdateInsertingAt(), XLogCtlData::xlblocks, XLOG_PAGE_MAGIC, XLogCtl, XLogRecPtrToBufIdx, and XLogSegmentOffset.

Referenced by CopyXLogRecordToWAL(), and CreateOverwriteContrecordRecord().

◆ GetXLogInsertRecPtr()

XLogRecPtr GetXLogInsertRecPtr ( void  )

Definition at line 9435 of file xlog.c.

9436 {
9438  uint64 current_bytepos;
9439 
9440  SpinLockAcquire(&Insert->insertpos_lck);
9441  current_bytepos = Insert->CurrBytePos;
9442  SpinLockRelease(&Insert->insertpos_lck);
9443 
9444  return XLogBytePosToRecPtr(current_bytepos);
9445 }

References XLogCtlData::Insert, Insert(), SpinLockAcquire, SpinLockRelease, XLogBytePosToRecPtr(), and XLogCtl.

Referenced by CreateOverwriteContrecordRecord(), gistGetFakeLSN(), logical_begin_heap_rewrite(), pg_current_wal_insert_lsn(), and ReplicationSlotReserveWal().

◆ GetXLogWriteRecPtr()

XLogRecPtr GetXLogWriteRecPtr ( void  )

◆ InitControlFile()

static void InitControlFile ( uint64  sysidentifier,
uint32  data_checksum_version 
)
static

Definition at line 4204 of file xlog.c.

4205 {
4206  char mock_auth_nonce[MOCK_AUTH_NONCE_LEN];
4207 
4208  /*
4209  * Generate a random nonce. This is used for authentication requests that
4210  * will fail because the user does not exist. The nonce is used to create
4211  * a genuine-looking password challenge for the non-existent user, in lieu
4212  * of an actual stored password.
4213  */
4214  if (!pg_strong_random(mock_auth_nonce, MOCK_AUTH_NONCE_LEN))
4215  ereport(PANIC,
4216  (errcode(ERRCODE_INTERNAL_ERROR),
4217  errmsg("could not generate secret authorization token")));
4218 
4219  memset(ControlFile, 0, sizeof(ControlFileData));
4220  /* Initialize pg_control status fields */
4221  ControlFile->system_identifier = sysidentifier;
4222  memcpy(ControlFile->mock_authentication_nonce, mock_auth_nonce, MOCK_AUTH_NONCE_LEN);
4225 
4226  /* Set important parameter values for use when replaying WAL */
4235  ControlFile->data_checksum_version = data_checksum_version;
4236 }
bool track_commit_timestamp
Definition: commit_ts.c:109
#define MOCK_AUTH_NONCE_LEN
Definition: pg_control.h:28
bool pg_strong_random(void *buf, size_t len)
bool track_commit_timestamp
Definition: pg_control.h:185
bool wal_log_hints
Definition: xlog.c:123
#define FirstNormalUnloggedLSN
Definition: xlogdefs.h:36

References ControlFile, ControlFileData::data_checksum_version, DB_SHUTDOWNED, ereport, errcode(), errmsg(), FirstNormalUnloggedLSN, max_locks_per_xact, ControlFileData::max_locks_per_xact, max_prepared_xacts, ControlFileData::max_prepared_xacts, max_wal_senders, ControlFileData::max_wal_senders, max_worker_processes, ControlFileData::max_worker_processes, MaxConnections, ControlFileData::MaxConnections, MOCK_AUTH_NONCE_LEN, ControlFileData::mock_authentication_nonce, PANIC, pg_strong_random(), ControlFileData::state, ControlFileData::system_identifier, track_commit_timestamp, ControlFileData::track_commit_timestamp, ControlFileData::unloggedLSN, wal_level, ControlFileData::wal_level, wal_log_hints, and ControlFileData::wal_log_hints.

Referenced by BootStrapXLOG().

◆ InitializeWalConsistencyChecking()

void InitializeWalConsistencyChecking ( void  )

Definition at line 4784 of file xlog.c.

4785 {
4787 
4789  {
4790  struct config_generic *guc;
4791 
4792  guc = find_option("wal_consistency_checking", false, false, ERROR);
4793 
4795 
4796  set_config_option_ext("wal_consistency_checking",
4798  guc->scontext, guc->source, guc->srole,
4799  GUC_ACTION_SET, true, ERROR, false);
4800 
4801  /* checking should not be deferred again */
4803  }
4804 }
struct config_generic * find_option(const char *name, bool create_placeholders, bool skip_errors, int elevel)
Definition: guc.c:1235
int set_config_option_ext(const char *name, const char *value, GucContext context, GucSource source, Oid srole, GucAction action, bool changeVal, int elevel, bool is_reload)
Definition: guc.c:3382
@ GUC_ACTION_SET
Definition: guc.h:199
GucContext scontext
Definition: guc_tables.h:167
GucSource source
Definition: guc_tables.h:165
char * wal_consistency_checking_string
Definition: xlog.c:125

References Assert, check_wal_consistency_checking_deferred, ERROR, find_option(), GUC_ACTION_SET, process_shared_preload_libraries_done, config_generic::scontext, set_config_option_ext(), config_generic::source, config_generic::srole, and wal_consistency_checking_string.

Referenced by PostgresSingleUserMain(), and PostmasterMain().

◆ InstallXLogFileSegment()

static bool InstallXLogFileSegment ( XLogSegNo segno,
char *  tmppath,
bool  find_free,
XLogSegNo  max_segno,
TimeLineID  tli 
)
static

Definition at line 3563 of file xlog.c.

3565 {
3566  char path[MAXPGPATH];
3567  struct stat stat_buf;
3568 
3569  Assert(tli != 0);
3570 
3571  XLogFilePath(path, tli, *segno, wal_segment_size);
3572 
3573  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
3575  {
3576  LWLockRelease(ControlFileLock);
3577  return false;
3578  }
3579 
3580  if (!find_free)
3581  {
3582  /* Force installation: get rid of any pre-existing segment file */
3583  durable_unlink(path, DEBUG1);
3584  }
3585  else
3586  {
3587  /* Find a free slot to put it in */
3588  while (stat(path, &stat_buf) == 0)
3589  {
3590  if ((*segno) >= max_segno)
3591  {
3592  /* Failed to find a free slot within specified range */
3593  LWLockRelease(ControlFileLock);
3594  return false;
3595  }
3596  (*segno)++;
3597  XLogFilePath(path, tli, *segno, wal_segment_size);
3598  }
3599  }
3600 
3601  Assert(access(path, F_OK) != 0 && errno == ENOENT);
3602  if (durable_rename(tmppath, path, LOG) != 0)
3603  {
3604  LWLockRelease(ControlFileLock);
3605  /* durable_rename already emitted log message */
3606  return false;
3607  }
3608 
3609  LWLockRelease(ControlFileLock);
3610 
3611  return true;
3612 }
int durable_unlink(const char *fname, int elevel)
Definition: fd.c:871
short access
Definition: preproc-type.c:36
bool InstallXLogFileSegmentActive
Definition: xlog.c:526
#define stat
Definition: win32_port.h:284

References Assert, DEBUG1, durable_rename(), durable_unlink(), XLogCtlData::InstallXLogFileSegmentActive, LOG, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), MAXPGPATH, stat, wal_segment_size, XLogCtl, and XLogFilePath().

Referenced by RemoveXlogFile(), XLogFileCopy(), and XLogFileInitInternal().

◆ IsInstallXLogFileSegmentActive()

bool IsInstallXLogFileSegmentActive ( void  )

Definition at line 9492 of file xlog.c.

9493 {
9494  bool result;
9495 
9496  LWLockAcquire(ControlFileLock, LW_SHARED);
9498  LWLockRelease(ControlFileLock);
9499 
9500  return result;
9501 }

References XLogCtlData::InstallXLogFileSegmentActive, LW_SHARED, LWLockAcquire(), LWLockRelease(), and XLogCtl.

Referenced by XLogFileRead().

◆ issue_xlog_fsync()

void issue_xlog_fsync ( int  fd,
XLogSegNo  segno,
TimeLineID  tli 
)

Definition at line 8683 of file xlog.c.

8684 {
8685  char *msg = NULL;
8686  instr_time start;
8687 
8688  Assert(tli != 0);
8689 
8690  /*
8691  * Quick exit if fsync is disabled or write() has already synced the WAL
8692  * file.
8693  */
8694  if (!enableFsync ||
8697  return;
8698 
8699  /* Measure I/O timing to sync the WAL file */
8700  if (track_wal_io_timing)
8702  else
8704 
8705  pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
8706  switch (wal_sync_method)
8707  {
8708  case WAL_SYNC_METHOD_FSYNC:
8709  if (pg_fsync_no_writethrough(fd) != 0)
8710  msg = _("could not fsync file \"%s\": %m");
8711  break;
8712 #ifdef HAVE_FSYNC_WRITETHROUGH
8714  if (pg_fsync_writethrough(fd) != 0)
8715  msg = _("could not fsync write-through file \"%s\": %m");
8716  break;
8717 #endif
8719  if (pg_fdatasync(fd) != 0)
8720  msg = _("could not fdatasync file \"%s\": %m");
8721  break;
8722  case WAL_SYNC_METHOD_OPEN:
8724  /* not reachable */
8725  Assert(false);
8726  break;
8727  default:
8728  ereport(PANIC,
8729  errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8730  errmsg_internal("unrecognized \"wal_sync_method\": %d", wal_sync_method));
8731  break;
8732  }
8733 
8734  /* PANIC if failed to fsync */
8735  if (msg)
8736  {
8737  char xlogfname[MAXFNAMELEN];
8738  int save_errno = errno;
8739 
8740  XLogFileName(xlogfname, tli, segno, wal_segment_size);
8741  errno = save_errno;
8742  ereport(PANIC,
8744  errmsg(msg, xlogfname)));
8745  }
8746 
8748 
8749  /*
8750  * Increment the I/O timing and the number of times WAL files were synced.
8751  */
8752  if (track_wal_io_timing)
8753  {
8754  instr_time end;
8755 
8758  }
8759 
8761 }
#define _(x)
Definition: elog.c:90
int pg_fsync_no_writethrough(int fd)
Definition: fd.c:440
int pg_fdatasync(int fd)
Definition: fd.c:479
int pg_fsync_writethrough(int fd)
Definition: fd.c:460
return str start
#define INSTR_TIME_SET_CURRENT(t)
Definition: instr_time.h:122
#define INSTR_TIME_SET_ZERO(t)
Definition: instr_time.h:172
#define INSTR_TIME_ACCUM_DIFF(x, y, z)
Definition: instr_time.h:184
static int fd(const char *x, int i)
Definition: preproc-init.c:105
instr_time wal_sync_time
Definition: pgstat.h:493
PgStat_Counter wal_sync
Definition: pgstat.h:491
bool track_wal_io_timing
Definition: xlog.c:137

References _, Assert, enableFsync, ereport, errcode(), errcode_for_file_access(), errmsg(), errmsg_internal(), fd(), INSTR_TIME_ACCUM_DIFF, INSTR_TIME_SET_CURRENT, INSTR_TIME_SET_ZERO, MAXFNAMELEN, PANIC, PendingWalStats, pg_fdatasync(), pg_fsync_no_writethrough(), pg_fsync_writethrough(), pgstat_report_wait_end(), pgstat_report_wait_start(), start, track_wal_io_timing, wal_segment_size, PgStat_PendingWalStats::wal_sync, wal_sync_method, WAL_SYNC_METHOD_FDATASYNC, WAL_SYNC_METHOD_FSYNC, WAL_SYNC_METHOD_FSYNC_WRITETHROUGH, WAL_SYNC_METHOD_OPEN, WAL_SYNC_METHOD_OPEN_DSYNC, PgStat_PendingWalStats::wal_sync_time, and XLogFileName().

Referenced by XLogWalRcvFlush(), and XLogWrite().

◆ KeepLogSeg()

static void KeepLogSeg ( XLogRecPtr  recptr,
XLogSegNo logSegNo 
)
static

Definition at line 7953 of file xlog.c.

7954 {
7955  XLogSegNo currSegNo;
7956  XLogSegNo segno;
7957  XLogRecPtr keep;
7958 
7959  XLByteToSeg(recptr, currSegNo, wal_segment_size);
7960  segno = currSegNo;
7961 
7962  /*
7963  * Calculate how many segments are kept by slots first, adjusting for
7964  * max_slot_wal_keep_size.
7965  */
7967  if (keep != InvalidXLogRecPtr && keep < recptr)
7968  {
7969  XLByteToSeg(keep, segno, wal_segment_size);
7970 
7971  /* Cap by max_slot_wal_keep_size ... */
7972  if (max_slot_wal_keep_size_mb >= 0)
7973  {
7974  uint64 slot_keep_segs;
7975 
7976  slot_keep_segs =
7978 
7979  if (currSegNo - segno > slot_keep_segs)
7980  segno = currSegNo - slot_keep_segs;
7981  }
7982  }
7983 
7984  /*
7985  * If WAL summarization is in use, don't remove WAL that has yet to be
7986  * summarized.
7987  */
7988  keep = GetOldestUnsummarizedLSN(NULL, NULL);
7989  if (keep != InvalidXLogRecPtr)
7990  {
7991  XLogSegNo unsummarized_segno;
7992 
7993  XLByteToSeg(keep, unsummarized_segno, wal_segment_size);
7994  if (unsummarized_segno < segno)
7995  segno = unsummarized_segno;
7996  }
7997 
7998  /* but, keep at least wal_keep_size if that's set */
7999  if (wal_keep_size_mb > 0)
8000  {
8001  uint64 keep_segs;
8002 
8004  if (currSegNo - segno < keep_segs)
8005  {
8006  /* avoid underflow, don't go below 1 */
8007  if (currSegNo <= keep_segs)
8008  segno = 1;
8009  else
8010  segno = currSegNo - keep_segs;
8011  }
8012  }
8013 
8014  /* don't delete WAL segments newer than the calculated segment */
8015  if (segno < *logSegNo)
8016  *logSegNo = segno;
8017 }
XLogRecPtr GetOldestUnsummarizedLSN(TimeLineID *tli, bool *lsn_is_exact)
int wal_keep_size_mb
Definition: xlog.c:116
static XLogRecPtr XLogGetReplicationSlotMinimumLSN(void)
Definition: xlog.c:2701
int max_slot_wal_keep_size_mb
Definition: xlog.c:135

References ConvertToXSegs, GetOldestUnsummarizedLSN(), InvalidXLogRecPtr, max_slot_wal_keep_size_mb, wal_keep_size_mb, wal_segment_size, XLByteToSeg, and XLogGetReplicationSlotMinimumLSN().

Referenced by CreateCheckPoint(), CreateRestartPoint(), and GetWALAvailability().

◆ LocalProcessControlFile()

void LocalProcessControlFile ( bool  reset)

Definition at line 4846 of file xlog.c.

4847 {
4848  Assert(reset || ControlFile == NULL);
4849  ControlFile = palloc(sizeof(ControlFileData));
4850  ReadControlFile();
4851 }
void reset(void)
Definition: sql-declare.c:600

References Assert, ControlFile, palloc(), ReadControlFile(), and reset().

Referenced by PostgresSingleUserMain(), PostmasterMain(), and PostmasterStateMachine().

◆ LocalSetXLogInsertAllowed()

static int LocalSetXLogInsertAllowed ( void  )
static

Definition at line 6422 of file xlog.c.

6423 {
6424  int oldXLogAllowed = LocalXLogInsertAllowed;
6425 
6427 
6428  return oldXLogAllowed;
6429 }

References LocalXLogInsertAllowed.

Referenced by CreateCheckPoint(), and StartupXLOG().

◆ LogCheckpointEnd()

static void LogCheckpointEnd ( bool  restartpoint)
static

Definition at line 6681 of file xlog.c.

6682 {
6683  long write_msecs,
6684  sync_msecs,
6685  total_msecs,
6686  longest_msecs,
6687  average_msecs;
6688  uint64 average_sync_time;
6689 
6691 
6694 
6697 
6698  /* Accumulate checkpoint timing summary data, in milliseconds. */
6699  PendingCheckpointerStats.write_time += write_msecs;
6700  PendingCheckpointerStats.sync_time += sync_msecs;
6701 
6702  /*
6703  * All of the published timing statistics are accounted for. Only
6704  * continue if a log message is to be written.
6705  */
6706  if (!log_checkpoints)
6707  return;
6708 
6711 
6712  /*
6713  * Timing values returned from CheckpointStats are in microseconds.
6714  * Convert to milliseconds for consistent printing.
6715  */
6716  longest_msecs = (long) ((CheckpointStats.ckpt_longest_sync + 999) / 1000);
6717 
6718  average_sync_time = 0;
6720  average_sync_time = CheckpointStats.ckpt_agg_sync_time /
6722  average_msecs = (long) ((average_sync_time + 999) / 1000);
6723 
6724  /*
6725  * ControlFileLock is not required to see ControlFile->checkPoint and
6726  * ->checkPointCopy here as we are the only updator of those variables at
6727  * this moment.
6728  */
6729  if (restartpoint)
6730  ereport(LOG,
6731  (errmsg("restartpoint complete: wrote %d buffers (%.1f%%), "
6732  "wrote %d SLRU buffers; %d WAL file(s) added, "
6733  "%d removed, %d recycled; write=%ld.%03d s, "
6734  "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6735  "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6736  "estimate=%d kB; lsn=%X/%X, redo lsn=%X/%X",
6738  (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6743  write_msecs / 1000, (int) (write_msecs % 1000),
6744  sync_msecs / 1000, (int) (sync_msecs % 1000),
6745  total_msecs / 1000, (int) (total_msecs % 1000),
6747  longest_msecs / 1000, (int) (longest_msecs % 1000),
6748  average_msecs / 1000, (int) (average_msecs % 1000),
6749  (int) (PrevCheckPointDistance / 1024.0),
6750  (int) (CheckPointDistanceEstimate / 1024.0),
6753  else
6754  ereport(LOG,
6755  (errmsg("checkpoint complete: wrote %d buffers (%.1f%%), "
6756  "wrote %d SLRU buffers; %d WAL file(s) added, "
6757  "%d removed, %d recycled; write=%ld.%03d s, "
6758  "sync=%ld.%03d s, total=%ld.%03d s; sync files=%d, "
6759  "longest=%ld.%03d s, average=%ld.%03d s; distance=%d kB, "
6760  "estimate=%d kB; lsn=%X/%X, redo lsn=%X/%X",
6762  (double) CheckpointStats.ckpt_bufs_written * 100 / NBuffers,
6767  write_msecs / 1000, (int) (write_msecs % 1000),
6768  sync_msecs / 1000, (int) (sync_msecs % 1000),
6769  total_msecs / 1000, (int) (total_msecs % 1000),
6771  longest_msecs / 1000, (int) (longest_msecs % 1000),
6772  average_msecs / 1000, (int) (average_msecs % 1000),
6773  (int) (PrevCheckPointDistance / 1024.0),
6774  (int) (CheckPointDistanceEstimate / 1024.0),
6777 }
long TimestampDifferenceMilliseconds(TimestampTz start_time, TimestampTz stop_time)
Definition: timestamp.c:1756
PgStat_CheckpointerStats PendingCheckpointerStats
uint64 ckpt_agg_sync_time
Definition: xlog.h:176
uint64 ckpt_longest_sync
Definition: xlog.h:175
TimestampTz ckpt_end_t
Definition: xlog.h:165
int ckpt_slru_written
Definition: xlog.h:168
int ckpt_sync_rels
Definition: xlog.h:174
PgStat_Counter sync_time
Definition: pgstat.h:302
PgStat_Counter write_time
Definition: pgstat.h:301
static double CheckPointDistanceEstimate
Definition: xlog.c:159
static double PrevCheckPointDistance
Definition: xlog.c:160

References ControlFileData::checkPoint, ControlFileData::checkPointCopy, CheckPointDistanceEstimate, CheckpointStats, CheckpointStatsData::ckpt_agg_sync_time, CheckpointStatsData::ckpt_bufs_written, CheckpointStatsData::ckpt_end_t, CheckpointStatsData::ckpt_longest_sync, CheckpointStatsData::ckpt_segs_added, CheckpointStatsData::ckpt_segs_recycled, CheckpointStatsData::ckpt_segs_removed, CheckpointStatsData::ckpt_slru_written, CheckpointStatsData::ckpt_start_t, CheckpointStatsData::ckpt_sync_end_t, CheckpointStatsData::ckpt_sync_rels, CheckpointStatsData::ckpt_sync_t, CheckpointStatsData::ckpt_write_t, ControlFile, ereport, errmsg(), GetCurrentTimestamp(), LOG, log_checkpoints, LSN_FORMAT_ARGS, NBuffers, PendingCheckpointerStats, PrevCheckPointDistance, CheckPoint::redo, PgStat_CheckpointerStats::sync_time, TimestampDifferenceMilliseconds(), and PgStat_CheckpointerStats::write_time.

Referenced by CreateCheckPoint(), and CreateRestartPoint().

◆ LogCheckpointStart()

static void LogCheckpointStart ( int  flags,
bool  restartpoint 
)
static

Definition at line 6649 of file xlog.c.

6650 {
6651  if (restartpoint)
6652  ereport(LOG,
6653  /* translator: the placeholders show checkpoint options */
6654  (errmsg("restartpoint starting:%s%s%s%s%s%s%s%s",
6655  (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6656  (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6657  (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
6658  (flags & CHECKPOINT_FORCE) ? " force" : "",
6659  (flags & CHECKPOINT_WAIT) ? " wait" : "",
6660  (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6661  (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6662  (flags & CHECKPOINT_FLUSH_ALL) ? " flush-all" : "")));
6663  else
6664  ereport(LOG,
6665  /* translator: the placeholders show checkpoint options */
6666  (errmsg("checkpoint starting:%s%s%s%s%s%s%s%s",
6667  (flags & CHECKPOINT_IS_SHUTDOWN) ? " shutdown" : "",
6668  (flags & CHECKPOINT_END_OF_RECOVERY) ? " end-of-recovery" : "",
6669  (flags & CHECKPOINT_IMMEDIATE) ? " immediate" : "",
6670  (flags & CHECKPOINT_FORCE) ? " force" : "",
6671  (flags & CHECKPOINT_WAIT) ? " wait" : "",
6672  (flags & CHECKPOINT_CAUSE_XLOG) ? " wal" : "",
6673  (flags & CHECKPOINT_CAUSE_TIME) ? " time" : "",
6674  (flags & CHECKPOINT_FLUSH_ALL) ? " flush-all" : "")));
6675 }
#define CHECKPOINT_CAUSE_XLOG
Definition: xlog.h:148
#define CHECKPOINT_FLUSH_ALL
Definition: xlog.h:143
#define CHECKPOINT_CAUSE_TIME
Definition: xlog.h:149

References CHECKPOINT_CAUSE_TIME, CHECKPOINT_CAUSE_XLOG, CHECKPOINT_END_OF_RECOVERY, CHECKPOINT_FLUSH_ALL, CHECKPOINT_FORCE, CHECKPOINT_IMMEDIATE, CHECKPOINT_IS_SHUTDOWN, CHECKPOINT_WAIT, ereport, errmsg(), and LOG.

Referenced by CreateCheckPoint(), and CreateRestartPoint().

◆ PerformRecoveryXLogAction()

static bool PerformRecoveryXLogAction ( void  )
static

Definition at line 6284 of file xlog.c.

6285 {
6286  bool promoted = false;
6287 
6288  /*
6289  * Perform a checkpoint to update all our recovery activity to disk.
6290  *
6291  * Note that we write a shutdown checkpoint rather than an on-line one.
6292  * This is not particularly critical, but since we may be assigning a new
6293  * TLI, using a shutdown checkpoint allows us to have the rule that TLI
6294  * only changes in shutdown checkpoints, which allows some extra error
6295  * checking in xlog_redo.
6296  *
6297  * In promotion, only create a lightweight end-of-recovery record instead
6298  * of a full checkpoint. A checkpoint is requested later, after we're
6299  * fully out of recovery mode and already accepting queries.
6300  */
6303  {
6304  promoted = true;
6305 
6306  /*
6307  * Insert a special WAL record to mark the end of recovery, since we
6308  * aren't doing a checkpoint. That means that the checkpointer process
6309  * may likely be in the middle of a time-smoothed restartpoint and
6310  * could continue to be for minutes after this. That sounds strange,
6311  * but the effect is roughly the same and it would be stranger to try
6312  * to come out of the restartpoint and then checkpoint. We request a
6313  * checkpoint later anyway, just for safety.
6314  */
6316  }
6317  else
6318  {
6321  CHECKPOINT_WAIT);
6322  }
6323 
6324  return promoted;
6325 }
static void CreateEndOfRecoveryRecord(void)
Definition: xlog.c:7375
bool PromoteIsTriggered(void)

References ArchiveRecoveryRequested, CHECKPOINT_END_OF_RECOVERY, CHECKPOINT_IMMEDIATE, CHECKPOINT_WAIT, CreateEndOfRecoveryRecord(), IsUnderPostmaster, PromoteIsTriggered(), and RequestCheckpoint().

Referenced by StartupXLOG().

◆ PreallocXlogFiles()

static void PreallocXlogFiles ( XLogRecPtr  endptr,
TimeLineID  tli 
)
static

Definition at line 3690 of file xlog.c.

3691 {
3692  XLogSegNo _logSegNo;
3693  int lf;
3694  bool added;
3695  char path[MAXPGPATH];
3696  uint64 offset;
3697 
3699  return; /* unlocked check says no */
3700 
3701  XLByteToPrevSeg(endptr, _logSegNo, wal_segment_size);
3702  offset = XLogSegmentOffset(endptr - 1, wal_segment_size);
3703  if (offset >= (uint32) (0.75 * wal_segment_size))
3704  {
3705  _logSegNo++;
3706  lf = XLogFileInitInternal(_logSegNo, tli, &added, path);
3707  if (lf >= 0)
3708  close(lf);
3709  if (added)
3711  }
3712 }
static int XLogFileInitInternal(XLogSegNo logsegno, TimeLineID logtli, bool *added, char *path)
Definition: xlog.c:3210

References CheckpointStats, CheckpointStatsData::ckpt_segs_added, close, XLogCtlData::InstallXLogFileSegmentActive, MAXPGPATH, wal_segment_size, XLByteToPrevSeg, XLogCtl, XLogFileInitInternal(), and XLogSegmentOffset.

Referenced by CreateCheckPoint(), CreateRestartPoint(), and StartupXLOG().

◆ ReachedEndOfBackup()

void ReachedEndOfBackup ( XLogRecPtr  EndRecPtr,
TimeLineID  tli 
)

Definition at line 6247 of file xlog.c.

6248 {
6249  /*
6250  * We have reached the end of base backup, as indicated by pg_control. The
6251  * data on disk is now consistent (unless minRecoveryPoint is further
6252  * ahead, which can happen if we crashed during previous recovery). Reset
6253  * backupStartPoint and backupEndPoint, and update minRecoveryPoint to
6254  * make sure we don't allow starting up at an earlier point even if
6255  * recovery is stopped and restarted soon after this.
6256  */
6257  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6258 
6259  if (ControlFile->minRecoveryPoint < EndRecPtr)
6260  {
6261  ControlFile->minRecoveryPoint = EndRecPtr;
6263  }
6264 
6267  ControlFile->backupEndRequired = false;
6269 
6270  LWLockRelease(ControlFileLock);
6271 }
XLogRecPtr backupStartPoint
Definition: pg_control.h:170
bool backupEndRequired
Definition: pg_control.h:172
XLogRecPtr backupEndPoint
Definition: pg_control.h:171

References ControlFileData::backupEndPoint, ControlFileData::backupEndRequired, ControlFileData::backupStartPoint, ControlFile, InvalidXLogRecPtr, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, and UpdateControlFile().

Referenced by CheckRecoveryConsistency().

◆ ReadControlFile()

static void ReadControlFile ( void  )
static

Definition at line 4321 of file xlog.c.

4322 {
4323  pg_crc32c crc;
4324  int fd;
4325  char wal_segsz_str[20];
4326  int r;
4327 
4328  /*
4329  * Read data...
4330  */
4332  O_RDWR | PG_BINARY);
4333  if (fd < 0)
4334  ereport(PANIC,
4336  errmsg("could not open file \"%s\": %m",
4337  XLOG_CONTROL_FILE)));
4338 
4339  pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_READ);
4340  r = read(fd, ControlFile, sizeof(ControlFileData));
4341  if (r != sizeof(ControlFileData))
4342  {
4343  if (r < 0)
4344  ereport(PANIC,
4346  errmsg("could not read file \"%s\": %m",
4347  XLOG_CONTROL_FILE)));
4348  else
4349  ereport(PANIC,
4351  errmsg("could not read file \"%s\": read %d of %zu",
4352  XLOG_CONTROL_FILE, r, sizeof(ControlFileData))));
4353  }
4355 
4356  close(fd);
4357 
4358  /*
4359  * Check for expected pg_control format version. If this is wrong, the
4360  * CRC check will likely fail because we'll be checking the wrong number
4361  * of bytes. Complaining about wrong version will probably be more
4362  * enlightening than complaining about wrong CRC.
4363  */
4364 
4366  ereport(FATAL,
4367  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4368  errmsg("database files are incompatible with server"),
4369  errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d (0x%08x),"
4370  " but the server was compiled with PG_CONTROL_VERSION %d (0x%08x).",
4373  errhint("This could be a problem of mismatched byte ordering. It looks like you need to initdb.")));
4374 
4376  ereport(FATAL,
4377  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4378  errmsg("database files are incompatible with server"),
4379  errdetail("The database cluster was initialized with PG_CONTROL_VERSION %d,"
4380  " but the server was compiled with PG_CONTROL_VERSION %d.",
4382  errhint("It looks like you need to initdb.")));
4383 
4384  /* Now check the CRC. */
4385  INIT_CRC32C(crc);
4386  COMP_CRC32C(crc,
4387  (char *) ControlFile,
4388  offsetof(ControlFileData, crc));
4389  FIN_CRC32C(crc);
4390 
4391  if (!EQ_CRC32C(crc, ControlFile->crc))
4392  ereport(FATAL,
4393  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4394  errmsg("incorrect checksum in control file")));
4395 
4396  /*
4397  * Do compatibility checking immediately. If the database isn't
4398  * compatible with the backend executable, we want to abort before we can
4399  * possibly do any damage.
4400  */
4402  ereport(FATAL,
4403  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4404  errmsg("database files are incompatible with server"),
4405  /* translator: %s is a variable name and %d is its value */
4406  errdetail("The database cluster was initialized with %s %d,"
4407  " but the server was compiled with %s %d.",
4408  "CATALOG_VERSION_NO", ControlFile->catalog_version_no,
4409  "CATALOG_VERSION_NO", CATALOG_VERSION_NO),
4410  errhint("It looks like you need to initdb.")));
4411  if (ControlFile->maxAlign != MAXIMUM_ALIGNOF)
4412  ereport(FATAL,
4413  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4414  errmsg("database files are incompatible with server"),
4415  /* translator: %s is a variable name and %d is its value */
4416  errdetail("The database cluster was initialized with %s %d,"
4417  " but the server was compiled with %s %d.",
4418  "MAXALIGN", ControlFile->maxAlign,
4419  "MAXALIGN", MAXIMUM_ALIGNOF),
4420  errhint("It looks like you need to initdb.")));
4422  ereport(FATAL,
4423  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4424  errmsg("database files are incompatible with server"),
4425  errdetail("The database cluster appears to use a different floating-point number format than the server executable."),
4426  errhint("It looks like you need to initdb.")));
4427  if (ControlFile->blcksz != BLCKSZ)
4428  ereport(FATAL,
4429  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4430  errmsg("database files are incompatible with server"),
4431  /* translator: %s is a variable name and %d is its value */
4432  errdetail("The database cluster was initialized with %s %d,"
4433  " but the server was compiled with %s %d.",
4434  "BLCKSZ", ControlFile->blcksz,
4435  "BLCKSZ", BLCKSZ),
4436  errhint("It looks like you need to recompile or initdb.")));
4437  if (ControlFile->relseg_size != RELSEG_SIZE)
4438  ereport(FATAL,
4439  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4440  errmsg("database files are incompatible with server"),
4441  /* translator: %s is a variable name and %d is its value */
4442  errdetail("The database cluster was initialized with %s %d,"
4443  " but the server was compiled with %s %d.",
4444  "RELSEG_SIZE", ControlFile->relseg_size,
4445  "RELSEG_SIZE", RELSEG_SIZE),
4446  errhint("It looks like you need to recompile or initdb.")));
4447  if (ControlFile->xlog_blcksz != XLOG_BLCKSZ)
4448  ereport(FATAL,
4449  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4450  errmsg("database files are incompatible with server"),
4451  /* translator: %s is a variable name and %d is its value */
4452  errdetail("The database cluster was initialized with %s %d,"
4453  " but the server was compiled with %s %d.",
4454  "XLOG_BLCKSZ", ControlFile->xlog_blcksz,
4455  "XLOG_BLCKSZ", XLOG_BLCKSZ),
4456  errhint("It looks like you need to recompile or initdb.")));
4458  ereport(FATAL,
4459  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4460  errmsg("database files are incompatible with server"),
4461  /* translator: %s is a variable name and %d is its value */
4462  errdetail("The database cluster was initialized with %s %d,"
4463  " but the server was compiled with %s %d.",
4464  "NAMEDATALEN", ControlFile->nameDataLen,
4465  "NAMEDATALEN", NAMEDATALEN),
4466  errhint("It looks like you need to recompile or initdb.")));
4468  ereport(FATAL,
4469  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4470  errmsg("database files are incompatible with server"),
4471  /* translator: %s is a variable name and %d is its value */
4472  errdetail("The database cluster was initialized with %s %d,"
4473  " but the server was compiled with %s %d.",
4474  "INDEX_MAX_KEYS", ControlFile->indexMaxKeys,
4475  "INDEX_MAX_KEYS", INDEX_MAX_KEYS),
4476  errhint("It looks like you need to recompile or initdb.")));
4478  ereport(FATAL,
4479  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4480  errmsg("database files are incompatible with server"),
4481  /* translator: %s is a variable name and %d is its value */
4482  errdetail("The database cluster was initialized with %s %d,"
4483  " but the server was compiled with %s %d.",
4484  "TOAST_MAX_CHUNK_SIZE", ControlFile->toast_max_chunk_size,
4485  "TOAST_MAX_CHUNK_SIZE", (int) TOAST_MAX_CHUNK_SIZE),
4486  errhint("It looks like you need to recompile or initdb.")));
4488  ereport(FATAL,
4489  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4490  errmsg("database files are incompatible with server"),
4491  /* translator: %s is a variable name and %d is its value */
4492  errdetail("The database cluster was initialized with %s %d,"
4493  " but the server was compiled with %s %d.",
4494  "LOBLKSIZE", ControlFile->loblksize,
4495  "LOBLKSIZE", (int) LOBLKSIZE),
4496  errhint("It looks like you need to recompile or initdb.")));
4497 
4498 #ifdef USE_FLOAT8_BYVAL
4499  if (ControlFile->float8ByVal != true)
4500  ereport(FATAL,
4501  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4502  errmsg("database files are incompatible with server"),
4503  errdetail("The database cluster was initialized without USE_FLOAT8_BYVAL"
4504  " but the server was compiled with USE_FLOAT8_BYVAL."),
4505  errhint("It looks like you need to recompile or initdb.")));
4506 #else
4507  if (ControlFile->float8ByVal != false)
4508  ereport(FATAL,
4509  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
4510  errmsg("database files are incompatible with server"),
4511  errdetail("The database cluster was initialized with USE_FLOAT8_BYVAL"
4512  " but the server was compiled without USE_FLOAT8_BYVAL."),
4513  errhint("It looks like you need to recompile or initdb.")));
4514 #endif
4515 
4517 
4519  ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4520  errmsg_plural("invalid WAL segment size in control file (%d byte)",
4521  "invalid WAL segment size in control file (%d bytes)",
4524  errdetail("The WAL segment size must be a power of two between 1 MB and 1 GB.")));
4525 
4526  snprintf(wal_segsz_str, sizeof(wal_segsz_str), "%d", wal_segment_size);
4527  SetConfigOption("wal_segment_size", wal_segsz_str, PGC_INTERNAL,
4529 
4530  /* check and update variables dependent on wal_segment_size */
4532  ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4533  /* translator: both %s are GUC names */
4534  errmsg("\"%s\" must be at least twice \"%s\"",
4535  "min_wal_size", "wal_segment_size")));
4536 
4538  ereport(ERROR, (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
4539  /* translator: both %s are GUC names */
4540  errmsg("\"%s\" must be at least twice \"%s\"",
4541  "max_wal_size", "wal_segment_size")));
4542 
4544  (wal_segment_size / XLOG_BLCKSZ * UsableBytesInPage) -
4546 
4548 
4549  /* Make the initdb settings visible as GUC variables, too */
4550  SetConfigOption("data_checksums", DataChecksumsEnabled() ? "yes" : "no",
4552 }
#define PG_BINARY
Definition: c.h:1276
#define CATALOG_VERSION_NO
Definition: catversion.h:60
int errmsg_plural(const char *fmt_singular, const char *fmt_plural, unsigned long n,...)
Definition: elog.c:1180
int BasicOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1086
void SetConfigOption(const char *name, const char *value, GucContext context, GucSource source)
Definition: guc.c:4332
@ PGC_S_DYNAMIC_DEFAULT
Definition: guc.h:110
@ PGC_INTERNAL
Definition: guc.h:69
#define TOAST_MAX_CHUNK_SIZE
Definition: heaptoast.h:84
#define read(a, b, c)
Definition: win32.h:13
#define LOBLKSIZE
Definition: large_object.h:70
#define INDEX_MAX_KEYS
#define NAMEDATALEN
#define FLOATFORMAT_VALUE
Definition: pg_control.h:201
#define PG_CONTROL_VERSION
Definition: pg_control.h:25
#define EQ_CRC32C(c1, c2)
Definition: pg_crc32c.h:42
uint32 pg_control_version
Definition: pg_control.h:125
uint32 xlog_seg_size
Definition: pg_control.h:211
uint32 nameDataLen
Definition: pg_control.h:213
uint32 indexMaxKeys
Definition: pg_control.h:214
uint32 relseg_size
Definition: pg_control.h:208
uint32 catalog_version_no
Definition: pg_control.h:126
double floatFormat
Definition: pg_control.h:200
uint32 xlog_blcksz
Definition: pg_control.h:210
uint32 loblksize
Definition: pg_control.h:217
pg_crc32c crc
Definition: pg_control.h:232
uint32 toast_max_chunk_size
Definition: pg_control.h:216
#define UsableBytesInPage
Definition: xlog.c:597
bool DataChecksumsEnabled(void)
Definition: xlog.c:4588
static int UsableBytesInSegment
Definition: xlog.c:606
int min_wal_size_mb
Definition: xlog.c:115
#define XLOG_CONTROL_FILE

References BasicOpenFile(), ControlFileData::blcksz, CalculateCheckpointSegments(), CATALOG_VERSION_NO, ControlFileData::catalog_version_no, close, COMP_CRC32C, ControlFile, ConvertToXSegs, ControlFileData::crc, crc, DataChecksumsEnabled(), EQ_CRC32C, ereport, errcode(), ERRCODE_DATA_CORRUPTED, errcode_for_file_access(), errdetail(), errhint(), errmsg(), errmsg_plural(), ERROR, FATAL, fd(), FIN_CRC32C, ControlFileData::float8ByVal, ControlFileData::floatFormat, FLOATFORMAT_VALUE, INDEX_MAX_KEYS, ControlFileData::indexMaxKeys, INIT_CRC32C, IsValidWalSegSize, ControlFileData::loblksize, LOBLKSIZE, max_wal_size_mb, ControlFileData::maxAlign, min_wal_size_mb, ControlFileData::nameDataLen, NAMEDATALEN, PANIC, PG_BINARY, PG_CONTROL_VERSION, ControlFileData::pg_control_version, PGC_INTERNAL, PGC_S_DYNAMIC_DEFAULT, pgstat_report_wait_end(), pgstat_report_wait_start(), read, ControlFileData::relseg_size, SetConfigOption(), SizeOfXLogLongPHD, SizeOfXLogShortPHD, snprintf, TOAST_MAX_CHUNK_SIZE, ControlFileData::toast_max_chunk_size, UsableBytesInPage, UsableBytesInSegment, wal_segment_size, ControlFileData::xlog_blcksz, XLOG_CONTROL_FILE, and ControlFileData::xlog_seg_size.

Referenced by BootStrapXLOG(), and LocalProcessControlFile().

◆ RecoveryInProgress()

bool RecoveryInProgress ( void  )

Definition at line 6334 of file xlog.c.

6335 {
6336  /*
6337  * We check shared state each time only until we leave recovery mode. We
6338  * can't re-enter recovery, so there's no need to keep checking after the
6339  * shared variable has once been seen false.
6340  */
6342  return false;
6343  else
6344  {
6345  /*
6346  * use volatile pointer to make sure we make a fresh read of the
6347  * shared variable.
6348  */
6349  volatile XLogCtlData *xlogctl = XLogCtl;
6350 
6352 
6353  /*
6354  * Note: We don't need a memory barrier when we're still in recovery.
6355  * We might exit recovery immediately after return, so the caller
6356  * can't rely on 'true' meaning that we're still in recovery anyway.
6357  */
6358 
6359  return LocalRecoveryInProgress;
6360  }
6361 }
static bool LocalRecoveryInProgress
Definition: xlog.c:224

References LocalRecoveryInProgress, RECOVERY_STATE_DONE, XLogCtlData::SharedRecoveryState, and XLogCtl.

Referenced by attribute_statistics_update(), BackgroundWriterMain(), BeginReportingGUCOptions(), brin_desummarize_range(), brin_summarize_range(), btree_index_mainfork_expected(), check_transaction_isolation(), check_transaction_read_only(), CheckArchiveTimeout(), CheckLogicalDecodingRequirements(), CheckpointerMain(), ComputeXidHorizons(), CreateCheckPoint(), CreateDecodingContext(), CreateEndOfRecoveryRecord(), CreateOverwriteContrecordRecord(), CreateRestartPoint(), do_pg_backup_start(), do_pg_backup_stop(), error_commit_ts_disabled(), get_relation_info(), GetCurrentLSN(), GetLatestLSN(), GetNewMultiXactId(), GetNewObjectId(), GetNewTransactionId(), GetOldestActiveTransactionId(), GetOldestSafeDecodingTransactionId(), GetRunningTransactionData(), GetSerializableTransactionSnapshot(), GetSerializableTransactionSnapshotInt(), GetSnapshotData(), GetStrictOldestNonRemovableTransactionId(), gin_clean_pending_list(), GlobalVisHorizonKindForRel(), heap_force_common(), heap_page_prune_opt(), IdentifySystem(), InitTempTableNamespace(), IsCheckpointOnSchedule(), LockAcquireExtended(), logical_read_xlog_page(), MaintainLatestCompletedXid(), MarkBufferDirtyHint(), perform_base_backup(), pg_clear_attribute_stats(), pg_create_restore_point(), pg_current_wal_flush_lsn(), pg_current_wal_insert_lsn(), pg_current_wal_lsn(), pg_get_sequence_data(), pg_get_wal_replay_pause_state(), pg_is_in_recovery(), pg_is_wal_replay_paused(), pg_log_standby_snapshot(), pg_logical_slot_get_changes_guts(), pg_promote(), pg_replication_slot_advance(), pg_sequence_last_value(), pg_switch_wal(), pg_sync_replication_slots(), pg_wal_replay_pause(), pg_wal_replay_resume(), pg_walfile_name(), pg_walfile_name_offset(), PhysicalWakeupLogicalWalSnd(), PrepareRedoAdd(), PrepareRedoRemove(), PreventCommandDuringRecovery(), ProcSleep(), read_local_xlog_page_guts(), ReadReplicationSlot(), recovery_create_dbdir(), relation_statistics_update(), ReplicationSlotAlter(), ReplicationSlotCreate(), ReplicationSlotDrop(), ReplicationSlotReserveWal(), replorigin_check_prerequisites(), ReportChangedGUCOptions(), sendDir(), SerialSetActiveSerXmin(), show_in_hot_standby(), ShutdownXLOG(), SnapBuildWaitSnapshot(), standard_ProcessUtility(), StandbySlotsHaveCaughtup(), StartLogicalReplication(), StartReplication(), StartTransaction(), TransactionIdIsInProgress(), TruncateMultiXact(), UpdateFullPageWrites(), verify_heapam(), WALReadFromBuffers(), WalReceiverMain(), WalSndWaitForWal(), xlog_decode(), XLogBackgroundFlush(), XLogFlush(), XLogInsertAllowed(), XLogNeedsFlush(), and XLogSendPhysical().

◆ RecoveryRestartPoint()

static void RecoveryRestartPoint ( const CheckPoint checkPoint,
XLogReaderState record 
)
static

Definition at line 7550 of file xlog.c.

7551 {
7552  /*
7553  * Also refrain from creating a restartpoint if we have seen any
7554  * references to non-existent pages. Restarting recovery from the
7555  * restartpoint would not see the references, so we would lose the
7556  * cross-check that the pages belonged to a relation that was dropped
7557  * later.
7558  */
7559  if (XLogHaveInvalidPages())
7560  {
7561  elog(DEBUG2,
7562  "could not record restart point at %X/%X because there "
7563  "are unresolved references to invalid pages",
7564  LSN_FORMAT_ARGS(checkPoint->redo));
7565  return;
7566  }
7567 
7568  /*
7569  * Copy the checkpoint record to shared memory, so that checkpointer can
7570  * work out the next time it wants to perform a restartpoint.
7571  */
7575  XLogCtl->lastCheckPoint = *checkPoint;
7577 }
XLogRecPtr EndRecPtr
Definition: xlogreader.h:207
XLogRecPtr ReadRecPtr
Definition: xlogreader.h:206
bool XLogHaveInvalidPages(void)
Definition: xlogutils.c:235

References DEBUG2, elog, XLogReaderState::EndRecPtr, XLogCtlData::info_lck, XLogCtlData::lastCheckPoint, XLogCtlData::lastCheckPointEndPtr, XLogCtlData::lastCheckPointRecPtr, LSN_FORMAT_ARGS, XLogReaderState::ReadRecPtr, CheckPoint::redo, SpinLockAcquire, SpinLockRelease, XLogCtl, and XLogHaveInvalidPages().

Referenced by xlog_redo().

◆ register_persistent_abort_backup_handler()

void register_persistent_abort_backup_handler ( void  )

Definition at line 9421 of file xlog.c.

9422 {
9423  static bool already_done = false;
9424 
9425  if (already_done)
9426  return;
9428  already_done = true;
9429 }
void before_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:337

References before_shmem_exit(), DatumGetBool(), and do_pg_abort_backup().

Referenced by pg_backup_start().

◆ RemoveNonParentXlogFiles()

void RemoveNonParentXlogFiles ( XLogRecPtr  switchpoint,
TimeLineID  newTLI 
)

Definition at line 3940 of file xlog.c.

3941 {
3942  DIR *xldir;
3943  struct dirent *xlde;
3944  char switchseg[MAXFNAMELEN];
3945  XLogSegNo endLogSegNo;
3946  XLogSegNo switchLogSegNo;
3947  XLogSegNo recycleSegNo;
3948 
3949  /*
3950  * Initialize info about where to begin the work. This will recycle,
3951  * somewhat arbitrarily, 10 future segments.
3952  */
3953  XLByteToPrevSeg(switchpoint, switchLogSegNo, wal_segment_size);
3954  XLByteToSeg(switchpoint, endLogSegNo, wal_segment_size);
3955  recycleSegNo = endLogSegNo + 10;
3956 
3957  /*
3958  * Construct a filename of the last segment to be kept.
3959  */
3960  XLogFileName(switchseg, newTLI, switchLogSegNo, wal_segment_size);
3961 
3962  elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3963  switchseg);
3964 
3965  xldir = AllocateDir(XLOGDIR);
3966 
3967  while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3968  {
3969  /* Ignore files that are not XLOG segments */
3970  if (!IsXLogFileName(xlde->d_name))
3971  continue;
3972 
3973  /*
3974  * Remove files that are on a timeline older than the new one we're
3975  * switching to, but with a segment number >= the first segment on the
3976  * new timeline.
3977  */
3978  if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
3979  strcmp(xlde->d_name + 8, switchseg + 8) > 0)
3980  {
3981  /*
3982  * If the file has already been marked as .ready, however, don't
3983  * remove it yet. It should be OK to remove it - files that are
3984  * not part of our timeline history are not required for recovery
3985  * - but seems safer to let them be archived and removed later.
3986  */
3987  if (!XLogArchiveIsReady(xlde->d_name))
3988  RemoveXlogFile(xlde, recycleSegNo, &endLogSegNo, newTLI);
3989  }
3990  }
3991 
3992  FreeDir(xldir);
3993 }
static void RemoveXlogFile(const struct dirent *segment_de, XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo, TimeLineID insertTLI)
Definition: xlog.c:4009
static bool IsXLogFileName(const char *fname)
bool XLogArchiveIsReady(const char *xlog)
Definition: xlogarchive.c:694

References AllocateDir(), dirent::d_name, DEBUG2, elog, FreeDir(), IsXLogFileName(), MAXFNAMELEN, ReadDir(), RemoveXlogFile(), wal_segment_size, XLByteToPrevSeg, XLByteToSeg, XLogArchiveIsReady(), XLOGDIR, and XLogFileName().

Referenced by ApplyWalRecord(), and CleanupAfterArchiveRecovery().

◆ RemoveOldXlogFiles()

static void RemoveOldXlogFiles ( XLogSegNo  segno,
XLogRecPtr  lastredoptr,
XLogRecPtr  endptr,
TimeLineID  insertTLI 
)
static

Definition at line 3865 of file xlog.c.

3867 {
3868  DIR *xldir;
3869  struct dirent *xlde;
3870  char lastoff[MAXFNAMELEN];
3871  XLogSegNo endlogSegNo;
3872  XLogSegNo recycleSegNo;
3873 
3874  /* Initialize info about where to try to recycle to */
3875  XLByteToSeg(endptr, endlogSegNo, wal_segment_size);
3876  recycleSegNo = XLOGfileslop(lastredoptr);
3877 
3878  /*
3879  * Construct a filename of the last segment to be kept. The timeline ID
3880  * doesn't matter, we ignore that in the comparison. (During recovery,
3881  * InsertTimeLineID isn't set, so we can't use that.)
3882  */
3883  XLogFileName(lastoff, 0, segno, wal_segment_size);
3884 
3885  elog(DEBUG2, "attempting to remove WAL segments older than log file %s",
3886  lastoff);
3887 
3888  xldir = AllocateDir(XLOGDIR);
3889 
3890  while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3891  {
3892  /* Ignore files that are not XLOG segments */
3893  if (!IsXLogFileName(xlde->d_name) &&
3894  !IsPartialXLogFileName(xlde->d_name))
3895  continue;
3896 
3897  /*
3898  * We ignore the timeline part of the XLOG segment identifiers in
3899  * deciding whether a segment is still needed. This ensures that we
3900  * won't prematurely remove a segment from a parent timeline. We could
3901  * probably be a little more proactive about removing segments of
3902  * non-parent timelines, but that would be a whole lot more
3903  * complicated.
3904  *
3905  * We use the alphanumeric sorting property of the filenames to decide
3906  * which ones are earlier than the lastoff segment.
3907  */
3908  if (strcmp(xlde->d_name + 8, lastoff + 8) <= 0)
3909  {
3910  if (XLogArchiveCheckDone(xlde->d_name))
3911  {
3912  /* Update the last removed location in shared memory first */
3914 
3915  RemoveXlogFile(xlde, recycleSegNo, &endlogSegNo, insertTLI);
3916  }
3917  }
3918  }
3919 
3920  FreeDir(xldir);
3921 }
static XLogSegNo XLOGfileslop(XLogRecPtr lastredoptr)
Definition: xlog.c:2240
static void UpdateLastRemovedPtr(char *filename)
Definition: xlog.c:3812
static bool IsPartialXLogFileName(const char *fname)

References AllocateDir(), dirent::d_name, DEBUG2, elog, FreeDir(), IsPartialXLogFileName(), IsXLogFileName(), MAXFNAMELEN, ReadDir(), RemoveXlogFile(), UpdateLastRemovedPtr(), wal_segment_size, XLByteToSeg, XLogArchiveCheckDone(), XLOGDIR, XLogFileName(), and XLOGfileslop().

Referenced by CreateCheckPoint(), and CreateRestartPoint().

◆ RemoveTempXlogFiles()

static void RemoveTempXlogFiles ( void  )
static

Definition at line 3832 of file xlog.c.

3833 {
3834  DIR *xldir;
3835  struct dirent *xlde;
3836 
3837  elog(DEBUG2, "removing all temporary WAL segments");
3838 
3839  xldir = AllocateDir(XLOGDIR);
3840  while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3841  {
3842  char path[MAXPGPATH];
3843 
3844  if (strncmp(xlde->d_name, "xlogtemp.", 9) != 0)
3845  continue;
3846 
3847  snprintf(path, MAXPGPATH, XLOGDIR "/%s", xlde->d_name);
3848  unlink(path);
3849  elog(DEBUG2, "removed temporary WAL segment \"%s\"", path);
3850  }
3851  FreeDir(xldir);
3852 }

References AllocateDir(), dirent::d_name, DEBUG2, elog, FreeDir(), MAXPGPATH, ReadDir(), snprintf, and XLOGDIR.

Referenced by StartupXLOG().

◆ RemoveXlogFile()

static void RemoveXlogFile ( const struct dirent segment_de,
XLogSegNo  recycleSegNo,
XLogSegNo endlogSegNo,
TimeLineID  insertTLI 
)
static

Definition at line 4009 of file xlog.c.

4012 {
4013  char path[MAXPGPATH];
4014 #ifdef WIN32
4015  char newpath[MAXPGPATH];
4016 #endif
4017  const char *segname = segment_de->d_name;
4018 
4019  snprintf(path, MAXPGPATH, XLOGDIR "/%s", segname);
4020 
4021  /*
4022  * Before deleting the file, see if it can be recycled as a future log
4023  * segment. Only recycle normal files, because we don't want to recycle
4024  * symbolic links pointing to a separate archive directory.
4025  */
4026  if (wal_recycle &&
4027  *endlogSegNo <= recycleSegNo &&
4028  XLogCtl->InstallXLogFileSegmentActive && /* callee rechecks this */
4029  get_dirent_type(path, segment_de, false, DEBUG2) == PGFILETYPE_REG &&
4030  InstallXLogFileSegment(endlogSegNo, path,
4031  true, recycleSegNo, insertTLI))
4032  {
4033  ereport(DEBUG2,
4034  (errmsg_internal("recycled write-ahead log file \"%s\"",
4035  segname)));
4037  /* Needn't recheck that slot on future iterations */
4038  (*endlogSegNo)++;
4039  }
4040  else
4041  {
4042  /* No need for any more future segments, or recycling failed ... */
4043  int rc;
4044 
4045  ereport(DEBUG2,
4046  (errmsg_internal("removing write-ahead log file \"%s\"",
4047  segname)));
4048 
4049 #ifdef WIN32
4050 
4051  /*
4052  * On Windows, if another process (e.g another backend) holds the file
4053  * open in FILE_SHARE_DELETE mode, unlink will succeed, but the file
4054  * will still show up in directory listing until the last handle is
4055  * closed. To avoid confusing the lingering deleted file for a live
4056  * WAL file that needs to be archived, rename it before deleting it.
4057  *
4058  * If another process holds the file open without FILE_SHARE_DELETE
4059  * flag, rename will fail. We'll try again at the next checkpoint.
4060  */
4061  snprintf(newpath, MAXPGPATH, "%s.deleted", path);
4062  if (rename(path, newpath) != 0)
4063  {
4064  ereport(LOG,
4066  errmsg("could not rename file \"%s\": %m",
4067  path)));
4068  return;
4069  }
4070  rc = durable_unlink(newpath, LOG);
4071 #else
4072  rc = durable_unlink(path, LOG);
4073 #endif
4074  if (rc != 0)
4075  {
4076  /* Message already logged by durable_unlink() */
4077  return;
4078  }
4080  }
4081 
4082  XLogArchiveCleanup(segname);
4083 }
@ PGFILETYPE_REG
Definition: file_utils.h:22
static bool InstallXLogFileSegment(XLogSegNo *segno, char *tmppath, bool find_free, XLogSegNo max_segno, TimeLineID tli)
Definition: xlog.c:3563
bool wal_recycle
Definition: xlog.c:128

References CheckpointStats, CheckpointStatsData::ckpt_segs_recycled, CheckpointStatsData::ckpt_segs_removed, dirent::d_name, DEBUG2, durable_unlink(), ereport, errcode_for_file_access(), errmsg(), errmsg_internal(), get_dirent_type(), InstallXLogFileSegment(), XLogCtlData::InstallXLogFileSegmentActive, LOG, MAXPGPATH, PGFILETYPE_REG, snprintf, wal_recycle, XLogArchiveCleanup(), XLogCtl, and XLOGDIR.

Referenced by RemoveNonParentXlogFiles(), and RemoveOldXlogFiles().

◆ RequestXLogSwitch()

XLogRecPtr RequestXLogSwitch ( bool  mark_unimportant)

Definition at line 8060 of file xlog.c.

8061 {
8062  XLogRecPtr RecPtr;
8063 
8064  /* XLOG SWITCH has no data */
8065  XLogBeginInsert();
8066 
8067  if (mark_unimportant)
8069  RecPtr = XLogInsert(RM_XLOG_ID, XLOG_SWITCH);
8070 
8071  return RecPtr;
8072 }
#define XLOG_SWITCH
Definition: pg_control.h:72
#define XLOG_MARK_UNIMPORTANT
Definition: xlog.h:155
void XLogSetRecordFlags(uint8 flags)
Definition: xloginsert.c:456

References XLOG_MARK_UNIMPORTANT, XLOG_SWITCH, XLogBeginInsert(), XLogInsert(), and XLogSetRecordFlags().

Referenced by CheckArchiveTimeout(), do_pg_backup_start(), do_pg_backup_stop(), pg_switch_wal(), and ShutdownXLOG().

◆ ReserveXLogInsertLocation()

static pg_attribute_always_inline void ReserveXLogInsertLocation ( int  size,
XLogRecPtr StartPos,
XLogRecPtr EndPos,
XLogRecPtr PrevPtr 
)
static

Definition at line 1108 of file xlog.c.

1110 {
1112  uint64 startbytepos;
1113  uint64 endbytepos;
1114  uint64 prevbytepos;
1115 
1116  size = MAXALIGN(size);
1117 
1118  /* All (non xlog-switch) records should contain data. */
1120 
1121  /*
1122  * The duration the spinlock needs to be held is minimized by minimizing
1123  * the calculations that have to be done while holding the lock. The
1124  * current tip of reserved WAL is kept in CurrBytePos, as a byte position
1125  * that only counts "usable" bytes in WAL, that is, it excludes all WAL
1126  * page headers. The mapping between "usable" byte positions and physical
1127  * positions (XLogRecPtrs) can be done outside the locked region, and
1128  * because the usable byte position doesn't include any headers, reserving
1129  * X bytes from WAL is almost as simple as "CurrBytePos += X".
1130  */
1131  SpinLockAcquire(&Insert->insertpos_lck);
1132 
1133  startbytepos = Insert->CurrBytePos;
1134  endbytepos = startbytepos + size;
1135  prevbytepos = Insert->PrevBytePos;
1136  Insert->CurrBytePos = endbytepos;
1137  Insert->PrevBytePos = startbytepos;
1138 
1139  SpinLockRelease(&Insert->insertpos_lck);
1140 
1141  *StartPos = XLogBytePosToRecPtr(startbytepos);
1142  *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1143  *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1144 
1145  /*
1146  * Check that the conversions between "usable byte positions" and
1147  * XLogRecPtrs work consistently in both directions.
1148  */
1149  Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1150  Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1151  Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1152 }
#define MAXALIGN(LEN)
Definition: c.h:814
static pg_noinline void Size size
Definition: slab.c:607
static XLogRecPtr XLogBytePosToEndRecPtr(uint64 bytepos)
Definition: xlog.c:1898
static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr)
Definition: xlog.c:1941

References Assert, XLogCtlData::Insert, Insert(), MAXALIGN, size, SizeOfXLogRecord, SpinLockAcquire, SpinLockRelease, XLogBytePosToEndRecPtr(), XLogBytePosToRecPtr(), XLogCtl, and XLogRecPtrToBytePos().

Referenced by XLogInsertRecord().

◆ ReserveXLogSwitch()

static bool ReserveXLogSwitch ( XLogRecPtr StartPos,
XLogRecPtr EndPos,
XLogRecPtr PrevPtr 
)
static

Definition at line 1164 of file xlog.c.

1165 {
1167  uint64 startbytepos;
1168  uint64 endbytepos;
1169  uint64 prevbytepos;
1171  XLogRecPtr ptr;
1172  uint32 segleft;
1173 
1174  /*
1175  * These calculations are a bit heavy-weight to be done while holding a
1176  * spinlock, but since we're holding all the WAL insertion locks, there
1177  * are no other inserters competing for it. GetXLogInsertRecPtr() does
1178  * compete for it, but that's not called very frequently.
1179  */
1180  SpinLockAcquire(&Insert->insertpos_lck);
1181 
1182  startbytepos = Insert->CurrBytePos;
1183 
1184  ptr = XLogBytePosToEndRecPtr(startbytepos);
1185  if (XLogSegmentOffset(ptr, wal_segment_size) == 0)
1186  {
1187  SpinLockRelease(&Insert->insertpos_lck);
1188  *EndPos = *StartPos = ptr;
1189  return false;
1190  }
1191 
1192  endbytepos = startbytepos + size;
1193  prevbytepos = Insert->PrevBytePos;
1194 
1195  *StartPos = XLogBytePosToRecPtr(startbytepos);
1196  *EndPos = XLogBytePosToEndRecPtr(endbytepos);
1197 
1198  segleft = wal_segment_size - XLogSegmentOffset(*EndPos, wal_segment_size);
1199  if (segleft != wal_segment_size)
1200  {
1201  /* consume the rest of the segment */
1202  *EndPos += segleft;
1203  endbytepos = XLogRecPtrToBytePos(*EndPos);
1204  }
1205  Insert->CurrBytePos = endbytepos;
1206  Insert->PrevBytePos = startbytepos;
1207 
1208  SpinLockRelease(&Insert->insertpos_lck);
1209 
1210  *PrevPtr = XLogBytePosToRecPtr(prevbytepos);
1211 
1212  Assert(XLogSegmentOffset(*EndPos, wal_segment_size) == 0);
1213  Assert(XLogRecPtrToBytePos(*EndPos) == endbytepos);
1214  Assert(XLogRecPtrToBytePos(*StartPos) == startbytepos);
1215  Assert(XLogRecPtrToBytePos(*PrevPtr) == prevbytepos);
1216 
1217  return true;
1218 }

References Assert, XLogCtlData::Insert, Insert(), MAXALIGN, size, SizeOfXLogRecord, SpinLockAcquire, SpinLockRelease, wal_segment_size, XLogBytePosToEndRecPtr(), XLogBytePosToRecPtr(), XLogCtl, XLogRecPtrToBytePos(), and XLogSegmentOffset.

Referenced by XLogInsertRecord().

◆ SetInstallXLogFileSegmentActive()

void SetInstallXLogFileSegmentActive ( void  )

Definition at line 9484 of file xlog.c.

9485 {
9486  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9488  LWLockRelease(ControlFileLock);
9489 }

References XLogCtlData::InstallXLogFileSegmentActive, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), and XLogCtl.

Referenced by BootStrapXLOG(), StartupXLOG(), and WaitForWALToBecomeAvailable().

◆ SetWalWriterSleeping()

void SetWalWriterSleeping ( bool  sleeping)

Definition at line 9507 of file xlog.c.

9508 {
9510  XLogCtl->WalWriterSleeping = sleeping;
9512 }
bool WalWriterSleeping
Definition: xlog.c:533

References XLogCtlData::info_lck, SpinLockAcquire, SpinLockRelease, XLogCtlData::WalWriterSleeping, and XLogCtl.

Referenced by WalWriterMain().

◆ show_archive_command()

const char* show_archive_command ( void  )

Definition at line 4810 of file xlog.c.

4811 {
4812  if (XLogArchivingActive())
4813  return XLogArchiveCommand;
4814  else
4815  return "(disabled)";
4816 }
char * XLogArchiveCommand
Definition: xlog.c:120

References XLogArchiveCommand, and XLogArchivingActive.

◆ show_in_hot_standby()

const char* show_in_hot_standby ( void  )

Definition at line 4822 of file xlog.c.

4823 {
4824  /*
4825  * We display the actual state based on shared memory, so that this GUC
4826  * reports up-to-date state if examined intra-query. The underlying
4827  * variable (in_hot_standby_guc) changes only when we transmit a new value
4828  * to the client.
4829  */
4830  return RecoveryInProgress() ? "on" : "off";
4831 }

References RecoveryInProgress().

◆ ShutdownXLOG()

void ShutdownXLOG ( int  code,
Datum  arg 
)

Definition at line 6602 of file xlog.c.

6603 {
6604  /*
6605  * We should have an aux process resource owner to use, and we should not
6606  * be in a transaction that's installed some other resowner.
6607  */
6609  Assert(CurrentResourceOwner == NULL ||
6612 
6613  /* Don't be chatty in standalone mode */
6615  (errmsg("shutting down")));
6616 
6617  /*
6618  * Signal walsenders to move to stopping state.
6619  */
6621 
6622  /*
6623  * Wait for WAL senders to be in stopping state. This prevents commands
6624  * from writing new WAL.
6625  */
6627 
6628  if (RecoveryInProgress())
6630  else
6631  {
6632  /*
6633  * If archiving is enabled, rotate the last XLOG file so that all the
6634  * remaining records are archived (postmaster wakes up the archiver
6635  * process one more time at the end of shutdown). The checkpoint
6636  * record will go to the next XLOG file and won't be archived (yet).
6637  */
6638  if (XLogArchivingActive())
6639  RequestXLogSwitch(false);
6640 
6642  }
6643 }
bool IsPostmasterEnvironment
Definition: globals.c:118
ResourceOwner CurrentResourceOwner
Definition: resowner.c:165
ResourceOwner AuxProcessResourceOwner
Definition: resowner.c:168
void WalSndInitStopping(void)
Definition: walsender.c:3717
void WalSndWaitStopping(void)
Definition: walsender.c:3743
bool CreateRestartPoint(int flags)
Definition: xlog.c:7591
bool CreateCheckPoint(int flags)
Definition: xlog.c:6889

References Assert, AuxProcessResourceOwner, CHECKPOINT_IMMEDIATE, CHECKPOINT_IS_SHUTDOWN, CreateCheckPoint(), CreateRestartPoint(), CurrentResourceOwner, ereport, errmsg(), IsPostmasterEnvironment, LOG, NOTICE, RecoveryInProgress(), RequestXLogSwitch(), WalSndInitStopping(), WalSndWaitStopping(), and XLogArchivingActive.

Referenced by HandleCheckpointerInterrupts(), and InitPostgres().

◆ StartupXLOG()

void StartupXLOG ( void  )

Definition at line 5429 of file xlog.c.

5430 {
5432  CheckPoint checkPoint;
5433  bool wasShutdown;
5434  bool didCrash;
5435  bool haveTblspcMap;
5436  bool haveBackupLabel;
5437  XLogRecPtr EndOfLog;
5438  TimeLineID EndOfLogTLI;
5439  TimeLineID newTLI;
5440  bool performedWalRecovery;
5441  EndOfWalRecoveryInfo *endOfRecoveryInfo;
5444  TransactionId oldestActiveXID;
5445  bool promoted = false;
5446 
5447  /*
5448  * We should have an aux process resource owner to use, and we should not
5449  * be in a transaction that's installed some other resowner.
5450  */
5452  Assert(CurrentResourceOwner == NULL ||
5455 
5456  /*
5457  * Check that contents look valid.
5458  */
5460  ereport(FATAL,
5462  errmsg("control file contains invalid checkpoint location")));
5463 
5464  switch (ControlFile->state)
5465  {
5466  case DB_SHUTDOWNED:
5467 
5468  /*
5469  * This is the expected case, so don't be chatty in standalone
5470  * mode
5471  */
5473  (errmsg("database system was shut down at %s",
5474  str_time(ControlFile->time))));
5475  break;
5476 
5478  ereport(LOG,
5479  (errmsg("database system was shut down in recovery at %s",
5480  str_time(ControlFile->time))));
5481  break;
5482 
5483  case DB_SHUTDOWNING:
5484  ereport(LOG,
5485  (errmsg("database system shutdown was interrupted; last known up at %s",
5486  str_time(ControlFile->time))));
5487  break;
5488 
5489  case DB_IN_CRASH_RECOVERY:
5490  ereport(LOG,
5491  (errmsg("database system was interrupted while in recovery at %s",
5493  errhint("This probably means that some data is corrupted and"
5494  " you will have to use the last backup for recovery.")));
5495  break;
5496 
5498  ereport(LOG,
5499  (errmsg("database system was interrupted while in recovery at log time %s",
5501  errhint("If this has occurred more than once some data might be corrupted"
5502  " and you might need to choose an earlier recovery target.")));
5503  break;
5504 
5505  case DB_IN_PRODUCTION:
5506  ereport(LOG,
5507  (errmsg("database system was interrupted; last known up at %s",
5508  str_time(ControlFile->time))));
5509  break;
5510 
5511  default:
5512  ereport(FATAL,
5514  errmsg("control file contains invalid database cluster state")));
5515  }
5516 
5517  /* This is just to allow attaching to startup process with a debugger */
5518 #ifdef XLOG_REPLAY_DELAY
5520  pg_usleep(60000000L);
5521 #endif
5522 
5523  /*
5524  * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
5525  * In cases where someone has performed a copy for PITR, these directories
5526  * may have been excluded and need to be re-created.
5527  */
5529 
5530  /* Set up timeout handler needed to report startup progress. */
5534 
5535  /*----------
5536  * If we previously crashed, perform a couple of actions:
5537  *
5538  * - The pg_wal directory may still include some temporary WAL segments
5539  * used when creating a new segment, so perform some clean up to not
5540  * bloat this path. This is done first as there is no point to sync
5541  * this temporary data.
5542  *
5543  * - There might be data which we had written, intending to fsync it, but
5544  * which we had not actually fsync'd yet. Therefore, a power failure in
5545  * the near future might cause earlier unflushed writes to be lost, even
5546  * though more recent data written to disk from here on would be
5547  * persisted. To avoid that, fsync the entire data directory.
5548  */
5549  if (ControlFile->state != DB_SHUTDOWNED &&
5551  {
5554  didCrash = true;
5555  }
5556  else
5557  didCrash = false;
5558 
5559  /*
5560  * Prepare for WAL recovery if needed.
5561  *
5562  * InitWalRecovery analyzes the control file and the backup label file, if
5563  * any. It updates the in-memory ControlFile buffer according to the
5564  * starting checkpoint, and sets InRecovery and ArchiveRecoveryRequested.
5565  * It also applies the tablespace map file, if any.
5566  */
5567  InitWalRecovery(ControlFile, &wasShutdown,
5568  &haveBackupLabel, &haveTblspcMap);
5569  checkPoint = ControlFile->checkPointCopy;
5570 
5571  /* initialize shared memory variables from the checkpoint record */
5572  TransamVariables->nextXid = checkPoint.nextXid;
5573  TransamVariables->nextOid = checkPoint.nextOid;
5575  MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5576  AdvanceOldestClogXid(checkPoint.oldestXid);
5577  SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5578  SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5580  checkPoint.newestCommitTsXid);
5581  XLogCtl->ckptFullXid = checkPoint.nextXid;
5582 
5583  /*
5584  * Clear out any old relcache cache files. This is *necessary* if we do
5585  * any WAL replay, since that would probably result in the cache files
5586  * being out of sync with database reality. In theory we could leave them
5587  * in place if the database had been cleanly shut down, but it seems
5588  * safest to just remove them always and let them be rebuilt during the
5589  * first backend startup. These files needs to be removed from all
5590  * directories including pg_tblspc, however the symlinks are created only
5591  * after reading tablespace_map file in case of archive recovery from
5592  * backup, so needs to clear old relcache files here after creating
5593  * symlinks.
5594  */
5596 
5597  /*
5598  * Initialize replication slots, before there's a chance to remove
5599  * required resources.
5600  */
5602 
5603  /*
5604  * Startup logical state, needs to be setup now so we have proper data
5605  * during crash recovery.
5606  */
5608 
5609  /*
5610  * Startup CLOG. This must be done after TransamVariables->nextXid has
5611  * been initialized and before we accept connections or begin WAL replay.
5612  */
5613  StartupCLOG();
5614 
5615  /*
5616  * Startup MultiXact. We need to do this early to be able to replay
5617  * truncations.
5618  */
5619  StartupMultiXact();
5620 
5621  /*
5622  * Ditto for commit timestamps. Activate the facility if the setting is
5623  * enabled in the control file, as there should be no tracking of commit
5624  * timestamps done when the setting was disabled. This facility can be
5625  * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
5626  */
5628  StartupCommitTs();
5629 
5630  /*
5631  * Recover knowledge about replay progress of known replication partners.
5632  */
5634 
5635  /*
5636  * Initialize unlogged LSN. On a clean shutdown, it's restored from the
5637  * control file. On recovery, all unlogged relations are blown away, so
5638  * the unlogged LSN counter can be reset too.
5639  */
5643  else
5646 
5647  /*
5648  * Copy any missing timeline history files between 'now' and the recovery
5649  * target timeline from archive to pg_wal. While we don't need those files
5650  * ourselves - the history file of the recovery target timeline covers all
5651  * the previous timelines in the history too - a cascading standby server
5652  * might be interested in them. Or, if you archive the WAL from this
5653  * server to a different archive than the primary, it'd be good for all
5654  * the history files to get archived there after failover, so that you can
5655  * use one of the old timelines as a PITR target. Timeline history files
5656  * are small, so it's better to copy them unnecessarily than not copy them
5657  * and regret later.
5658  */
5660 
5661  /*
5662  * Before running in recovery, scan pg_twophase and fill in its status to
5663  * be able to work on entries generated by redo. Doing a scan before
5664  * taking any recovery action has the merit to discard any 2PC files that
5665  * are newer than the first record to replay, saving from any conflicts at
5666  * replay. This avoids as well any subsequent scans when doing recovery
5667  * of the on-disk two-phase data.
5668  */
5670 
5671  /*
5672  * When starting with crash recovery, reset pgstat data - it might not be
5673  * valid. Otherwise restore pgstat data. It's safe to do this here,
5674  * because postmaster will not yet have started any other processes.
5675  *
5676  * NB: Restoring replication slot stats relies on slot state to have
5677  * already been restored from disk.
5678  *
5679  * TODO: With a bit of extra work we could just start with a pgstat file
5680  * associated with the checkpoint redo location we're starting from.
5681  */
5682  if (didCrash)
5684  else
5685  pgstat_restore_stats(checkPoint.redo);
5686 
5687  lastFullPageWrites = checkPoint.fullPageWrites;
5688 
5691 
5692  /* REDO */
5693  if (InRecovery)
5694  {
5695  /* Initialize state for RecoveryInProgress() */
5697  if (InArchiveRecovery)
5699  else
5702 
5703  /*
5704  * Update pg_control to show that we are recovering and to show the
5705  * selected checkpoint as the place we are starting from. We also mark
5706  * pg_control with any minimum recovery stop point obtained from a
5707  * backup history file.
5708  *
5709  * No need to hold ControlFileLock yet, we aren't up far enough.
5710  */
5712 
5713  /*
5714  * If there was a backup label file, it's done its job and the info
5715  * has now been propagated into pg_control. We must get rid of the
5716  * label file so that if we crash during recovery, we'll pick up at
5717  * the latest recovery restartpoint instead of going all the way back
5718  * to the backup start point. It seems prudent though to just rename
5719  * the file out of the way rather than delete it completely.
5720  */
5721  if (haveBackupLabel)
5722  {
5723  unlink(BACKUP_LABEL_OLD);
5725  }
5726 
5727  /*
5728  * If there was a tablespace_map file, it's done its job and the
5729  * symlinks have been created. We must get rid of the map file so
5730  * that if we crash during recovery, we don't create symlinks again.
5731  * It seems prudent though to just rename the file out of the way
5732  * rather than delete it completely.
5733  */
5734  if (haveTblspcMap)
5735  {
5736  unlink(TABLESPACE_MAP_OLD);
5738  }
5739 
5740  /*
5741  * Initialize our local copy of minRecoveryPoint. When doing crash
5742  * recovery we want to replay up to the end of WAL. Particularly, in
5743  * the case of a promoted standby minRecoveryPoint value in the
5744  * control file is only updated after the first checkpoint. However,
5745  * if the instance crashes before the first post-recovery checkpoint
5746  * is completed then recovery will use a stale location causing the
5747  * startup process to think that there are still invalid page
5748  * references when checking for data consistency.
5749  */
5750  if (InArchiveRecovery)
5751  {
5754  }
5755  else
5756  {
5759  }
5760 
5761  /* Check that the GUCs used to generate the WAL allow recovery */
5763 
5764  /*
5765  * We're in recovery, so unlogged relations may be trashed and must be
5766  * reset. This should be done BEFORE allowing Hot Standby
5767  * connections, so that read-only backends don't try to read whatever
5768  * garbage is left over from before.
5769  */
5771 
5772  /*
5773  * Likewise, delete any saved transaction snapshot files that got left
5774  * behind by crashed backends.
5775  */
5777 
5778  /*
5779  * Initialize for Hot Standby, if enabled. We won't let backends in
5780  * yet, not until we've reached the min recovery point specified in
5781  * control file and we've established a recovery snapshot from a
5782  * running-xacts WAL record.
5783  */
5785  {
5786  TransactionId *xids;
5787  int nxids;
5788 
5789  ereport(DEBUG1,
5790  (errmsg_internal("initializing for hot standby")));
5791 
5793 
5794  if (wasShutdown)
5795  oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
5796  else
5797  oldestActiveXID = checkPoint.oldestActiveXid;
5798  Assert(TransactionIdIsValid(oldestActiveXID));
5799 
5800  /* Tell procarray about the range of xids it has to deal with */
5802 
5803  /*
5804  * Startup subtrans only. CLOG, MultiXact and commit timestamp
5805  * have already been started up and other SLRUs are not maintained
5806  * during recovery and need not be started yet.
5807  */
5808  StartupSUBTRANS(oldestActiveXID);
5809 
5810  /*
5811  * If we're beginning at a shutdown checkpoint, we know that
5812  * nothing was running on the primary at this point. So fake-up an
5813  * empty running-xacts record and use that here and now. Recover
5814  * additional standby state for prepared transactions.
5815  */
5816  if (wasShutdown)
5817  {
5818  RunningTransactionsData running;
5819  TransactionId latestCompletedXid;
5820 
5821  /* Update pg_subtrans entries for any prepared transactions */
5823 
5824  /*
5825  * Construct a RunningTransactions snapshot representing a
5826  * shut down server, with only prepared transactions still
5827  * alive. We're never overflowed at this point because all
5828  * subxids are listed with their parent prepared transactions.
5829  */
5830  running.xcnt = nxids;
5831  running.subxcnt = 0;
5833  running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5834  running.oldestRunningXid = oldestActiveXID;
5835  latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5836  TransactionIdRetreat(latestCompletedXid);
5837  Assert(TransactionIdIsNormal(latestCompletedXid));
5838  running.latestCompletedXid = latestCompletedXid;
5839  running.xids = xids;
5840 
5841  ProcArrayApplyRecoveryInfo(&running);
5842  }
5843  }
5844 
5845  /*
5846  * We're all set for replaying the WAL now. Do it.
5847  */
5849  performedWalRecovery = true;
5850  }
5851  else
5852  performedWalRecovery = false;
5853 
5854  /*
5855  * Finish WAL recovery.
5856  */
5857  endOfRecoveryInfo = FinishWalRecovery();
5858  EndOfLog = endOfRecoveryInfo->endOfLog;
5859  EndOfLogTLI = endOfRecoveryInfo->endOfLogTLI;
5860  abortedRecPtr = endOfRecoveryInfo->abortedRecPtr;
5861  missingContrecPtr = endOfRecoveryInfo->missingContrecPtr;
5862 
5863  /*
5864  * Reset ps status display, so as no information related to recovery shows
5865  * up.
5866  */
5867  set_ps_display("");
5868 
5869  /*
5870  * When recovering from a backup (we are in recovery, and archive recovery
5871  * was requested), complain if we did not roll forward far enough to reach
5872  * the point where the database is consistent. For regular online
5873  * backup-from-primary, that means reaching the end-of-backup WAL record
5874  * (at which point we reset backupStartPoint to be Invalid), for
5875  * backup-from-replica (which can't inject records into the WAL stream),
5876  * that point is when we reach the minRecoveryPoint in pg_control (which
5877  * we purposefully copy last when backing up from a replica). For
5878  * pg_rewind (which creates a backup_label with a method of "pg_rewind")
5879  * or snapshot-style backups (which don't), backupEndRequired will be set
5880  * to false.
5881  *
5882  * Note: it is indeed okay to look at the local variable
5883  * LocalMinRecoveryPoint here, even though ControlFile->minRecoveryPoint
5884  * might be further ahead --- ControlFile->minRecoveryPoint cannot have
5885  * been advanced beyond the WAL we processed.
5886  */
5887  if (InRecovery &&
5888  (EndOfLog < LocalMinRecoveryPoint ||
5890  {
5891  /*
5892  * Ran off end of WAL before reaching end-of-backup WAL record, or
5893  * minRecoveryPoint. That's a bad sign, indicating that you tried to
5894  * recover from an online backup but never called pg_backup_stop(), or
5895  * you didn't archive all the WAL needed.
5896  */
5898  {
5900  ereport(FATAL,
5901  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5902  errmsg("WAL ends before end of online backup"),
5903  errhint("All WAL generated while online backup was taken must be available at recovery.")));
5904  else
5905  ereport(FATAL,
5906  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
5907  errmsg("WAL ends before consistent recovery point")));
5908  }
5909  }
5910 
5911  /*
5912  * Reset unlogged relations to the contents of their INIT fork. This is
5913  * done AFTER recovery is complete so as to include any unlogged relations
5914  * created during recovery, but BEFORE recovery is marked as having
5915  * completed successfully. Otherwise we'd not retry if any of the post
5916  * end-of-recovery steps fail.
5917  */
5918  if (InRecovery)
5920 
5921  /*
5922  * Pre-scan prepared transactions to find out the range of XIDs present.
5923  * This information is not quite needed yet, but it is positioned here so
5924  * as potential problems are detected before any on-disk change is done.
5925  */
5926  oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
5927 
5928  /*
5929  * Allow ordinary WAL segment creation before possibly switching to a new
5930  * timeline, which creates a new segment, and after the last ReadRecord().
5931  */
5933 
5934  /*
5935  * Consider whether we need to assign a new timeline ID.
5936  *
5937  * If we did archive recovery, we always assign a new ID. This handles a
5938  * couple of issues. If we stopped short of the end of WAL during
5939  * recovery, then we are clearly generating a new timeline and must assign
5940  * it a unique new ID. Even if we ran to the end, modifying the current
5941  * last segment is problematic because it may result in trying to
5942  * overwrite an already-archived copy of that segment, and we encourage
5943  * DBAs to make their archive_commands reject that. We can dodge the
5944  * problem by making the new active segment have a new timeline ID.
5945  *
5946  * In a normal crash recovery, we can just extend the timeline we were in.
5947  */
5948  newTLI = endOfRecoveryInfo->lastRecTLI;
5950  {
5951  newTLI = findNewestTimeLine(recoveryTargetTLI) + 1;
5952  ereport(LOG,
5953  (errmsg("selected new timeline ID: %u", newTLI)));
5954 
5955  /*
5956  * Make a writable copy of the last WAL segment. (Note that we also
5957  * have a copy of the last block of the old WAL in
5958  * endOfRecovery->lastPage; we will use that below.)
5959  */
5960  XLogInitNewTimeline(EndOfLogTLI, EndOfLog, newTLI);
5961 
5962  /*
5963  * Remove the signal files out of the way, so that we don't
5964  * accidentally re-enter archive recovery mode in a subsequent crash.
5965  */
5966  if (endOfRecoveryInfo->standby_signal_file_found)
5968 
5969  if (endOfRecoveryInfo->recovery_signal_file_found)
5971 
5972  /*
5973  * Write the timeline history file, and have it archived. After this
5974  * point (or rather, as soon as the file is archived), the timeline
5975  * will appear as "taken" in the WAL archive and to any standby
5976  * servers. If we crash before actually switching to the new
5977  * timeline, standby servers will nevertheless think that we switched
5978  * to the new timeline, and will try to connect to the new timeline.
5979  * To minimize the window for that, try to do as little as possible
5980  * between here and writing the end-of-recovery record.
5981  */
5983  EndOfLog, endOfRecoveryInfo->recoveryStopReason);
5984 
5985  ereport(LOG,
5986  (errmsg("archive recovery complete")));
5987  }
5988 
5989  /* Save the selected TimeLineID in shared memory, too */
5991  XLogCtl->InsertTimeLineID = newTLI;
5992  XLogCtl->PrevTimeLineID = endOfRecoveryInfo->lastRecTLI;
5994 
5995  /*
5996  * Actually, if WAL ended in an incomplete record, skip the parts that
5997  * made it through and start writing after the portion that persisted.
5998  * (It's critical to first write an OVERWRITE_CONTRECORD message, which
5999  * we'll do as soon as we're open for writing new WAL.)
6000  */
6002  {
6003  /*
6004  * We should only have a missingContrecPtr if we're not switching to a
6005  * new timeline. When a timeline switch occurs, WAL is copied from the
6006  * old timeline to the new only up to the end of the last complete
6007  * record, so there can't be an incomplete WAL record that we need to
6008  * disregard.
6009  */
6010  Assert(newTLI == endOfRecoveryInfo->lastRecTLI);
6012  EndOfLog = missingContrecPtr;
6013  }
6014 
6015  /*
6016  * Prepare to write WAL starting at EndOfLog location, and init xlog
6017  * buffer cache using the block containing the last record from the
6018  * previous incarnation.
6019  */
6020  Insert = &XLogCtl->Insert;
6021  Insert->PrevBytePos = XLogRecPtrToBytePos(endOfRecoveryInfo->lastRec);
6022  Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
6023 
6024  /*
6025  * Tricky point here: lastPage contains the *last* block that the LastRec
6026  * record spans, not the one it starts in. The last block is indeed the
6027  * one we want to use.
6028  */
6029  if (EndOfLog % XLOG_BLCKSZ != 0)
6030  {
6031  char *page;
6032  int len;
6033  int firstIdx;
6034 
6035  firstIdx = XLogRecPtrToBufIdx(EndOfLog);
6036  len = EndOfLog - endOfRecoveryInfo->lastPageBeginPtr;
6037  Assert(len < XLOG_BLCKSZ);
6038 
6039  /* Copy the valid part of the last block, and zero the rest */
6040  page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
6041  memcpy(page, endOfRecoveryInfo->lastPage, len);
6042  memset(page + len, 0, XLOG_BLCKSZ - len);
6043 
6044  pg_atomic_write_u64(&XLogCtl->xlblocks[firstIdx], endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ);
6045  XLogCtl->InitializedUpTo = endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ;
6046  }
6047  else
6048  {
6049  /*
6050  * There is no partial block to copy. Just set InitializedUpTo, and
6051  * let the first attempt to insert a log record to initialize the next
6052  * buffer.
6053  */
6054  XLogCtl->InitializedUpTo = EndOfLog;
6055  }
6056 
6057  /*
6058  * Update local and shared status. This is OK to do without any locks
6059  * because no other process can be reading or writing WAL yet.
6060  */
6061  LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
6065  XLogCtl->LogwrtRqst.Write = EndOfLog;
6066  XLogCtl->LogwrtRqst.Flush = EndOfLog;
6067 
6068  /*
6069  * Preallocate additional log files, if wanted.
6070  */
6071  PreallocXlogFiles(EndOfLog, newTLI);
6072 
6073  /*
6074  * Okay, we're officially UP.
6075  */
6076  InRecovery = false;
6077 
6078  /* start the archive_timeout timer and LSN running */
6079  XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
6080  XLogCtl->lastSegSwitchLSN = EndOfLog;
6081 
6082  /* also initialize latestCompletedXid, to nextXid - 1 */
6083  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
6086  LWLockRelease(ProcArrayLock);
6087 
6088  /*
6089  * Start up subtrans, if not already done for hot standby. (commit
6090  * timestamps are started below, if necessary.)
6091  */
6093  StartupSUBTRANS(oldestActiveXID);
6094 
6095  /*
6096  * Perform end of recovery actions for any SLRUs that need it.
6097  */
6098  TrimCLOG();
6099  TrimMultiXact();
6100 
6101  /*
6102  * Reload shared-memory state for prepared transactions. This needs to
6103  * happen before renaming the last partial segment of the old timeline as
6104  * it may be possible that we have to recovery some transactions from it.
6105  */
6107 
6108  /* Shut down xlogreader */
6110 
6111  /* Enable WAL writes for this backend only. */
6113 
6114  /* If necessary, write overwrite-contrecord before doing anything else */
6116  {
6119  }
6120 
6121  /*
6122  * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
6123  * record before resource manager writes cleanup WAL records or checkpoint
6124  * record is written.
6125  */
6126  Insert->fullPageWrites = lastFullPageWrites;
6128 
6129  /*
6130  * Emit checkpoint or end-of-recovery record in XLOG, if required.
6131  */
6132  if (performedWalRecovery)
6133  promoted = PerformRecoveryXLogAction();
6134 
6135  /*
6136  * If any of the critical GUCs have changed, log them before we allow
6137  * backends to write WAL.
6138  */
6140 
6141  /* If this is archive recovery, perform post-recovery cleanup actions. */
6143  CleanupAfterArchiveRecovery(EndOfLogTLI, EndOfLog, newTLI);
6144 
6145  /*
6146  * Local WAL inserts enabled, so it's time to finish initialization of
6147  * commit timestamp.
6148  */
6150 
6151  /*
6152  * All done with end-of-recovery actions.
6153  *
6154  * Now allow backends to write WAL and update the control file status in
6155  * consequence. SharedRecoveryState, that controls if backends can write
6156  * WAL, is updated while holding ControlFileLock to prevent other backends
6157  * to look at an inconsistent state of the control file in shared memory.
6158  * There is still a small window during which backends can write WAL and
6159  * the control file is still referring to a system not in DB_IN_PRODUCTION
6160  * state while looking at the on-disk control file.
6161  *
6162  * Also, we use info_lck to update SharedRecoveryState to ensure that
6163  * there are no race conditions concerning visibility of other recent
6164  * updates to shared memory.
6165  */
6166  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6168 
6172 
6174  LWLockRelease(ControlFileLock);
6175 
6176  /*
6177  * Shutdown the recovery environment. This must occur after
6178  * RecoverPreparedTransactions() (see notes in lock_twophase_recover())
6179  * and after switching SharedRecoveryState to RECOVERY_STATE_DONE so as
6180  * any session building a snapshot will not rely on KnownAssignedXids as
6181  * RecoveryInProgress() would return false at this stage. This is
6182  * particularly critical for prepared 2PC transactions, that would still
6183  * need to be included in snapshots once recovery has ended.
6184  */
6187 
6188  /*
6189  * If there were cascading standby servers connected to us, nudge any wal
6190  * sender processes to notice that we've been promoted.
6191  */
6192  WalSndWakeup(true, true);
6193 
6194  /*
6195  * If this was a promotion, request an (online) checkpoint now. This isn't
6196  * required for consistency, but the last restartpoint might be far back,
6197  * and in case of a crash, recovering from it might take a longer than is
6198  * appropriate now that we're not in standby mode anymore.
6199  */
6200  if (promoted)
6202 }
static void pg_atomic_write_membarrier_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:494
TimeLineID findNewestTimeLine(TimeLineID startTLI)
Definition: timeline.c:264
void restoreTimeLineHistoryFiles(TimeLineID begin, TimeLineID end)
Definition: timeline.c:50
void writeTimeLineHistory(TimeLineID newTLI, TimeLineID parentTLI, XLogRecPtr switchpoint, char *reason)
Definition: timeline.c:304
void startup_progress_timeout_handler(void)
Definition: startup.c:303
uint32 TransactionId
Definition: c.h:655
void StartupCLOG(void)
Definition: clog.c:877
void TrimCLOG(void)
Definition: clog.c:892
void StartupCommitTs(void)
Definition: commit_ts.c:632
void CompleteCommitTsInitialization(void)
Definition: commit_ts.c:642
void SyncDataDirectory(void)
Definition: fd.c:3567
#define IsBootstrapProcessingMode()
Definition: miscadmin.h:454
void TrimMultiXact(void)
Definition: multixact.c:2178
void StartupMultiXact(void)
Definition: multixact.c:2153
void StartupReplicationOrigin(void)
Definition: origin.c:703
@ DB_IN_PRODUCTION
Definition: pg_control.h:97
@ DB_IN_CRASH_RECOVERY
Definition: pg_control.h:95
const void size_t len
void pgstat_restore_stats(XLogRecPtr redo)
Definition: pgstat.c:476
void pgstat_discard_stats(void)
Definition: pgstat.c:488
void ProcArrayApplyRecoveryInfo(RunningTransactions running)
Definition: procarray.c:1054
void ProcArrayInitRecovery(TransactionId initializedUptoXID)
Definition: procarray.c:1023
static void set_ps_display(const char *activity)
Definition: ps_status.h:40
void ResetUnloggedRelations(int op)
Definition: reinit.c:47
#define UNLOGGED_RELATION_INIT
Definition: reinit.h:28
#define UNLOGGED_RELATION_CLEANUP
Definition: reinit.h:27
void RelationCacheInitFileRemove(void)
Definition: relcache.c:6797
void StartupReorderBuffer(void)
void StartupReplicationSlots(void)
Definition: slot.c:1924
void DeleteAllExportedSnapshotFiles(void)
Definition: snapmgr.c:1567
void InitRecoveryTransactionEnvironment(void)
Definition: standby.c:94
void ShutdownRecoveryTransactionEnvironment(void)
Definition: standby.c:160
@ SUBXIDS_IN_SUBTRANS
Definition: standby.h:82
XLogRecPtr lastPageBeginPtr
Definition: xlogrecovery.h:111
XLogRecPtr abortedRecPtr
Definition: xlogrecovery.h:120
XLogRecPtr missingContrecPtr
Definition: xlogrecovery.h:121
TimeLineID endOfLogTLI
Definition: xlogrecovery.h:109
TransactionId oldestRunningXid
Definition: standby.h:92
TransactionId nextXid
Definition: standby.h:91
TransactionId latestCompletedXid
Definition: standby.h:95
subxids_array_status subxid_status
Definition: standby.h:90
TransactionId * xids
Definition: standby.h:97
FullTransactionId latestCompletedXid
Definition: transam.h:238
pg_atomic_uint64 logInsertResult
Definition: xlog.c:471
void StartupSUBTRANS(TransactionId oldestActiveXID)
Definition: subtrans.c:309
TimeoutId RegisterTimeout(TimeoutId id, timeout_handler_proc handler)
Definition: timeout.c:505
@ STARTUP_PROGRESS_TIMEOUT
Definition: timeout.h:38
#define TransactionIdRetreat(dest)
Definition: transam.h:141
static void FullTransactionIdRetreat(FullTransactionId *dest)
Definition: transam.h:103
#define XidFromFullTransactionId(x)
Definition: transam.h:48
#define TransactionIdIsValid(xid)
Definition: transam.h:41
#define TransactionIdIsNormal(xid)
Definition: transam.h:42
void RecoverPreparedTransactions(void)
Definition: twophase.c:2090
void restoreTwoPhaseData(void)
Definition: twophase.c:1905
TransactionId PrescanPreparedTransactions(TransactionId **xids_p, int *nxids_p)
Definition: twophase.c:1969
void StandbyRecoverPreparedTransactions(void)
Definition: twophase.c:2049
void WalSndWakeup(bool physical, bool logical)
Definition: walsender.c:3638
void UpdateFullPageWrites(void)
Definition: xlog.c:8166
static char * str_time(pg_time_t tnow)
Definition: xlog.c:5199
static void ValidateXLOGDirectoryStructure(void)
Definition: xlog.c:4099
static XLogRecPtr CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn, XLogRecPtr pagePtr, TimeLineID newTLI)
Definition: xlog.c:7440
static void XLogReportParameters(void)
Definition: xlog.c:8103
static bool PerformRecoveryXLogAction(void)
Definition: xlog.c:6284
static void CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog, TimeLineID newTLI)
Definition: xlog.c:5289
static bool lastFullPageWrites
Definition: xlog.c:217
static void XLogInitNewTimeline(TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
Definition: xlog.c:5214
static void CheckRequiredParameterValues(void)
Definition: xlog.c:5385
static void RemoveTempXlogFiles(void)
Definition: xlog.c:3832
#define TABLESPACE_MAP_OLD
Definition: xlog.h:306
#define TABLESPACE_MAP
Definition: xlog.h:305
#define STANDBY_SIGNAL_FILE
Definition: xlog.h:301
#define BACKUP_LABEL_OLD
Definition: xlog.h:303
#define BACKUP_LABEL_FILE
Definition: xlog.h:302
#define RECOVERY_SIGNAL_FILE
Definition: xlog.h:300
@ RECOVERY_STATE_CRASH
Definition: xlog.h:91
@ RECOVERY_STATE_ARCHIVE
Definition: xlog.h:92
#define XRecOffIsValid(xlrp)
void ShutdownWalRecovery(void)
bool InArchiveRecovery
Definition: xlogrecovery.c:138
void PerformWalRecovery(void)
EndOfWalRecoveryInfo * FinishWalRecovery(void)
static XLogRecPtr missingContrecPtr
Definition: xlogrecovery.c:373
static XLogRecPtr abortedRecPtr
Definition: xlogrecovery.c:372
void InitWalRecovery(ControlFileData *ControlFile, bool *wasShutdown_ptr, bool *haveBackupLabel_ptr, bool *haveTblspcMap_ptr)
Definition: xlogrecovery.c:512
TimeLineID recoveryTargetTLI
Definition: xlogrecovery.c:122
HotStandbyState standbyState
Definition: xlogutils.c:53
bool InRecovery
Definition: xlogutils.c:50
@ STANDBY_DISABLED
Definition: xlogutils.h:52

References abortedRecPtr, EndOfWalRecoveryInfo::abortedRecPtr, AdvanceOldestClogXid(), ArchiveRecoveryRequested, Assert, AuxProcessResourceOwner, BACKUP_LABEL_FILE, BACKUP_LABEL_OLD, ControlFileData::backupEndRequired, ControlFileData::backupStartPoint, ControlFileData::checkPoint, CHECKPOINT_FORCE, ControlFileData::checkPointCopy, CheckRequiredParameterValues(), XLogCtlData::ckptFullXid, CleanupAfterArchiveRecovery(), CompleteCommitTsInitialization(), ControlFile, CreateOverwriteContrecordRecord(), CurrentResourceOwner, DB_IN_ARCHIVE_RECOVERY, DB_IN_CRASH_RECOVERY, DB_IN_PRODUCTION, DB_SHUTDOWNED, DB_SHUTDOWNED_IN_RECOVERY, DB_SHUTDOWNING, DEBUG1, DeleteAllExportedSnapshotFiles(), doPageWrites, durable_rename(), durable_unlink(), EnableHotStandby, EndOfWalRecoveryInfo::endOfLog, EndOfWalRecoveryInfo::endOfLogTLI, ereport, errcode(), ERRCODE_DATA_CORRUPTED, errhint(), errmsg(), errmsg_internal(), FATAL, findNewestTimeLine(), FinishWalRecovery(), FirstNormalUnloggedLSN, XLogwrtRqst::Flush, XLogwrtResult::Flush, CheckPoint::fullPageWrites, FullTransactionIdRetreat(), InArchiveRecovery, XLogCtlData::info_lck, XLogCtlData::InitializedUpTo, InitRecoveryTransactionEnvironment(), InitWalRecovery(), InRecovery, XLogCtlData::Insert, Insert(), XLogCtlData::InsertTimeLineID, InvalidXLogRecPtr, IsBootstrapProcessingMode, IsPostmasterEnvironment, lastFullPageWrites, EndOfWalRecoveryInfo::lastPage, EndOfWalRecoveryInfo::lastPageBeginPtr, EndOfWalRecoveryInfo::lastRec, EndOfWalRecoveryInfo::lastRecTLI, XLogCtlData::lastSegSwitchLSN, XLogCtlData::lastSegSwitchTime, TransamVariablesData::latestCompletedXid, RunningTransactionsData::latestCompletedXid, len, LocalMinRecoveryPoint, LocalMinRecoveryPointTLI, LocalSetXLogInsertAllowed(), LOG, XLogCtlData::logFlushResult, XLogCtlData::logInsertResult, XLogCtlData::logWriteResult, LogwrtResult, XLogCtlData::LogwrtRqst, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, missingContrecPtr, EndOfWalRecoveryInfo::missingContrecPtr, MultiXactSetNextMXact(), CheckPoint::newestCommitTsXid, CheckPoint::nextMulti, CheckPoint::nextMultiOffset, TransamVariablesData::nextOid, CheckPoint::nextOid, TransamVariablesData::nextXid, CheckPoint::nextXid, RunningTransactionsData::nextXid, NOTICE, TransamVariablesData::oidCount, CheckPoint::oldestActiveXid, CheckPoint::oldestCommitTsXid, CheckPoint::oldestMulti, CheckPoint::oldestMultiDB, RunningTransactionsData::oldestRunningXid, CheckPoint::oldestXid, CheckPoint::oldestXidDB, XLogCtlData::pages, PerformRecoveryXLogAction(), PerformWalRecovery(), pg_atomic_write_membarrier_u64(), pg_atomic_write_u64(), pg_usleep(), pgstat_discard_stats(), pgstat_restore_stats(), PreallocXlogFiles(), PrescanPreparedTransactions(), XLogCtlData::PrevTimeLineID, ProcArrayApplyRecoveryInfo(), ProcArrayInitRecovery(), RecoverPreparedTransactions(), RECOVERY_SIGNAL_FILE, EndOfWalRecoveryInfo::recovery_signal_file_found, RECOVERY_STATE_ARCHIVE, RECOVERY_STATE_CRASH, RECOVERY_STATE_DONE, EndOfWalRecoveryInfo::recoveryStopReason, recoveryTargetTLI, CheckPoint::redo, RedoRecPtr, XLogCtlInsert::RedoRecPtr, XLogCtlData::RedoRecPtr, RegisterTimeout(), RelationCacheInitFileRemove(), RemoveTempXlogFiles(), RequestCheckpoint(), ResetUnloggedRelations(), restoreTimeLineHistoryFiles(), restoreTwoPhaseData(), set_ps_display(), SetCommitTsLimit(), SetInstallXLogFileSegmentActive(), SetMultiXactIdLimit(), SetTransactionIdLimit(), XLogCtlData::SharedRecoveryState, ShutdownRecoveryTransactionEnvironment(), ShutdownWalRecovery(), SpinLockAcquire, SpinLockRelease, STANDBY_DISABLED, STANDBY_SIGNAL_FILE, EndOfWalRecoveryInfo::standby_signal_file_found, StandbyRecoverPreparedTransactions(), standbyState, STARTUP_PROGRESS_TIMEOUT, startup_progress_timeout_handler(), StartupCLOG(), StartupCommitTs(), StartupMultiXact(), StartupReorderBuffer(), StartupReplicationOrigin(), StartupReplicationSlots(), StartupSUBTRANS(), ControlFileData::state, str_time(), RunningTransactionsData::subxcnt, RunningTransactionsData::subxid_status, SUBXIDS_IN_SUBTRANS, SyncDataDirectory(), TABLESPACE_MAP, TABLESPACE_MAP_OLD, CheckPoint::ThisTimeLineID, CheckPoint::time, ControlFileData::time, ControlFileData::track_commit_timestamp, TransactionIdIsNormal, TransactionIdIsValid, TransactionIdRetreat, TransamVariables, TrimCLOG(), TrimMultiXact(), UNLOGGED_RELATION_CLEANUP, UNLOGGED_RELATION_INIT, XLogCtlData::unloggedLSN, ControlFileData::unloggedLSN, UpdateControlFile(), UpdateFullPageWrites(), ValidateXLOGDirectoryStructure(), WalSndWakeup(), XLogwrtRqst::Write, XLogwrtResult::Write, writeTimeLineHistory(), RunningTransactionsData::xcnt, XidFromFullTransactionId, RunningTransactionsData::xids, XLogCtlData::xlblocks, XLogCtl, XLogInitNewTimeline(), XLogRecPtrIsInvalid, XLogRecPtrToBufIdx, XLogRecPtrToBytePos(), XLogReportParameters(), and XRecOffIsValid.

Referenced by InitPostgres(), and StartupProcessMain().

◆ str_time()

static char * str_time ( pg_time_t  tnow)
static

Definition at line 5199 of file xlog.c.

5200 {
5201  char *buf = palloc(128);
5202 
5203  pg_strftime(buf, 128,
5204  "%Y-%m-%d %H:%M:%S %Z",
5205  pg_localtime(&tnow, log_timezone));
5206 
5207  return buf;
5208 }
static char * buf
Definition: pg_test_fsync.c:72
struct pg_tm * pg_localtime(const pg_time_t *timep, const pg_tz *tz)
Definition: localtime.c:1344
size_t pg_strftime(char *s, size_t maxsize, const char *format, const struct pg_tm *t)
Definition: strftime.c:128
PGDLLIMPORT pg_tz * log_timezone
Definition: pgtz.c:31

References buf, log_timezone, palloc(), pg_localtime(), and pg_strftime().

Referenced by StartupXLOG().

◆ SwitchIntoArchiveRecovery()

void SwitchIntoArchiveRecovery ( XLogRecPtr  EndRecPtr,
TimeLineID  replayTLI 
)

Definition at line 6209 of file xlog.c.

6210 {
6211  /* initialize minRecoveryPoint to this record */
6212  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6214  if (ControlFile->minRecoveryPoint < EndRecPtr)
6215  {
6216  ControlFile->minRecoveryPoint = EndRecPtr;
6217  ControlFile->minRecoveryPointTLI = replayTLI;
6218  }
6219  /* update local copy */
6222 
6223  /*
6224  * The startup process can update its local copy of minRecoveryPoint from
6225  * this point.
6226  */
6227  updateMinRecoveryPoint = true;
6228 
6230 
6231  /*
6232  * We update SharedRecoveryState while holding the lock on ControlFileLock
6233  * so both states are consistent in shared memory.
6234  */
6238 
6239  LWLockRelease(ControlFileLock);
6240 }
static bool updateMinRecoveryPoint
Definition: xlog.c:648

References ControlFile, DB_IN_ARCHIVE_RECOVERY, XLogCtlData::info_lck, LocalMinRecoveryPoint, LocalMinRecoveryPointTLI, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, RECOVERY_STATE_ARCHIVE, XLogCtlData::SharedRecoveryState, SpinLockAcquire, SpinLockRelease, ControlFileData::state, UpdateControlFile(), updateMinRecoveryPoint, and XLogCtl.

Referenced by ReadRecord().

◆ update_checkpoint_display()

static void update_checkpoint_display ( int  flags,
bool  restartpoint,
bool  reset 
)
static

Definition at line 6824 of file xlog.c.

6825 {
6826  /*
6827  * The status is reported only for end-of-recovery and shutdown
6828  * checkpoints or shutdown restartpoints. Updating the ps display is
6829  * useful in those situations as it may not be possible to rely on
6830  * pg_stat_activity to see the status of the checkpointer or the startup
6831  * process.
6832  */
6833  if ((flags & (CHECKPOINT_END_OF_RECOVERY | CHECKPOINT_IS_SHUTDOWN)) == 0)
6834  return;
6835 
6836  if (reset)
6837  set_ps_display("");
6838  else
6839  {
6840  char activitymsg[128];
6841 
6842  snprintf(activitymsg, sizeof(activitymsg), "performing %s%s%s",
6843  (flags & CHECKPOINT_END_OF_RECOVERY) ? "end-of-recovery " : "",
6844  (flags & CHECKPOINT_IS_SHUTDOWN) ? "shutdown " : "",
6845  restartpoint ? "restartpoint" : "checkpoint");
6846  set_ps_display(activitymsg);
6847  }
6848 }

References CHECKPOINT_END_OF_RECOVERY, CHECKPOINT_IS_SHUTDOWN, reset(), set_ps_display(), and snprintf.

Referenced by CreateCheckPoint(), and CreateRestartPoint().

◆ UpdateCheckPointDistanceEstimate()

static void UpdateCheckPointDistanceEstimate ( uint64  nbytes)
static

Definition at line 6786 of file xlog.c.

6787 {
6788  /*
6789  * To estimate the number of segments consumed between checkpoints, keep a
6790  * moving average of the amount of WAL generated in previous checkpoint
6791  * cycles. However, if the load is bursty, with quiet periods and busy
6792  * periods, we want to cater for the peak load. So instead of a plain
6793  * moving average, let the average decline slowly if the previous cycle
6794  * used less WAL than estimated, but bump it up immediately if it used
6795  * more.
6796  *
6797  * When checkpoints are triggered by max_wal_size, this should converge to
6798  * CheckpointSegments * wal_segment_size,
6799  *
6800  * Note: This doesn't pay any attention to what caused the checkpoint.
6801  * Checkpoints triggered manually with CHECKPOINT command, or by e.g.
6802  * starting a base backup, are counted the same as those created
6803  * automatically. The slow-decline will largely mask them out, if they are
6804  * not frequent. If they are frequent, it seems reasonable to count them
6805  * in as any others; if you issue a manual checkpoint every 5 minutes and
6806  * never let a timed checkpoint happen, it makes sense to base the
6807  * preallocation on that 5 minute interval rather than whatever
6808  * checkpoint_timeout is set to.
6809  */
6810  PrevCheckPointDistance = nbytes;
6811  if (CheckPointDistanceEstimate < nbytes)
6812  CheckPointDistanceEstimate = nbytes;
6813  else
6815  (0.90 * CheckPointDistanceEstimate + 0.10 * (double) nbytes);
6816 }

References CheckPointDistanceEstimate, and PrevCheckPointDistance.

Referenced by CreateCheckPoint(), and CreateRestartPoint().

◆ UpdateControlFile()

static void UpdateControlFile ( void  )
static

Definition at line 4559 of file xlog.c.

4560 {
4562 }
void update_controlfile(const char *DataDir, ControlFileData *ControlFile, bool do_sync)

References ControlFile, DataDir, and update_controlfile().

Referenced by CreateCheckPoint(), CreateEndOfRecoveryRecord(), CreateRestartPoint(), ReachedEndOfBackup(), StartupXLOG(), SwitchIntoArchiveRecovery(), UpdateMinRecoveryPoint(), xlog_redo(), and XLogReportParameters().

◆ UpdateFullPageWrites()

void UpdateFullPageWrites ( void  )

Definition at line 8166 of file xlog.c.

8167 {
8169  bool recoveryInProgress;
8170 
8171  /*
8172  * Do nothing if full_page_writes has not been changed.
8173  *
8174  * It's safe to check the shared full_page_writes without the lock,
8175  * because we assume that there is no concurrently running process which
8176  * can update it.
8177  */
8178  if (fullPageWrites == Insert->fullPageWrites)
8179  return;
8180 
8181  /*
8182  * Perform this outside critical section so that the WAL insert
8183  * initialization done by RecoveryInProgress() doesn't trigger an
8184  * assertion failure.
8185  */
8186  recoveryInProgress = RecoveryInProgress();
8187 
8189 
8190  /*
8191  * It's always safe to take full page images, even when not strictly
8192  * required, but not the other round. So if we're setting full_page_writes
8193  * to true, first set it true and then write the WAL record. If we're
8194  * setting it to false, first write the WAL record and then set the global
8195  * flag.
8196  */
8197  if (fullPageWrites)
8198  {
8200  Insert->fullPageWrites = true;
8202  }
8203 
8204  /*
8205  * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
8206  * full_page_writes during archive recovery, if required.
8207  */
8208  if (XLogStandbyInfoActive() && !recoveryInProgress)
8209  {
8210  XLogBeginInsert();
8211  XLogRegisterData((char *) (&fullPageWrites), sizeof(bool));
8212 
8213  XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
8214  }
8215 
8216  if (!fullPageWrites)
8217  {
8219  Insert->fullPageWrites = false;
8221  }
8222  END_CRIT_SECTION();
8223 }
#define XLOG_FPW_CHANGE
Definition: pg_control.h:76

References END_CRIT_SECTION, fullPageWrites, XLogCtlData::Insert, Insert(), RecoveryInProgress(), START_CRIT_SECTION, WALInsertLockAcquireExclusive(), WALInsertLockRelease(), XLOG_FPW_CHANGE, XLogBeginInsert(), XLogCtl, XLogInsert(), XLogRegisterData(), and XLogStandbyInfoActive.

Referenced by StartupXLOG(), and UpdateSharedMemoryConfig().

◆ UpdateLastRemovedPtr()

static void UpdateLastRemovedPtr ( char *  filename)
static

Definition at line 3812 of file xlog.c.

3813 {
3814  uint32 tli;
3815  XLogSegNo segno;
3816 
3817  XLogFromFileName(filename, &tli, &segno, wal_segment_size);
3818 
3820  if (segno > XLogCtl->lastRemovedSegNo)
3821  XLogCtl->lastRemovedSegNo = segno;
3823 }
static void XLogFromFileName(const char *fname, TimeLineID *tli, XLogSegNo *logSegNo, int wal_segsz_bytes)

References filename, XLogCtlData::info_lck, XLogCtlData::lastRemovedSegNo, SpinLockAcquire, SpinLockRelease, wal_segment_size, XLogCtl, and XLogFromFileName().

Referenced by RemoveOldXlogFiles().

◆ UpdateMinRecoveryPoint()

static void UpdateMinRecoveryPoint ( XLogRecPtr  lsn,
bool  force 
)
static

Definition at line 2722 of file xlog.c.

2723 {
2724  /* Quick check using our local copy of the variable */
2725  if (!updateMinRecoveryPoint || (!force && lsn <= LocalMinRecoveryPoint))
2726  return;
2727 
2728  /*
2729  * An invalid minRecoveryPoint means that we need to recover all the WAL,
2730  * i.e., we're doing crash recovery. We never modify the control file's
2731  * value in that case, so we can short-circuit future checks here too. The
2732  * local values of minRecoveryPoint and minRecoveryPointTLI should not be
2733  * updated until crash recovery finishes. We only do this for the startup
2734  * process as it should not update its own reference of minRecoveryPoint
2735  * until it has finished crash recovery to make sure that all WAL
2736  * available is replayed in this case. This also saves from extra locks
2737  * taken on the control file from the startup process.
2738  */
2740  {
2741  updateMinRecoveryPoint = false;
2742  return;
2743  }
2744 
2745  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
2746 
2747  /* update local copy */
2750 
2752  updateMinRecoveryPoint = false;
2753  else if (force || LocalMinRecoveryPoint < lsn)
2754  {
2755  XLogRecPtr newMinRecoveryPoint;
2756  TimeLineID newMinRecoveryPointTLI;
2757 
2758  /*
2759  * To avoid having to update the control file too often, we update it
2760  * all the way to the last record being replayed, even though 'lsn'
2761  * would suffice for correctness. This also allows the 'force' case
2762  * to not need a valid 'lsn' value.
2763  *
2764  * Another important reason for doing it this way is that the passed
2765  * 'lsn' value could be bogus, i.e., past the end of available WAL, if
2766  * the caller got it from a corrupted heap page. Accepting such a
2767  * value as the min recovery point would prevent us from coming up at
2768  * all. Instead, we just log a warning and continue with recovery.
2769  * (See also the comments about corrupt LSNs in XLogFlush.)
2770  */
2771  newMinRecoveryPoint = GetCurrentReplayRecPtr(&newMinRecoveryPointTLI);
2772  if (!force && newMinRecoveryPoint < lsn)
2773  elog(WARNING,
2774  "xlog min recovery request %X/%X is past current point %X/%X",
2775  LSN_FORMAT_ARGS(lsn), LSN_FORMAT_ARGS(newMinRecoveryPoint));
2776 
2777  /* update control file */
2778  if (ControlFile->minRecoveryPoint < newMinRecoveryPoint)
2779  {
2780  ControlFile->minRecoveryPoint = newMinRecoveryPoint;
2781  ControlFile->minRecoveryPointTLI = newMinRecoveryPointTLI;
2783  LocalMinRecoveryPoint = newMinRecoveryPoint;
2784  LocalMinRecoveryPointTLI = newMinRecoveryPointTLI;
2785 
2786  ereport(DEBUG2,
2787  (errmsg_internal("updated min recovery point to %X/%X on timeline %u",
2788  LSN_FORMAT_ARGS(newMinRecoveryPoint),
2789  newMinRecoveryPointTLI)));
2790  }
2791  }
2792  LWLockRelease(ControlFileLock);
2793 }
XLogRecPtr GetCurrentReplayRecPtr(TimeLineID *replayEndTLI)

References ControlFile, DEBUG2, elog, ereport, errmsg_internal(), GetCurrentReplayRecPtr(), InRecovery, LocalMinRecoveryPoint, LocalMinRecoveryPointTLI, LSN_FORMAT_ARGS, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, UpdateControlFile(), updateMinRecoveryPoint, WARNING, and XLogRecPtrIsInvalid.

Referenced by CreateRestartPoint(), XLogFlush(), and XLogInitNewTimeline().

◆ ValidateXLOGDirectoryStructure()

static void ValidateXLOGDirectoryStructure ( void  )
static

Definition at line 4099 of file xlog.c.

4100 {
4101  char path[MAXPGPATH];
4102  struct stat stat_buf;
4103 
4104  /* Check for pg_wal; if it doesn't exist, error out */
4105  if (stat(XLOGDIR, &stat_buf) != 0 ||
4106  !S_ISDIR(stat_buf.st_mode))
4107  ereport(FATAL,
4109  errmsg("required WAL directory \"%s\" does not exist",
4110  XLOGDIR)));
4111 
4112  /* Check for archive_status */
4113  snprintf(path, MAXPGPATH, XLOGDIR "/archive_status");
4114  if (stat(path, &stat_buf) == 0)
4115  {
4116  /* Check for weird cases where it exists but isn't a directory */
4117  if (!S_ISDIR(stat_buf.st_mode))
4118  ereport(FATAL,
4120  errmsg("required WAL directory \"%s\" does not exist",
4121  path)));
4122  }
4123  else
4124  {
4125  ereport(LOG,
4126  (errmsg("creating missing WAL directory \"%s\"", path)));
4127  if (MakePGDirectory(path) < 0)
4128  ereport(FATAL,
4130  errmsg("could not create missing directory \"%s\": %m",
4131  path)));
4132  }
4133 
4134  /* Check for summaries */
4135  snprintf(path, MAXPGPATH, XLOGDIR "/summaries");
4136  if (stat(path, &stat_buf) == 0)
4137  {
4138  /* Check for weird cases where it exists but isn't a directory */
4139  if (!S_ISDIR(stat_buf.st_mode))
4140  ereport(FATAL,
4141  (errmsg("required WAL directory \"%s\" does not exist",
4142  path)));
4143  }
4144  else
4145  {
4146  ereport(LOG,
4147  (errmsg("creating missing WAL directory \"%s\"", path)));
4148  if (MakePGDirectory(path) < 0)
4149  ereport(FATAL,
4150  (errmsg("could not create missing directory \"%s\": %m",
4151  path)));
4152  }
4153 }
int MakePGDirectory(const char *directoryName)
Definition: fd.c:3936
#define S_ISDIR(m)
Definition: win32_port.h:325

References ereport, errcode_for_file_access(), errmsg(), FATAL, LOG, MakePGDirectory(), MAXPGPATH, S_ISDIR, snprintf, stat::st_mode, stat, and XLOGDIR.

Referenced by StartupXLOG().

◆ WaitXLogInsertionsToFinish()

static XLogRecPtr WaitXLogInsertionsToFinish ( XLogRecPtr  upto)
static

Definition at line 1504 of file xlog.c.

1505 {
1506  uint64 bytepos;
1507  XLogRecPtr inserted;
1508  XLogRecPtr reservedUpto;
1509  XLogRecPtr finishedUpto;
1511  int i;
1512 
1513  if (MyProc == NULL)
1514  elog(PANIC, "cannot wait without a PGPROC structure");
1515 
1516  /*
1517  * Check if there's any work to do. Use a barrier to ensure we get the
1518  * freshest value.
1519  */
1521  if (upto <= inserted)
1522  return inserted;
1523 
1524  /* Read the current insert position */
1525  SpinLockAcquire(&Insert->insertpos_lck);
1526  bytepos = Insert->CurrBytePos;
1527  SpinLockRelease(&Insert->insertpos_lck);
1528  reservedUpto = XLogBytePosToEndRecPtr(bytepos);
1529 
1530  /*
1531  * No-one should request to flush a piece of WAL that hasn't even been
1532  * reserved yet. However, it can happen if there is a block with a bogus
1533  * LSN on disk, for example. XLogFlush checks for that situation and
1534  * complains, but only after the flush. Here we just assume that to mean
1535  * that all WAL that has been reserved needs to be finished. In this
1536  * corner-case, the return value can be smaller than 'upto' argument.
1537  */
1538  if (upto > reservedUpto)
1539  {
1540  ereport(LOG,
1541  (errmsg("request to flush past end of generated WAL; request %X/%X, current position %X/%X",
1542  LSN_FORMAT_ARGS(upto), LSN_FORMAT_ARGS(reservedUpto))));
1543  upto = reservedUpto;
1544  }
1545 
1546  /*
1547  * Loop through all the locks, sleeping on any in-progress insert older
1548  * than 'upto'.
1549  *
1550  * finishedUpto is our return value, indicating the point upto which all
1551  * the WAL insertions have been finished. Initialize it to the head of
1552  * reserved WAL, and as we iterate through the insertion locks, back it
1553  * out for any insertion that's still in progress.
1554  */
1555  finishedUpto = reservedUpto;
1556  for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1557  {
1558  XLogRecPtr insertingat = InvalidXLogRecPtr;
1559 
1560  do
1561  {
1562  /*
1563  * See if this insertion is in progress. LWLockWaitForVar will
1564  * wait for the lock to be released, or for the 'value' to be set
1565  * by a LWLockUpdateVar call. When a lock is initially acquired,
1566  * its value is 0 (InvalidXLogRecPtr), which means that we don't
1567  * know where it's inserting yet. We will have to wait for it. If
1568  * it's a small insertion, the record will most likely fit on the
1569  * same page and the inserter will release the lock without ever
1570  * calling LWLockUpdateVar. But if it has to sleep, it will
1571  * advertise the insertion point with LWLockUpdateVar before
1572  * sleeping.
1573  *
1574  * In this loop we are only waiting for insertions that started
1575  * before WaitXLogInsertionsToFinish was called. The lack of
1576  * memory barriers in the loop means that we might see locks as
1577  * "unused" that have since become used. This is fine because
1578  * they only can be used for later insertions that we would not
1579  * want to wait on anyway. Not taking a lock to acquire the
1580  * current insertingAt value means that we might see older
1581  * insertingAt values. This is also fine, because if we read a
1582  * value too old, we will add ourselves to the wait queue, which
1583  * contains atomic operations.
1584  */
1585  if (LWLockWaitForVar(&WALInsertLocks[i].l.lock,
1587  insertingat, &insertingat))
1588  {
1589  /* the lock was free, so no insertion in progress */
1590  insertingat = InvalidXLogRecPtr;
1591  break;
1592  }
1593 
1594  /*
1595  * This insertion is still in progress. Have to wait, unless the
1596  * inserter has proceeded past 'upto'.
1597  */
1598  } while (insertingat < upto);
1599 
1600  if (insertingat != InvalidXLogRecPtr && insertingat < finishedUpto)
1601  finishedUpto = insertingat;
1602  }
1603 
1604  /*
1605  * Advance the limit we know to have been inserted and return the freshest
1606  * value we know of, which might be beyond what we requested if somebody
1607  * is concurrently doing this with an 'upto' pointer ahead of us.
1608  */
1610  finishedUpto);
1611 
1612  return finishedUpto;
1613 }
static uint64 pg_atomic_monotonic_advance_u64(volatile pg_atomic_uint64 *ptr, uint64 target)
Definition: atomics.h:585
bool LWLockWaitForVar(LWLock *lock, pg_atomic_uint64 *valptr, uint64 oldval, uint64 *newval)
Definition: lwlock.c:1584
PGPROC * MyProc
Definition: proc.c:66
pg_atomic_uint64 insertingAt
Definition: xlog.c:370

References elog, ereport, errmsg(), i, XLogCtlData::Insert, Insert(), WALInsertLock::insertingAt, InvalidXLogRecPtr, WALInsertLockPadded::l, LOG, XLogCtlData::logInsertResult, LSN_FORMAT_ARGS, LWLockWaitForVar(), MyProc, NUM_XLOGINSERT_LOCKS, PANIC, pg_atomic_monotonic_advance_u64(), pg_atomic_read_membarrier_u64(), SpinLockAcquire, SpinLockRelease, WALInsertLocks, XLogBytePosToEndRecPtr(), and XLogCtl.

Referenced by AdvanceXLInsertBuffer(), XLogBackgroundFlush(), and XLogFlush().

◆ WALInsertLockAcquire()

static void WALInsertLockAcquire ( void  )
static

Definition at line 1371 of file xlog.c.

1372 {
1373  bool immed;
1374 
1375  /*
1376  * It doesn't matter which of the WAL insertion locks we acquire, so try
1377  * the one we used last time. If the system isn't particularly busy, it's
1378  * a good bet that it's still available, and it's good to have some
1379  * affinity to a particular lock so that you don't unnecessarily bounce
1380  * cache lines between processes when there's no contention.
1381  *
1382  * If this is the first time through in this backend, pick a lock
1383  * (semi-)randomly. This allows the locks to be used evenly if you have a
1384  * lot of very short connections.
1385  */
1386  static int lockToTry = -1;
1387 
1388  if (lockToTry == -1)
1389  lockToTry = MyProcNumber % NUM_XLOGINSERT_LOCKS;
1390  MyLockNo = lockToTry;
1391 
1392  /*
1393  * The insertingAt value is initially set to 0, as we don't know our
1394  * insert location yet.
1395  */
1397  if (!immed)
1398  {
1399  /*
1400  * If we couldn't get the lock immediately, try another lock next
1401  * time. On a system with more insertion locks than concurrent
1402  * inserters, this causes all the inserters to eventually migrate to a
1403  * lock that no-one else is using. On a system with more inserters
1404  * than locks, it still helps to distribute the inserters evenly
1405  * across the locks.
1406  */
1407  lockToTry = (lockToTry + 1) % NUM_XLOGINSERT_LOCKS;
1408  }
1409 }
ProcNumber MyProcNumber
Definition: globals.c:89
static int MyLockNo
Definition: xlog.c:651

References LW_EXCLUSIVE, LWLockAcquire(), MyLockNo, MyProcNumber, NUM_XLOGINSERT_LOCKS, and WALInsertLocks.

Referenced by CreateOverwriteContrecordRecord(), and XLogInsertRecord().

◆ WALInsertLockAcquireExclusive()

static void WALInsertLockAcquireExclusive ( void  )
static

Definition at line 1416 of file xlog.c.

1417 {
1418  int i;
1419 
1420  /*
1421  * When holding all the locks, all but the last lock's insertingAt
1422  * indicator is set to 0xFFFFFFFFFFFFFFFF, which is higher than any real
1423  * XLogRecPtr value, to make sure that no-one blocks waiting on those.
1424  */
1425  for (i = 0; i < NUM_XLOGINSERT_LOCKS - 1; i++)
1426  {
1428  LWLockUpdateVar(&WALInsertLocks[i].l.lock,
1430  PG_UINT64_MAX);
1431  }
1432  /* Variable value reset to 0 at release */
1434 
1435  holdingAllLocks = true;
1436 }
#define PG_UINT64_MAX
Definition: c.h:596
void LWLockUpdateVar(LWLock *lock, pg_atomic_uint64 *valptr, uint64 val)
Definition: lwlock.c:1720
static bool holdingAllLocks
Definition: xlog.c:652

References holdingAllLocks, i, WALInsertLock::insertingAt, WALInsertLockPadded::l, LW_EXCLUSIVE, LWLockAcquire(), LWLockUpdateVar(), NUM_XLOGINSERT_LOCKS, PG_UINT64_MAX, and WALInsertLocks.

Referenced by CreateCheckPoint(), CreateEndOfRecoveryRecord(), CreateRestartPoint(), do_pg_abort_backup(), do_pg_backup_start(), do_pg_backup_stop(), UpdateFullPageWrites(), and XLogInsertRecord().

◆ WALInsertLockRelease()

static void WALInsertLockRelease ( void  )
static

Definition at line 1445 of file xlog.c.

1446 {
1447  if (holdingAllLocks)
1448  {
1449  int i;
1450 
1451  for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
1454  0);
1455 
1456  holdingAllLocks = false;
1457  }
1458  else
1459  {
1462  0);
1463  }
1464 }
void LWLockReleaseClearVar(LWLock *lock, pg_atomic_uint64 *valptr, uint64 val)
Definition: lwlock.c:1854

References holdingAllLocks, i, WALInsertLock::insertingAt, WALInsertLockPadded::l, LWLockReleaseClearVar(), MyLockNo, NUM_XLOGINSERT_LOCKS, and WALInsertLocks.

Referenced by CreateCheckPoint(), CreateEndOfRecoveryRecord(), CreateOverwriteContrecordRecord(), CreateRestartPoint(), do_pg_abort_backup(), do_pg_backup_start(), do_pg_backup_stop(), UpdateFullPageWrites(), and XLogInsertRecord().

◆ WALInsertLockUpdateInsertingAt()

static void WALInsertLockUpdateInsertingAt ( XLogRecPtr  insertingAt)
static

Definition at line 1471 of file xlog.c.

1472 {
1473  if (holdingAllLocks)
1474  {
1475  /*
1476  * We use the last lock to mark our actual position, see comments in
1477  * WALInsertLockAcquireExclusive.
1478  */
1481  insertingAt);
1482  }
1483  else
1486  insertingAt);
1487 }

References holdingAllLocks, WALInsertLock::insertingAt, WALInsertLockPadded::l, LWLockUpdateVar(), MyLockNo, NUM_XLOGINSERT_LOCKS, and WALInsertLocks.

Referenced by GetXLogBuffer().

◆ WALReadFromBuffers()

Size WALReadFromBuffers ( char *  dstbuf,
XLogRecPtr  startptr,
Size  count,
TimeLineID  tli 
)

Definition at line 1748 of file xlog.c.

1750 {
1751  char *pdst = dstbuf;
1752  XLogRecPtr recptr = startptr;
1753  XLogRecPtr inserted;
1754  Size nbytes = count;
1755 
1756  if (RecoveryInProgress() || tli != GetWALInsertionTimeLine())
1757  return 0;
1758 
1759  Assert(!XLogRecPtrIsInvalid(startptr));
1760 
1761  /*
1762  * Caller should ensure that the requested data has been inserted into WAL
1763  * buffers before we try to read it.
1764  */
1766  if (startptr + count > inserted)
1767  ereport(ERROR,
1768  errmsg("cannot read past end of generated WAL: requested %X/%X, current position %X/%X",
1769  LSN_FORMAT_ARGS(startptr + count),
1770  LSN_FORMAT_ARGS(inserted)));
1771 
1772  /*
1773  * Loop through the buffers without a lock. For each buffer, atomically
1774  * read and verify the end pointer, then copy the data out, and finally
1775  * re-read and re-verify the end pointer.
1776  *
1777  * Once a page is evicted, it never returns to the WAL buffers, so if the
1778  * end pointer matches the expected end pointer before and after we copy
1779  * the data, then the right page must have been present during the data
1780  * copy. Read barriers are necessary to ensure that the data copy actually
1781  * happens between the two verification steps.
1782  *
1783  * If either verification fails, we simply terminate the loop and return
1784  * with the data that had been already copied out successfully.
1785  */
1786  while (nbytes > 0)
1787  {
1788  uint32 offset = recptr % XLOG_BLCKSZ;
1789  int idx = XLogRecPtrToBufIdx(recptr);
1790  XLogRecPtr expectedEndPtr;
1791  XLogRecPtr endptr;
1792  const char *page;
1793  const char *psrc;
1794  Size npagebytes;
1795 
1796  /*
1797  * Calculate the end pointer we expect in the xlblocks array if the
1798  * correct page is present.
1799  */
1800  expectedEndPtr = recptr + (XLOG_BLCKSZ - offset);
1801 
1802  /*
1803  * First verification step: check that the correct page is present in
1804  * the WAL buffers.
1805  */
1806  endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1807  if (expectedEndPtr != endptr)
1808  break;
1809 
1810  /*
1811  * The correct page is present (or was at the time the endptr was
1812  * read; must re-verify later). Calculate pointer to source data and
1813  * determine how much data to read from this page.
1814  */
1815  page = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1816  psrc = page + offset;
1817  npagebytes = Min(nbytes, XLOG_BLCKSZ - offset);
1818 
1819  /*
1820  * Ensure that the data copy and the first verification step are not
1821  * reordered.
1822  */
1823  pg_read_barrier();
1824 
1825  /* data copy */
1826  memcpy(pdst, psrc, npagebytes);
1827 
1828  /*
1829  * Ensure that the data copy and the second verification step are not
1830  * reordered.
1831  */
1832  pg_read_barrier();
1833 
1834  /*
1835  * Second verification step: check that the page we read from wasn't
1836  * evicted while we were copying the data.
1837  */
1838  endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1839  if (expectedEndPtr != endptr)
1840  break;
1841 
1842  pdst += npagebytes;
1843  recptr += npagebytes;
1844  nbytes -= npagebytes;
1845  }
1846 
1847  Assert(pdst - dstbuf <= count);
1848 
1849  return pdst - dstbuf;
1850 }
#define pg_read_barrier()
Definition: atomics.h:156
#define Min(x, y)
Definition: c.h:1007
TimeLineID GetWALInsertionTimeLine(void)
Definition: xlog.c:6520

References Assert, ereport, errmsg(), ERROR, GetWALInsertionTimeLine(), idx(), XLogCtlData::logInsertResult, LSN_FORMAT_ARGS, Min, XLogCtlData::pages, pg_atomic_read_u64(), pg_read_barrier, RecoveryInProgress(), XLogCtlData::xlblocks, XLogCtl, XLogRecPtrIsInvalid, and XLogRecPtrToBufIdx.

Referenced by XLogSendPhysical().

◆ WriteControlFile()

static void WriteControlFile ( void  )
static

Definition at line 4239 of file xlog.c.

4240 {
4241  int fd;
4242  char buffer[PG_CONTROL_FILE_SIZE]; /* need not be aligned */
4243 
4244  /*
4245  * Initialize version and compatibility-check fields
4246  */
4249 
4250  ControlFile->maxAlign = MAXIMUM_ALIGNOF;
4252 
4253  ControlFile->blcksz = BLCKSZ;
4254  ControlFile->relseg_size = RELSEG_SIZE;
4255  ControlFile->xlog_blcksz = XLOG_BLCKSZ;
4257 
4260 
4263 
4265 
4266  /* Contents are protected with a CRC */
4269  (char *) ControlFile,
4270  offsetof(ControlFileData, crc));
4272 
4273  /*
4274  * We write out PG_CONTROL_FILE_SIZE bytes into pg_control, zero-padding
4275  * the excess over sizeof(ControlFileData). This reduces the odds of
4276  * premature-EOF errors when reading pg_control. We'll still fail when we
4277  * check the contents of the file, but hopefully with a more specific
4278  * error than "couldn't read pg_control".
4279  */
4280  memset(buffer, 0, PG_CONTROL_FILE_SIZE);
4281  memcpy(buffer, ControlFile, sizeof(ControlFileData));
4282 
4284  O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
4285  if (fd < 0)
4286  ereport(PANIC,
4288  errmsg("could not create file \"%s\": %m",
4289  XLOG_CONTROL_FILE)));
4290 
4291  errno = 0;
4292  pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_WRITE);
4294  {
4295  /* if write didn't set errno, assume problem is no disk space */
4296  if (errno == 0)
4297  errno = ENOSPC;
4298  ereport(PANIC,
4300  errmsg("could not write to file \"%s\": %m",
4301  XLOG_CONTROL_FILE)));
4302  }
4304 
4305  pgstat_report_wait_start(WAIT_EVENT_CONTROL_FILE_SYNC);
4306  if (pg_fsync(fd) != 0)
4307  ereport(PANIC,
4309  errmsg("could not fsync file \"%s\": %m",
4310  XLOG_CONTROL_FILE)));
4312 
4313  if (close(fd) != 0)
4314  ereport(PANIC,
4316  errmsg("could not close file \"%s\": %m",
4317  XLOG_CONTROL_FILE)));
4318 }
#define FLOAT8PASSBYVAL
Definition: c.h:638
#define PG_CONTROL_FILE_SIZE
Definition: pg_control.h:250

References BasicOpenFile(), ControlFileData::blcksz, CATALOG_VERSION_NO, ControlFileData::catalog_version_no, close, COMP_CRC32C, ControlFile, ControlFileData::crc, crc, ereport, errcode_for_file_access(), errmsg(), fd(), FIN_CRC32C, ControlFileData::float8ByVal, FLOAT8PASSBYVAL, ControlFileData::floatFormat, FLOATFORMAT_VALUE, INDEX_MAX_KEYS, ControlFileData::indexMaxKeys, INIT_CRC32C, ControlFileData::loblksize, LOBLKSIZE, ControlFileData::maxAlign, ControlFileData::nameDataLen, NAMEDATALEN, PANIC, PG_BINARY, PG_CONTROL_FILE_SIZE, PG_CONTROL_VERSION, ControlFileData::pg_control_version, pg_fsync(), pgstat_report_wait_end(), pgstat_report_wait_start(), ControlFileData::relseg_size, TOAST_MAX_CHUNK_SIZE, ControlFileData::toast_max_chunk_size, wal_segment_size, write, ControlFileData::xlog_blcksz, XLOG_CONTROL_FILE, and ControlFileData::xlog_seg_size.

Referenced by BootStrapXLOG().

◆ xlog_redo()

void xlog_redo ( XLogReaderState record)

Definition at line 8235 of file xlog.c.

8236 {
8237  uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
8238  XLogRecPtr lsn = record->EndRecPtr;
8239 
8240  /*
8241  * In XLOG rmgr, backup blocks are only used by XLOG_FPI and
8242  * XLOG_FPI_FOR_HINT records.
8243  */
8244  Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
8245  !XLogRecHasAnyBlockRefs(record));
8246 
8247  if (info == XLOG_NEXTOID)
8248  {
8249  Oid nextOid;
8250 
8251  /*
8252  * We used to try to take the maximum of TransamVariables->nextOid and
8253  * the recorded nextOid, but that fails if the OID counter wraps
8254  * around. Since no OID allocation should be happening during replay
8255  * anyway, better to just believe the record exactly. We still take
8256  * OidGenLock while setting the variable, just in case.
8257  */
8258  memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
8259  LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8260  TransamVariables->nextOid = nextOid;
8262  LWLockRelease(OidGenLock);
8263  }
8264  else if (info == XLOG_CHECKPOINT_SHUTDOWN)
8265  {
8266  CheckPoint checkPoint;
8267  TimeLineID replayTLI;
8268 
8269  memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8270  /* In a SHUTDOWN checkpoint, believe the counters exactly */
8271  LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8272  TransamVariables->nextXid = checkPoint.nextXid;
8273  LWLockRelease(XidGenLock);
8274  LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8275  TransamVariables->nextOid = checkPoint.nextOid;
8277  LWLockRelease(OidGenLock);
8278  MultiXactSetNextMXact(checkPoint.nextMulti,
8279  checkPoint.nextMultiOffset);
8280 
8282  checkPoint.oldestMultiDB);
8283 
8284  /*
8285  * No need to set oldestClogXid here as well; it'll be set when we
8286  * redo an xl_clog_truncate if it changed since initialization.
8287  */
8288  SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
8289 
8290  /*
8291  * If we see a shutdown checkpoint while waiting for an end-of-backup
8292  * record, the backup was canceled and the end-of-backup record will
8293  * never arrive.
8294  */
8298  ereport(PANIC,
8299  (errmsg("online backup was canceled, recovery cannot continue")));
8300 
8301  /*
8302  * If we see a shutdown checkpoint, we know that nothing was running
8303  * on the primary at this point. So fake-up an empty running-xacts
8304  * record and use that here and now. Recover additional standby state
8305  * for prepared transactions.
8306  */
8308  {
8309  TransactionId *xids;
8310  int nxids;
8311  TransactionId oldestActiveXID;
8312  TransactionId latestCompletedXid;
8313  RunningTransactionsData running;
8314 
8315  oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
8316 
8317  /* Update pg_subtrans entries for any prepared transactions */
8319 
8320  /*
8321  * Construct a RunningTransactions snapshot representing a shut
8322  * down server, with only prepared transactions still alive. We're
8323  * never overflowed at this point because all subxids are listed
8324  * with their parent prepared transactions.
8325  */
8326  running.xcnt = nxids;
8327  running.subxcnt = 0;
8329  running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
8330  running.oldestRunningXid = oldestActiveXID;
8331  latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
8332  TransactionIdRetreat(latestCompletedXid);
8333  Assert(TransactionIdIsNormal(latestCompletedXid));
8334  running.latestCompletedXid = latestCompletedXid;
8335  running.xids = xids;
8336 
8337  ProcArrayApplyRecoveryInfo(&running);
8338  }
8339 
8340  /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8341  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8342  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8343  LWLockRelease(ControlFileLock);
8344 
8345  /* Update shared-memory copy of checkpoint XID/epoch */
8347  XLogCtl->ckptFullXid = checkPoint.nextXid;
8349 
8350  /*
8351  * We should've already switched to the new TLI before replaying this
8352  * record.
8353  */
8354  (void) GetCurrentReplayRecPtr(&replayTLI);
8355  if (checkPoint.ThisTimeLineID != replayTLI)
8356  ereport(PANIC,
8357  (errmsg("unexpected timeline ID %u (should be %u) in shutdown checkpoint record",
8358  checkPoint.ThisTimeLineID, replayTLI)));
8359 
8360  RecoveryRestartPoint(&checkPoint, record);
8361  }
8362  else if (info == XLOG_CHECKPOINT_ONLINE)
8363  {
8364  CheckPoint checkPoint;
8365  TimeLineID replayTLI;
8366 
8367  memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8368  /* In an ONLINE checkpoint, treat the XID counter as a minimum */
8369  LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8371  checkPoint.nextXid))
8372  TransamVariables->nextXid = checkPoint.nextXid;
8373  LWLockRelease(XidGenLock);
8374 
8375  /*
8376  * We ignore the nextOid counter in an ONLINE checkpoint, preferring
8377  * to track OID assignment through XLOG_NEXTOID records. The nextOid
8378  * counter is from the start of the checkpoint and might well be stale
8379  * compared to later XLOG_NEXTOID records. We could try to take the
8380  * maximum of the nextOid counter and our latest value, but since
8381  * there's no particular guarantee about the speed with which the OID
8382  * counter wraps around, that's a risky thing to do. In any case,
8383  * users of the nextOid counter are required to avoid assignment of
8384  * duplicates, so that a somewhat out-of-date value should be safe.
8385  */
8386 
8387  /* Handle multixact */
8389  checkPoint.nextMultiOffset);
8390 
8391  /*
8392  * NB: This may perform multixact truncation when replaying WAL
8393  * generated by an older primary.
8394  */
8396  checkPoint.oldestMultiDB);
8398  checkPoint.oldestXid))
8399  SetTransactionIdLimit(checkPoint.oldestXid,
8400  checkPoint.oldestXidDB);
8401  /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8402  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8403  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8404  LWLockRelease(ControlFileLock);
8405 
8406  /* Update shared-memory copy of checkpoint XID/epoch */
8408  XLogCtl->ckptFullXid = checkPoint.nextXid;
8410 
8411  /* TLI should not change in an on-line checkpoint */
8412  (void) GetCurrentReplayRecPtr(&replayTLI);
8413  if (checkPoint.ThisTimeLineID != replayTLI)
8414  ereport(PANIC,
8415  (errmsg("unexpected timeline ID %u (should be %u) in online checkpoint record",
8416  checkPoint.ThisTimeLineID, replayTLI)));
8417 
8418  RecoveryRestartPoint(&checkPoint, record);
8419  }
8420  else if (info == XLOG_OVERWRITE_CONTRECORD)
8421  {
8422  /* nothing to do here, handled in xlogrecovery_redo() */
8423  }
8424  else if (info == XLOG_END_OF_RECOVERY)
8425  {
8426  xl_end_of_recovery xlrec;
8427  TimeLineID replayTLI;
8428 
8429  memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
8430 
8431  /*
8432  * For Hot Standby, we could treat this like a Shutdown Checkpoint,
8433  * but this case is rarer and harder to test, so the benefit doesn't
8434  * outweigh the potential extra cost of maintenance.
8435  */
8436 
8437  /*
8438  * We should've already switched to the new TLI before replaying this
8439  * record.
8440  */
8441  (void) GetCurrentReplayRecPtr(&replayTLI);
8442  if (xlrec.ThisTimeLineID != replayTLI)
8443  ereport(PANIC,
8444  (errmsg("unexpected timeline ID %u (should be %u) in end-of-recovery record",
8445  xlrec.ThisTimeLineID, replayTLI)));
8446  }
8447  else if (info == XLOG_NOOP)
8448  {
8449  /* nothing to do here */
8450  }
8451  else if (info == XLOG_SWITCH)
8452  {
8453  /* nothing to do here */
8454  }
8455  else if (info == XLOG_RESTORE_POINT)
8456  {
8457  /* nothing to do here, handled in xlogrecovery.c */
8458  }
8459  else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
8460  {
8461  /*
8462  * XLOG_FPI records contain nothing else but one or more block
8463  * references. Every block reference must include a full-page image
8464  * even if full_page_writes was disabled when the record was generated
8465  * - otherwise there would be no point in this record.
8466  *
8467  * XLOG_FPI_FOR_HINT records are generated when a page needs to be
8468  * WAL-logged because of a hint bit update. They are only generated
8469  * when checksums and/or wal_log_hints are enabled. They may include
8470  * no full-page images if full_page_writes was disabled when they were
8471  * generated. In this case there is nothing to do here.
8472  *
8473  * No recovery conflicts are generated by these generic records - if a
8474  * resource manager needs to generate conflicts, it has to define a
8475  * separate WAL record type and redo routine.
8476  */
8477  for (uint8 block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
8478  {
8479  Buffer buffer;
8480 
8481  if (!XLogRecHasBlockImage(record, block_id))
8482  {
8483  if (info == XLOG_FPI)
8484  elog(ERROR, "XLOG_FPI record did not contain a full-page image");
8485  continue;
8486  }
8487 
8488  if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
8489  elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
8490  UnlockReleaseBuffer(buffer);
8491  }
8492  }
8493  else if (info == XLOG_BACKUP_END)
8494  {
8495  /* nothing to do here, handled in xlogrecovery_redo() */
8496  }
8497  else if (info == XLOG_PARAMETER_CHANGE)
8498  {
8499  xl_parameter_change xlrec;
8500 
8501  /* Update our copy of the parameters in pg_control */
8502  memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
8503 
8504  /*
8505  * Invalidate logical slots if we are in hot standby and the primary
8506  * does not have a WAL level sufficient for logical decoding. No need
8507  * to search for potentially conflicting logically slots if standby is
8508  * running with wal_level lower than logical, because in that case, we
8509  * would have either disallowed creation of logical slots or
8510  * invalidated existing ones.
8511  */
8512  if (InRecovery && InHotStandby &&
8513  xlrec.wal_level < WAL_LEVEL_LOGICAL &&
8516  0, InvalidOid,
8518 
8519  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8525  ControlFile->wal_level = xlrec.wal_level;
8527 
8528  /*
8529  * Update minRecoveryPoint to ensure that if recovery is aborted, we
8530  * recover back up to this point before allowing hot standby again.
8531  * This is important if the max_* settings are decreased, to ensure
8532  * you don't run queries against the WAL preceding the change. The
8533  * local copies cannot be updated as long as crash recovery is
8534  * happening and we expect all the WAL to be replayed.
8535  */
8536  if (InArchiveRecovery)
8537  {
8540  }
8542  {
8543  TimeLineID replayTLI;
8544 
8545  (void) GetCurrentReplayRecPtr(&replayTLI);
8547  ControlFile->minRecoveryPointTLI = replayTLI;
8548  }
8549 
8553 
8555  LWLockRelease(ControlFileLock);
8556 
8557  /* Check to see if any parameter change gives a problem on recovery */
8559  }
8560  else if (info == XLOG_FPW_CHANGE)
8561  {
8562  bool fpw;
8563 
8564  memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
8565 
8566  /*
8567  * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
8568  * do_pg_backup_start() and do_pg_backup_stop() can check whether
8569  * full_page_writes has been disabled during online backup.
8570  */
8571  if (!fpw)
8572  {
8574  if (XLogCtl->lastFpwDisableRecPtr < record->ReadRecPtr)
8577  }
8578 
8579  /* Keep track of full_page_writes */
8580  lastFullPageWrites = fpw;
8581  }
8582  else if (info == XLOG_CHECKPOINT_REDO)
8583  {
8584  /* nothing to do here, just for informational purposes */
8585  }
8586 }
int Buffer
Definition: buf.h:23
void UnlockReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:4941
unsigned char uint8
Definition: c.h:516
void CommitTsParameterChange(bool newvalue, bool oldvalue)
Definition: commit_ts.c:664
void MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
Definition: multixact.c:2536
void MultiXactAdvanceNextMXact(MultiXactId minMulti, MultiXactOffset minMultiOffset)
Definition: multixact.c:2511
#define XLOG_RESTORE_POINT
Definition: pg_control.h:75
#define XLOG_FPI
Definition: pg_control.h:79
#define XLOG_FPI_FOR_HINT
Definition: pg_control.h:78
#define XLOG_NEXTOID
Definition: pg_control.h:71
#define XLOG_NOOP
Definition: pg_control.h:70
#define XLOG_PARAMETER_CHANGE
Definition: pg_control.h:74
@ RS_INVAL_WAL_LEVEL
Definition: slot.h:58
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:280
#define FullTransactionIdPrecedes(a, b)
Definition: transam.h:51
static void RecoveryRestartPoint(const CheckPoint *checkPoint, XLogReaderState *record)
Definition: xlog.c:7550
@ WAL_LEVEL_LOGICAL
Definition: xlog.h:76
#define XLogRecGetInfo(decoder)
Definition: xlogreader.h:410
#define XLogRecGetData(decoder)
Definition: xlogreader.h:415
#define XLogRecMaxBlockId(decoder)
Definition: xlogreader.h:418
#define XLogRecHasBlockImage(decoder, block_id)
Definition: xlogreader.h:423
#define XLogRecHasAnyBlockRefs(decoder)
Definition: xlogreader.h:417
#define XLR_INFO_MASK
Definition: xlogrecord.h:62
XLogRedoAction XLogReadBufferForRedo(XLogReaderState *record, uint8 block_id, Buffer *buf)
Definition: xlogutils.c:314
@ STANDBY_INITIALIZED
Definition: xlogutils.h:53
#define InHotStandby
Definition: xlogutils.h:60
@ BLK_RESTORED
Definition: xlogutils.h:76

References ArchiveRecoveryRequested, Assert, ControlFileData::backupEndPoint, ControlFileData::backupStartPoint, BLK_RESTORED, ControlFileData::checkPointCopy, CheckRequiredParameterValues(), XLogCtlData::ckptFullXid, CommitTsParameterChange(), ControlFile, elog, XLogReaderState::EndRecPtr, ereport, errmsg(), ERROR, FullTransactionIdPrecedes, GetCurrentReplayRecPtr(), InArchiveRecovery, XLogCtlData::info_lck, InHotStandby, InRecovery, InvalidateObsoleteReplicationSlots(), InvalidOid, InvalidTransactionId, InvalidXLogRecPtr, XLogCtlData::lastFpwDisableRecPtr, lastFullPageWrites, RunningTransactionsData::latestCompletedXid, LocalMinRecoveryPoint, LocalMinRecoveryPointTLI, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), xl_parameter_change::max_locks_per_xact, ControlFileData::max_locks_per_xact, xl_parameter_change::max_prepared_xacts, ControlFileData::max_prepared_xacts, xl_parameter_change::max_wal_senders, ControlFileData::max_wal_senders, xl_parameter_change::max_worker_processes, ControlFileData::max_worker_processes, xl_parameter_change::MaxConnections, ControlFileData::MaxConnections, ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, MultiXactAdvanceNextMXact(), MultiXactAdvanceOldest(), MultiXactSetNextMXact(), CheckPoint::nextMulti, CheckPoint::nextMultiOffset, TransamVariablesData::nextOid, CheckPoint::nextOid, TransamVariablesData::nextXid, CheckPoint::nextXid, RunningTransactionsData::nextXid, TransamVariablesData::oidCount, CheckPoint::oldestMulti, CheckPoint::oldestMultiDB, RunningTransactionsData::oldestRunningXid, TransamVariablesData::oldestXid, CheckPoint::oldestXid, CheckPoint::oldestXidDB, PANIC, PrescanPreparedTransactions(), ProcArrayApplyRecoveryInfo(), XLogReaderState::ReadRecPtr, RecoveryRestartPoint(), RS_INVAL_WAL_LEVEL, SetTransactionIdLimit(), SpinLockAcquire, SpinLockRelease, STANDBY_INITIALIZED, StandbyRecoverPreparedTransactions(), standbyState, RunningTransactionsData::subxcnt, RunningTransactionsData::subxid_status, SUBXIDS_IN_SUBTRANS, xl_end_of_recovery::ThisTimeLineID, CheckPoint::ThisTimeLineID, xl_parameter_change::track_commit_timestamp, ControlFileData::track_commit_timestamp, TransactionIdIsNormal, TransactionIdPrecedes(), TransactionIdRetreat, TransamVariables, UnlockReleaseBuffer(), UpdateControlFile(), wal_level, xl_parameter_change::wal_level, ControlFileData::wal_level, WAL_LEVEL_LOGICAL, xl_parameter_change::wal_log_hints, ControlFileData::wal_log_hints, RunningTransactionsData::xcnt, XidFromFullTransactionId, RunningTransactionsData::xids, XLOG_BACKUP_END, XLOG_CHECKPOINT_ONLINE, XLOG_CHECKPOINT_REDO, XLOG_CHECKPOINT_SHUTDOWN, XLOG_END_OF_RECOVERY, XLOG_FPI, XLOG_FPI_FOR_HINT, XLOG_FPW_CHANGE, XLOG_NEXTOID, XLOG_NOOP, XLOG_OVERWRITE_CONTRECORD, XLOG_PARAMETER_CHANGE, XLOG_RESTORE_POINT, XLOG_SWITCH, XLogCtl, XLogReadBufferForRedo(), XLogRecGetData, XLogRecGetInfo, XLogRecHasAnyBlockRefs, XLogRecHasBlockImage, XLogRecMaxBlockId, XLogRecPtrIsInvalid, and XLR_INFO_MASK.

◆ XLogBackgroundFlush()

bool XLogBackgroundFlush ( void  )

Definition at line 2990 of file xlog.c.

2991 {
2992  XLogwrtRqst WriteRqst;
2993  bool flexible = true;
2994  static TimestampTz lastflush;
2995  TimestampTz now;
2996  int flushblocks;
2997  TimeLineID insertTLI;
2998 
2999  /* XLOG doesn't need flushing during recovery */
3000  if (RecoveryInProgress())
3001  return false;
3002 
3003  /*
3004  * Since we're not in recovery, InsertTimeLineID is set and can't change,
3005  * so we can read it without a lock.
3006  */
3007  insertTLI = XLogCtl->InsertTimeLineID;
3008 
3009  /* read updated LogwrtRqst */
3011  WriteRqst = XLogCtl->LogwrtRqst;
3013 
3014  /* back off to last completed page boundary */
3015  WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
3016 
3017  /* if we have already flushed that far, consider async commit records */
3019  if (WriteRqst.Write <= LogwrtResult.Flush)
3020  {
3022  WriteRqst.Write = XLogCtl->asyncXactLSN;
3024  flexible = false; /* ensure it all gets written */
3025  }
3026 
3027  /*
3028  * If already known flushed, we're done. Just need to check if we are
3029  * holding an open file handle to a logfile that's no longer in use,
3030  * preventing the file from being deleted.
3031  */
3032  if (WriteRqst.Write <= LogwrtResult.Flush)
3033  {
3034  if (openLogFile >= 0)
3035  {
3038  {
3039  XLogFileClose();
3040  }
3041  }
3042  return false;
3043  }
3044 
3045  /*
3046  * Determine how far to flush WAL, based on the wal_writer_delay and
3047  * wal_writer_flush_after GUCs.
3048  *
3049  * Note that XLogSetAsyncXactLSN() performs similar calculation based on
3050  * wal_writer_flush_after, to decide when to wake us up. Make sure the
3051  * logic is the same in both places if you change this.
3052  */
3054  flushblocks =
3055  WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
3056 
3057  if (WalWriterFlushAfter == 0 || lastflush == 0)
3058  {
3059  /* first call, or block based limits disabled */
3060  WriteRqst.Flush = WriteRqst.Write;
3061  lastflush = now;
3062  }
3063  else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
3064  {
3065  /*
3066  * Flush the writes at least every WalWriterDelay ms. This is
3067  * important to bound the amount of time it takes for an asynchronous
3068  * commit to hit disk.
3069  */
3070  WriteRqst.Flush = WriteRqst.Write;
3071  lastflush = now;
3072  }
3073  else if (flushblocks >= WalWriterFlushAfter)
3074  {
3075  /* exceeded wal_writer_flush_after blocks, flush */
3076  WriteRqst.Flush = WriteRqst.Write;
3077  lastflush = now;
3078  }
3079  else
3080  {
3081  /* no flushing, this time round */
3082  WriteRqst.Flush = 0;
3083  }
3084 
3085 #ifdef WAL_DEBUG
3086  if (XLOG_DEBUG)
3087  elog(LOG, "xlog bg flush request write %X/%X; flush: %X/%X, current is write %X/%X; flush %X/%X",
3088  LSN_FORMAT_ARGS(WriteRqst.Write),
3089  LSN_FORMAT_ARGS(WriteRqst.Flush),
3092 #endif
3093 
3095 
3096  /* now wait for any in-progress insertions to finish and get write lock */
3097  WaitXLogInsertionsToFinish(WriteRqst.Write);
3098  LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
3100  if (WriteRqst.Write > LogwrtResult.Write ||
3101  WriteRqst.Flush > LogwrtResult.Flush)
3102  {
3103  XLogWrite(WriteRqst, insertTLI, flexible);
3104  }
3105  LWLockRelease(WALWriteLock);
3106 
3107  END_CRIT_SECTION();
3108 
3109  /* wake up walsenders now that we've released heavily contended locks */
3111 
3112  /*
3113  * Great, done. To take some work off the critical path, try to initialize
3114  * as many of the no-longer-needed WAL buffers for future use as we can.
3115  */
3116  AdvanceXLInsertBuffer(InvalidXLogRecPtr, insertTLI, true);
3117 
3118  /*
3119  * If we determined that we need to write data, but somebody else
3120  * wrote/flushed already, it should be considered as being active, to
3121  * avoid hibernating too early.
3122  */
3123  return true;
3124 }
bool TimestampDifferenceExceeds(TimestampTz start_time, TimestampTz stop_time, int msec)
Definition: timestamp.c:1780
Datum now(PG_FUNCTION_ARGS)
Definition: timestamp.c:1608
XLogRecPtr asyncXactLSN
Definition: xlog.c:458
static void WalSndWakeupProcessRequests(bool physical, bool logical)
Definition: walsender.h:65
int WalWriterFlushAfter
Definition: walwriter.c:70
int WalWriterDelay
Definition: walwriter.c:69
#define XLByteInPrevSeg(xlrp, logSegNo, wal_segsz_bytes)

References AdvanceXLInsertBuffer(), XLogCtlData::asyncXactLSN, elog, END_CRIT_SECTION, XLogwrtRqst::Flush, XLogwrtResult::Flush, GetCurrentTimestamp(), XLogCtlData::info_lck, XLogCtlData::InsertTimeLineID, InvalidXLogRecPtr, LOG, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), now(), openLogFile, openLogSegNo, RecoveryInProgress(), RefreshXLogWriteResult, SpinLockAcquire, SpinLockRelease, START_CRIT_SECTION, TimestampDifferenceExceeds(), WaitXLogInsertionsToFinish(), wal_segment_size, WalSndWakeupProcessRequests(), WalWriterDelay, WalWriterFlushAfter, XLogwrtRqst::Write, XLogwrtResult::Write, XLByteInPrevSeg, XLogCtl, XLogFileClose(), and XLogWrite().

Referenced by WalSndWaitForWal(), and WalWriterMain().

◆ XLogBytePosToEndRecPtr()

static XLogRecPtr XLogBytePosToEndRecPtr ( uint64  bytepos)
static

Definition at line 1898 of file xlog.c.

1899 {
1900  uint64 fullsegs;
1901  uint64 fullpages;
1902  uint64 bytesleft;
1903  uint32 seg_offset;
1904  XLogRecPtr result;
1905 
1906  fullsegs = bytepos / UsableBytesInSegment;
1907  bytesleft = bytepos % UsableBytesInSegment;
1908 
1909  if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1910  {
1911  /* fits on first page of segment */
1912  if (bytesleft == 0)
1913  seg_offset = 0;
1914  else
1915  seg_offset = bytesleft + SizeOfXLogLongPHD;
1916  }
1917  else
1918  {
1919  /* account for the first page on segment with long header */
1920  seg_offset = XLOG_BLCKSZ;
1921  bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1922 
1923  fullpages = bytesleft / UsableBytesInPage;
1924  bytesleft = bytesleft % UsableBytesInPage;
1925 
1926  if (bytesleft == 0)
1927  seg_offset += fullpages * XLOG_BLCKSZ + bytesleft;
1928  else
1929  seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1930  }
1931 
1932  XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1933 
1934  return result;
1935 }
#define XLogSegNoOffsetToRecPtr(segno, offset, wal_segsz_bytes, dest)

References SizeOfXLogLongPHD, SizeOfXLogShortPHD, UsableBytesInPage, UsableBytesInSegment, wal_segment_size, and XLogSegNoOffsetToRecPtr.

Referenced by ReserveXLogInsertLocation(), ReserveXLogSwitch(), and WaitXLogInsertionsToFinish().

◆ XLogBytePosToRecPtr()

static XLogRecPtr XLogBytePosToRecPtr ( uint64  bytepos)
static

Definition at line 1858 of file xlog.c.

1859 {
1860  uint64 fullsegs;
1861  uint64 fullpages;
1862  uint64 bytesleft;
1863  uint32 seg_offset;
1864  XLogRecPtr result;
1865 
1866  fullsegs = bytepos / UsableBytesInSegment;
1867  bytesleft = bytepos % UsableBytesInSegment;
1868 
1869  if (bytesleft < XLOG_BLCKSZ - SizeOfXLogLongPHD)
1870  {
1871  /* fits on first page of segment */
1872  seg_offset = bytesleft + SizeOfXLogLongPHD;
1873  }
1874  else
1875  {
1876  /* account for the first page on segment with long header */
1877  seg_offset = XLOG_BLCKSZ;
1878  bytesleft -= XLOG_BLCKSZ - SizeOfXLogLongPHD;
1879 
1880  fullpages = bytesleft / UsableBytesInPage;
1881  bytesleft = bytesleft % UsableBytesInPage;
1882 
1883  seg_offset += fullpages * XLOG_BLCKSZ + bytesleft + SizeOfXLogShortPHD;
1884  }
1885 
1886  XLogSegNoOffsetToRecPtr(fullsegs, seg_offset, wal_segment_size, result);
1887 
1888  return result;
1889 }

References SizeOfXLogLongPHD, SizeOfXLogShortPHD, UsableBytesInPage, UsableBytesInSegment, wal_segment_size, and XLogSegNoOffsetToRecPtr.

Referenced by CreateCheckPoint(), GetXLogInsertRecPtr(), ReserveXLogInsertLocation(), and ReserveXLogSwitch().

◆ XLogCheckpointNeeded()

bool XLogCheckpointNeeded ( XLogSegNo  new_segno)

Definition at line 2290 of file xlog.c.

2291 {
2292  XLogSegNo old_segno;
2293 
2295 
2296  if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
2297  return true;
2298  return false;
2299 }

References CheckPointSegments, RedoRecPtr, wal_segment_size, and XLByteToSeg.

Referenced by XLogPageRead(), and XLogWrite().

◆ XLOGChooseNumBuffers()

static int XLOGChooseNumBuffers ( void  )
static

Definition at line 4621 of file xlog.c.

4622 {
4623  int xbuffers;
4624 
4625  xbuffers = NBuffers / 32;
4626  if (xbuffers > (wal_segment_size / XLOG_BLCKSZ))
4627  xbuffers = (wal_segment_size / XLOG_BLCKSZ);
4628  if (xbuffers < 8)
4629  xbuffers = 8;
4630  return xbuffers;
4631 }

References NBuffers, and wal_segment_size.

Referenced by check_wal_buffers(), and XLOGShmemSize().

◆ XLogFileClose()

static void XLogFileClose ( void  )
static

Definition at line 3639 of file xlog.c.

3640 {
3641  Assert(openLogFile >= 0);
3642 
3643  /*
3644  * WAL segment files will not be re-read in normal operation, so we advise
3645  * the OS to release any cached pages. But do not do so if WAL archiving
3646  * or streaming is active, because archiver and walsender process could
3647  * use the cache to read the WAL segment.
3648  */
3649 #if defined(USE_POSIX_FADVISE) && defined(POSIX_FADV_DONTNEED)
3650  if (!XLogIsNeeded() && (io_direct_flags & IO_DIRECT_WAL) == 0)
3651  (void) posix_fadvise(openLogFile, 0, 0, POSIX_FADV_DONTNEED);
3652 #endif
3653 
3654  if (close(openLogFile) != 0)
3655  {
3656  char xlogfname[MAXFNAMELEN];
3657  int save_errno = errno;
3658 
3660  errno = save_errno;
3661  ereport(PANIC,
3663  errmsg("could not close file \"%s\": %m", xlogfname)));
3664  }
3665 
3666  openLogFile = -1;
3668 }
void ReleaseExternalFD(void)
Definition: fd.c:1238

References Assert, close, ereport, errcode_for_file_access(), errmsg(), io_direct_flags, IO_DIRECT_WAL, MAXFNAMELEN, openLogFile, openLogSegNo, openLogTLI, PANIC, ReleaseExternalFD(), wal_segment_size, XLogFileName(), and XLogIsNeeded.

Referenced by assign_wal_sync_method(), XLogBackgroundFlush(), and XLogWrite().

◆ XLogFileCopy()

static void XLogFileCopy ( TimeLineID  destTLI,
XLogSegNo  destsegno,
TimeLineID  srcTLI,
XLogSegNo  srcsegno,
int  upto 
)
static

Definition at line 3418 of file xlog.c.

3421 {
3422  char path[MAXPGPATH];
3423  char tmppath[MAXPGPATH];
3424  PGAlignedXLogBlock buffer;
3425  int srcfd;
3426  int fd;
3427  int nbytes;
3428 
3429  /*
3430  * Open the source file
3431  */
3432  XLogFilePath(path, srcTLI, srcsegno, wal_segment_size);
3433  srcfd = OpenTransientFile(path, O_RDONLY | PG_BINARY);
3434  if (srcfd < 0)
3435  ereport(ERROR,
3437  errmsg("could not open file \"%s\": %m", path)));
3438 
3439  /*
3440  * Copy into a temp file name.
3441  */
3442  snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3443 
3444  unlink(tmppath);
3445 
3446  /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3447  fd = OpenTransientFile(tmppath, O_RDWR | O_CREAT | O_EXCL | PG_BINARY);
3448  if (fd < 0)
3449  ereport(ERROR,
3451  errmsg("could not create file \"%s\": %m", tmppath)));
3452 
3453  /*
3454  * Do the data copying.
3455  */
3456  for (nbytes = 0; nbytes < wal_segment_size; nbytes += sizeof(buffer))
3457  {
3458  int nread;
3459 
3460  nread = upto - nbytes;
3461 
3462  /*
3463  * The part that is not read from the source file is filled with
3464  * zeros.
3465  */
3466  if (nread < sizeof(buffer))
3467  memset(buffer.data, 0, sizeof(buffer));
3468 
3469  if (nread > 0)
3470  {
3471  int r;
3472 
3473  if (nread > sizeof(buffer))
3474  nread = sizeof(buffer);
3475  pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_READ);
3476  r = read(srcfd, buffer.data, nread);
3477  if (r != nread)
3478  {
3479  if (r < 0)
3480  ereport(ERROR,
3482  errmsg("could not read file \"%s\": %m",
3483  path)));
3484  else
3485  ereport(ERROR,
3487  errmsg("could not read file \"%s\": read %d of %zu",
3488  path, r, (Size) nread)));
3489  }
3491  }
3492  errno = 0;
3493  pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_WRITE);
3494  if ((int) write(fd, buffer.data, sizeof(buffer)) != (int) sizeof(buffer))
3495  {
3496  int save_errno = errno;
3497 
3498  /*
3499  * If we fail to make the file, delete it to release disk space
3500  */
3501  unlink(tmppath);
3502  /* if write didn't set errno, assume problem is no disk space */
3503  errno = save_errno ? save_errno : ENOSPC;
3504 
3505  ereport(ERROR,
3507  errmsg("could not write to file \"%s\": %m", tmppath)));
3508  }
3510  }
3511 
3512  pgstat_report_wait_start(WAIT_EVENT_WAL_COPY_SYNC);
3513  if (pg_fsync(fd) != 0)
3516  errmsg("could not fsync file \"%s\": %m", tmppath)));
3518 
3519  if (CloseTransientFile(fd) != 0)
3520  ereport(ERROR,
3522  errmsg("could not close file \"%s\": %m", tmppath)));
3523 
3524  if (CloseTransientFile(srcfd) != 0)
3525  ereport(ERROR,
3527  errmsg("could not close file \"%s\": %m", path)));
3528 
3529  /*
3530  * Now move the segment into place with its final name.
3531  */
3532  if (!InstallXLogFileSegment(&destsegno, tmppath, false, 0, destTLI))
3533  elog(ERROR, "InstallXLogFileSegment should not have failed");
3534 }
int CloseTransientFile(int fd)
Definition: fd.c:2831
int data_sync_elevel(int elevel)
Definition: fd.c:3959
int OpenTransientFile(const char *fileName, int fileFlags)
Definition: fd.c:2655
char data[XLOG_BLCKSZ]
Definition: c.h:1151

References CloseTransientFile(), PGAlignedXLogBlock::data, data_sync_elevel(), elog, ereport, errcode(), ERRCODE_DATA_CORRUPTED, errcode_for_file_access(), errmsg(), ERROR, fd(), InstallXLogFileSegment(), MAXPGPATH, OpenTransientFile(), PG_BINARY, pg_fsync(), pgstat_report_wait_end(), pgstat_report_wait_start(), read, snprintf, wal_segment_size, write, XLOGDIR, and XLogFilePath().

Referenced by XLogInitNewTimeline().

◆ XLogFileInit()

int XLogFileInit ( XLogSegNo  logsegno,
TimeLineID  logtli 
)

Definition at line 3380 of file xlog.c.

3381 {
3382  bool ignore_added;
3383  char path[MAXPGPATH];
3384  int fd;
3385 
3386  Assert(logtli != 0);
3387 
3388  fd = XLogFileInitInternal(logsegno, logtli, &ignore_added, path);
3389  if (fd >= 0)
3390  return fd;
3391 
3392  /* Now open original target segment (might not be file I just made) */
3393  fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3395  if (fd < 0)
3396  ereport(ERROR,
3398  errmsg("could not open file \"%s\": %m", path)));
3399  return fd;
3400 }
#define O_CLOEXEC
Definition: win32_port.h:359

References Assert, BasicOpenFile(), ereport, errcode_for_file_access(), errmsg(), ERROR, fd(), get_sync_bit(), MAXPGPATH, O_CLOEXEC, PG_BINARY, wal_sync_method, and XLogFileInitInternal().

Referenced by BootStrapXLOG(), XLogInitNewTimeline(), XLogWalRcvWrite(), and XLogWrite().

◆ XLogFileInitInternal()

static int XLogFileInitInternal ( XLogSegNo  logsegno,
TimeLineID  logtli,
bool added,
char *  path 
)
static

Definition at line 3210 of file xlog.c.

3212 {
3213  char tmppath[MAXPGPATH];
3214  XLogSegNo installed_segno;
3215  XLogSegNo max_segno;
3216  int fd;
3217  int save_errno;
3218  int open_flags = O_RDWR | O_CREAT | O_EXCL | PG_BINARY;
3219 
3220  Assert(logtli != 0);
3221 
3222  XLogFilePath(path, logtli, logsegno, wal_segment_size);
3223 
3224  /*
3225  * Try to use existent file (checkpoint maker may have created it already)
3226  */
3227  *added = false;
3228  fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3230  if (fd < 0)
3231  {
3232  if (errno != ENOENT)
3233  ereport(ERROR,
3235  errmsg("could not open file \"%s\": %m", path)));
3236  }
3237  else
3238  return fd;
3239 
3240  /*
3241  * Initialize an empty (all zeroes) segment. NOTE: it is possible that
3242  * another process is doing the same thing. If so, we will end up
3243  * pre-creating an extra log segment. That seems OK, and better than
3244  * holding the lock throughout this lengthy process.
3245  */
3246  elog(DEBUG2, "creating and filling new WAL file");
3247 
3248  snprintf(tmppath, MAXPGPATH, XLOGDIR "/xlogtemp.%d", (int) getpid());
3249 
3250  unlink(tmppath);
3251 
3253  open_flags |= PG_O_DIRECT;
3254 
3255  /* do not use get_sync_bit() here --- want to fsync only at end of fill */
3256  fd = BasicOpenFile(tmppath, open_flags);
3257  if (fd < 0)
3258  ereport(ERROR,
3260  errmsg("could not create file \"%s\": %m", tmppath)));
3261 
3262  pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_WRITE);
3263  save_errno = 0;
3264  if (wal_init_zero)
3265  {
3266  ssize_t rc;
3267 
3268  /*
3269  * Zero-fill the file. With this setting, we do this the hard way to
3270  * ensure that all the file space has really been allocated. On
3271  * platforms that allow "holes" in files, just seeking to the end
3272  * doesn't allocate intermediate space. This way, we know that we
3273  * have all the space and (after the fsync below) that all the
3274  * indirect blocks are down on disk. Therefore, fdatasync(2) or
3275  * O_DSYNC will be sufficient to sync future writes to the log file.
3276  */
3278 
3279  if (rc < 0)
3280  save_errno = errno;
3281  }
3282  else
3283  {
3284  /*
3285  * Otherwise, seeking to the end and writing a solitary byte is
3286  * enough.
3287  */
3288  errno = 0;
3289  if (pg_pwrite(fd, "\0", 1, wal_segment_size - 1) != 1)
3290  {
3291  /* if write didn't set errno, assume no disk space */
3292  save_errno = errno ? errno : ENOSPC;
3293  }
3294  }
3296 
3297  if (save_errno)
3298  {
3299  /*
3300  * If we fail to make the file, delete it to release disk space
3301  */
3302  unlink(tmppath);
3303 
3304  close(fd);
3305 
3306  errno = save_errno;
3307 
3308  ereport(ERROR,
3310  errmsg("could not write to file \"%s\": %m", tmppath)));
3311  }
3312 
3313  pgstat_report_wait_start(WAIT_EVENT_WAL_INIT_SYNC);
3314  if (pg_fsync(fd) != 0)
3315  {
3316  save_errno = errno;
3317  close(fd);
3318  errno = save_errno;
3319  ereport(ERROR,
3321  errmsg("could not fsync file \"%s\": %m", tmppath)));
3322  }
3324 
3325  if (close(fd) != 0)
3326  ereport(ERROR,
3328  errmsg("could not close file \"%s\": %m", tmppath)));
3329 
3330  /*
3331  * Now move the segment into place with its final name. Cope with
3332  * possibility that someone else has created the file while we were
3333  * filling ours: if so, use ours to pre-create a future log segment.
3334  */
3335  installed_segno = logsegno;
3336 
3337  /*
3338  * XXX: What should we use as max_segno? We used to use XLOGfileslop when
3339  * that was a constant, but that was always a bit dubious: normally, at a
3340  * checkpoint, XLOGfileslop was the offset from the checkpoint record, but
3341  * here, it was the offset from the insert location. We can't do the
3342  * normal XLOGfileslop calculation here because we don't have access to
3343  * the prior checkpoint's redo location. So somewhat arbitrarily, just use
3344  * CheckPointSegments.
3345  */
3346  max_segno = logsegno + CheckPointSegments;
3347  if (InstallXLogFileSegment(&installed_segno, tmppath, true, max_segno,
3348  logtli))
3349  {
3350  *added = true;
3351  elog(DEBUG2, "done creating and filling new WAL file");
3352  }
3353  else
3354  {
3355  /*
3356  * No need for any more future segments, or InstallXLogFileSegment()
3357  * failed to rename the file into place. If the rename failed, a
3358  * caller opening the file may fail.
3359  */
3360  unlink(tmppath);
3361  elog(DEBUG2, "abandoned new WAL file");
3362  }
3363 
3364  return -1;
3365 }
#define IO_DIRECT_WAL_INIT
Definition: fd.h:56
ssize_t pg_pwrite_zeros(int fd, size_t size, off_t offset)
Definition: file_utils.c:688
#define pg_pwrite
Definition: port.h:226
bool wal_init_zero
Definition: xlog.c:127

References Assert, BasicOpenFile(), CheckPointSegments, close, DEBUG2, elog, ereport, errcode_for_file_access(), errmsg(), ERROR, fd(), get_sync_bit(), InstallXLogFileSegment(), io_direct_flags, IO_DIRECT_WAL_INIT, MAXPGPATH, O_CLOEXEC, PG_BINARY, pg_fsync(), PG_O_DIRECT, pg_pwrite, pg_pwrite_zeros(), pgstat_report_wait_end(), pgstat_report_wait_start(), snprintf, wal_init_zero, wal_segment_size, wal_sync_method, XLOGDIR, and XLogFilePath().

Referenced by PreallocXlogFiles(), and XLogFileInit().

◆ XLogFileOpen()

int XLogFileOpen ( XLogSegNo  segno,
TimeLineID  tli 
)

Definition at line 3618 of file xlog.c.

3619 {
3620  char path[MAXPGPATH];
3621  int fd;
3622 
3623  XLogFilePath(path, tli, segno, wal_segment_size);
3624 
3625  fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3627  if (fd < 0)
3628  ereport(PANIC,
3630  errmsg("could not open file \"%s\": %m", path)));
3631 
3632  return fd;
3633 }

References BasicOpenFile(), ereport, errcode_for_file_access(), errmsg(), fd(), get_sync_bit(), MAXPGPATH, O_CLOEXEC, PANIC, PG_BINARY, wal_segment_size, wal_sync_method, and XLogFilePath().

Referenced by XLogWrite().

◆ XLOGfileslop()

static XLogSegNo XLOGfileslop ( XLogRecPtr  lastredoptr)
static

Definition at line 2240 of file xlog.c.

2241 {
2242  XLogSegNo minSegNo;
2243  XLogSegNo maxSegNo;
2244  double distance;
2245  XLogSegNo recycleSegNo;
2246 
2247  /*
2248  * Calculate the segment numbers that min_wal_size_mb and max_wal_size_mb
2249  * correspond to. Always recycle enough segments to meet the minimum, and
2250  * remove enough segments to stay below the maximum.
2251  */
2252  minSegNo = lastredoptr / wal_segment_size +
2254  maxSegNo = lastredoptr / wal_segment_size +
2256 
2257  /*
2258  * Between those limits, recycle enough segments to get us through to the
2259  * estimated end of next checkpoint.
2260  *
2261  * To estimate where the next checkpoint will finish, assume that the
2262  * system runs steadily consuming CheckPointDistanceEstimate bytes between
2263  * every checkpoint.
2264  */
2266  /* add 10% for good measure. */
2267  distance *= 1.10;
2268 
2269  recycleSegNo = (XLogSegNo) ceil(((double) lastredoptr + distance) /
2271 
2272  if (recycleSegNo < minSegNo)
2273  recycleSegNo = minSegNo;
2274  if (recycleSegNo > maxSegNo)
2275  recycleSegNo = maxSegNo;
2276 
2277  return recycleSegNo;
2278 }

References CheckPointCompletionTarget, CheckPointDistanceEstimate, ConvertToXSegs, max_wal_size_mb, min_wal_size_mb, and wal_segment_size.

Referenced by RemoveOldXlogFiles().

◆ XLogFlush()

void XLogFlush ( XLogRecPtr  record)

Definition at line 2802 of file xlog.c.

2803 {
2804  XLogRecPtr WriteRqstPtr;
2805  XLogwrtRqst WriteRqst;
2806  TimeLineID insertTLI = XLogCtl->InsertTimeLineID;
2807 
2808  /*
2809  * During REDO, we are reading not writing WAL. Therefore, instead of
2810  * trying to flush the WAL, we should update minRecoveryPoint instead. We
2811  * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2812  * to act this way too, and because when it tries to write the
2813  * end-of-recovery checkpoint, it should indeed flush.
2814  */
2815  if (!XLogInsertAllowed())
2816  {
2817  UpdateMinRecoveryPoint(record, false);
2818  return;
2819  }
2820 
2821  /* Quick exit if already known flushed */
2822  if (record <= LogwrtResult.Flush)
2823  return;
2824 
2825 #ifdef WAL_DEBUG
2826  if (XLOG_DEBUG)
2827  elog(LOG, "xlog flush request %X/%X; write %X/%X; flush %X/%X",
2828  LSN_FORMAT_ARGS(record),
2831 #endif
2832 
2834 
2835  /*
2836  * Since fsync is usually a horribly expensive operation, we try to
2837  * piggyback as much data as we can on each fsync: if we see any more data
2838  * entered into the xlog buffer, we'll write and fsync that too, so that
2839  * the final value of LogwrtResult.Flush is as large as possible. This
2840  * gives us some chance of avoiding another fsync immediately after.
2841  */
2842 
2843  /* initialize to given target; may increase below */
2844  WriteRqstPtr = record;
2845 
2846  /*
2847  * Now wait until we get the write lock, or someone else does the flush
2848  * for us.
2849  */
2850  for (;;)
2851  {
2852  XLogRecPtr insertpos;
2853 
2854  /* done already? */
2856  if (record <= LogwrtResult.Flush)
2857  break;
2858 
2859  /*
2860  * Before actually performing the write, wait for all in-flight
2861  * insertions to the pages we're about to write to finish.
2862  */
2864  if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2865  WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2867  insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2868 
2869  /*
2870  * Try to get the write lock. If we can't get it immediately, wait
2871  * until it's released, and recheck if we still need to do the flush
2872  * or if the backend that held the lock did it for us already. This
2873  * helps to maintain a good rate of group committing when the system
2874  * is bottlenecked by the speed of fsyncing.
2875  */
2876  if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2877  {
2878  /*
2879  * The lock is now free, but we didn't acquire it yet. Before we
2880  * do, loop back to check if someone else flushed the record for
2881  * us already.
2882  */
2883  continue;
2884  }
2885 
2886  /* Got the lock; recheck whether request is satisfied */
2888  if (record <= LogwrtResult.Flush)
2889  {
2890  LWLockRelease(WALWriteLock);
2891  break;
2892  }
2893 
2894  /*
2895  * Sleep before flush! By adding a delay here, we may give further
2896  * backends the opportunity to join the backlog of group commit
2897  * followers; this can significantly improve transaction throughput,
2898  * at the risk of increasing transaction latency.
2899  *
2900  * We do not sleep if enableFsync is not turned on, nor if there are
2901  * fewer than CommitSiblings other backends with active transactions.
2902  */
2903  if (CommitDelay > 0 && enableFsync &&
2905  {
2907 
2908  /*
2909  * Re-check how far we can now flush the WAL. It's generally not
2910  * safe to call WaitXLogInsertionsToFinish while holding
2911  * WALWriteLock, because an in-progress insertion might need to
2912  * also grab WALWriteLock to make progress. But we know that all
2913  * the insertions up to insertpos have already finished, because
2914  * that's what the earlier WaitXLogInsertionsToFinish() returned.
2915  * We're only calling it again to allow insertpos to be moved
2916  * further forward, not to actually wait for anyone.
2917  */
2918  insertpos = WaitXLogInsertionsToFinish(insertpos);
2919  }
2920 
2921  /* try to write/flush later additions to XLOG as well */
2922  WriteRqst.Write = insertpos;
2923  WriteRqst.Flush = insertpos;
2924 
2925  XLogWrite(WriteRqst, insertTLI, false);
2926 
2927  LWLockRelease(WALWriteLock);
2928  /* done */
2929  break;
2930  }
2931 
2932  END_CRIT_SECTION();
2933 
2934  /* wake up walsenders now that we've released heavily contended locks */
2936 
2937  /*
2938  * If we still haven't flushed to the request point then we have a
2939  * problem; most likely, the requested flush point is past end of XLOG.
2940  * This has been seen to occur when a disk page has a corrupted LSN.
2941  *
2942  * Formerly we treated this as a PANIC condition, but that hurts the
2943  * system's robustness rather than helping it: we do not want to take down
2944  * the whole system due to corruption on one data page. In particular, if
2945  * the bad page is encountered again during recovery then we would be
2946  * unable to restart the database at all! (This scenario actually
2947  * happened in the field several times with 7.1 releases.) As of 8.4, bad
2948  * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2949  * the only time we can reach here during recovery is while flushing the
2950  * end-of-recovery checkpoint record, and we don't expect that to have a
2951  * bad LSN.
2952  *
2953  * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2954  * since xact.c calls this routine inside a critical section. However,
2955  * calls from bufmgr.c are not within critical sections and so we will not
2956  * force a restart for a bad LSN on a data page.
2957  */
2958  if (LogwrtResult.Flush < record)
2959  elog(ERROR,
2960  "xlog flush request %X/%X is not satisfied --- flushed only to %X/%X",
2961  LSN_FORMAT_ARGS(record),
2963 }
bool LWLockAcquireOrWait(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1396
bool MinimumActiveBackends(int min)
Definition: procarray.c:3550
int CommitDelay
Definition: xlog.c:132
int CommitSiblings
Definition: xlog.c:133
bool XLogInsertAllowed(void)
Definition: xlog.c:6389

References CommitDelay, CommitSiblings, elog, enableFsync, END_CRIT_SECTION, ERROR, XLogwrtRqst::Flush, XLogwrtResult::Flush, XLogCtlData::info_lck, XLogCtlData::InsertTimeLineID, LOG, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, LW_EXCLUSIVE, LWLockAcquireOrWait(), LWLockRelease(), MinimumActiveBackends(), pg_usleep(), RecoveryInProgress(), RefreshXLogWriteResult, SpinLockAcquire, SpinLockRelease, START_CRIT_SECTION, UpdateMinRecoveryPoint(), WaitXLogInsertionsToFinish(), WalSndWakeupProcessRequests(), XLogwrtRqst::Write, XLogwrtResult::Write, XLogCtl, XLogInsertAllowed(), and XLogWrite().

Referenced by CheckPointReplicationOrigin(), CreateCheckPoint(), CreateEndOfRecoveryRecord(), CreateOverwriteContrecordRecord(), dropdb(), EndPrepare(), FlushBuffer(), LogLogicalMessage(), pg_attribute_noreturn(), RecordTransactionAbortPrepared(), RecordTransactionCommit(), RecordTransactionCommitPrepared(), RelationTruncate(), ReplicationSlotReserveWal(), replorigin_get_progress(), replorigin_session_get_progress(), SlruPhysicalWritePage(), smgr_redo(), write_relmap_file(), WriteMTruncateXlogRec(), WriteTruncateXlogRec(), xact_redo_abort(), xact_redo_commit(), XLogInsertRecord(), and XLogReportParameters().

◆ XLogGetLastRemovedSegno()

XLogSegNo XLogGetLastRemovedSegno ( void  )

Definition at line 3758 of file xlog.c.

3759 {
3760  XLogSegNo lastRemovedSegNo;
3761 
3763  lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3765 
3766  return lastRemovedSegNo;
3767 }

References XLogCtlData::info_lck, XLogCtlData::lastRemovedSegNo, SpinLockAcquire, SpinLockRelease, and XLogCtl.

Referenced by copy_replication_slot(), GetWALAvailability(), ReplicationSlotReserveWal(), and reserve_wal_for_local_slot().

◆ XLogGetOldestSegno()

XLogSegNo XLogGetOldestSegno ( TimeLineID  tli)

Definition at line 3774 of file xlog.c.

3775 {
3776  DIR *xldir;
3777  struct dirent *xlde;
3778  XLogSegNo oldest_segno = 0;
3779 
3780  xldir = AllocateDir(XLOGDIR);
3781  while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3782  {
3783  TimeLineID file_tli;
3784  XLogSegNo file_segno;
3785 
3786  /* Ignore files that are not XLOG segments. */
3787  if (!IsXLogFileName(xlde->d_name))
3788  continue;
3789 
3790  /* Parse filename to get TLI and segno. */
3791  XLogFromFileName(xlde->d_name, &file_tli, &file_segno,
3793 
3794  /* Ignore anything that's not from the TLI of interest. */
3795  if (tli != file_tli)
3796  continue;
3797 
3798  /* If it's the oldest so far, update oldest_segno. */
3799  if (oldest_segno == 0 || file_segno < oldest_segno)
3800  oldest_segno = file_segno;
3801  }
3802 
3803  FreeDir(xldir);
3804  return oldest_segno;
3805 }

References AllocateDir(), dirent::d_name, FreeDir(), IsXLogFileName(), ReadDir(), wal_segment_size, XLOGDIR, and XLogFromFileName().

Referenced by GetOldestUnsummarizedLSN(), MaybeRemoveOldWalSummaries(), and reserve_wal_for_local_slot().

◆ XLogGetReplicationSlotMinimumLSN()

static XLogRecPtr XLogGetReplicationSlotMinimumLSN ( void  )
static

Definition at line 2701 of file xlog.c.

2702 {
2703  XLogRecPtr retval;
2704 
2706  retval = XLogCtl->replicationSlotMinLSN;
2708 
2709  return retval;
2710 }
XLogRecPtr replicationSlotMinLSN
Definition: xlog.c:459

References XLogCtlData::info_lck, XLogCtlData::replicationSlotMinLSN, SpinLockAcquire, SpinLockRelease, and XLogCtl.

Referenced by KeepLogSeg().

◆ XLogInitNewTimeline()

static void XLogInitNewTimeline ( TimeLineID  endTLI,
XLogRecPtr  endOfLog,
TimeLineID  newTLI 
)
static

Definition at line 5214 of file xlog.c.

5215 {
5216  char xlogfname[MAXFNAMELEN];
5217  XLogSegNo endLogSegNo;
5218  XLogSegNo startLogSegNo;
5219 
5220  /* we always switch to a new timeline after archive recovery */
5221  Assert(endTLI != newTLI);
5222 
5223  /*
5224  * Update min recovery point one last time.
5225  */
5227 
5228  /*
5229  * Calculate the last segment on the old timeline, and the first segment
5230  * on the new timeline. If the switch happens in the middle of a segment,
5231  * they are the same, but if the switch happens exactly at a segment
5232  * boundary, startLogSegNo will be endLogSegNo + 1.
5233  */
5234  XLByteToPrevSeg(endOfLog, endLogSegNo, wal_segment_size);
5235  XLByteToSeg(endOfLog, startLogSegNo, wal_segment_size);
5236 
5237  /*
5238  * Initialize the starting WAL segment for the new timeline. If the switch
5239  * happens in the middle of a segment, copy data from the last WAL segment
5240  * of the old timeline up to the switch point, to the starting WAL segment
5241  * on the new timeline.
5242  */
5243  if (endLogSegNo == startLogSegNo)
5244  {
5245  /*
5246  * Make a copy of the file on the new timeline.
5247  *
5248  * Writing WAL isn't allowed yet, so there are no locking
5249  * considerations. But we should be just as tense as XLogFileInit to
5250  * avoid emplacing a bogus file.
5251  */
5252  XLogFileCopy(newTLI, endLogSegNo, endTLI, endLogSegNo,
5253  XLogSegmentOffset(endOfLog, wal_segment_size));
5254  }
5255  else
5256  {
5257  /*
5258  * The switch happened at a segment boundary, so just create the next
5259  * segment on the new timeline.
5260  */
5261  int fd;
5262 
5263  fd = XLogFileInit(startLogSegNo, newTLI);
5264 
5265  if (close(fd) != 0)
5266  {
5267  int save_errno = errno;
5268 
5269  XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
5270  errno = save_errno;
5271  ereport(ERROR,
5273  errmsg("could not close file \"%s\": %m", xlogfname)));
5274  }
5275  }
5276 
5277  /*
5278  * Let's just make real sure there are not .ready or .done flags posted
5279  * for the new segment.
5280  */
5281  XLogFileName(xlogfname, newTLI, startLogSegNo, wal_segment_size);
5282  XLogArchiveCleanup(xlogfname);
5283 }
static void XLogFileCopy(TimeLineID destTLI, XLogSegNo destsegno, TimeLineID srcTLI, XLogSegNo srcsegno, int upto)
Definition: xlog.c:3418

References Assert, close, ereport, errcode_for_file_access(), errmsg(), ERROR, fd(), InvalidXLogRecPtr, MAXFNAMELEN, UpdateMinRecoveryPoint(), wal_segment_size, XLByteToPrevSeg, XLByteToSeg, XLogArchiveCleanup(), XLogFileCopy(), XLogFileInit(), XLogFileName(), and XLogSegmentOffset.

Referenced by StartupXLOG().

◆ XLogInsertAllowed()

bool XLogInsertAllowed ( void  )

Definition at line 6389 of file xlog.c.

6390 {
6391  /*
6392  * If value is "unconditionally true" or "unconditionally false", just
6393  * return it. This provides the normal fast path once recovery is known
6394  * done.
6395  */
6396  if (LocalXLogInsertAllowed >= 0)
6397  return (bool) LocalXLogInsertAllowed;
6398 
6399  /*
6400  * Else, must check to see if we're still in recovery.
6401  */
6402  if (RecoveryInProgress())
6403  return false;
6404 
6405  /*
6406  * On exit from recovery, reset to "unconditionally true", since there is
6407  * no need to keep checking.
6408  */
6410  return true;
6411 }

References LocalXLogInsertAllowed, and RecoveryInProgress().

Referenced by XLogBeginInsert(), XLogFlush(), and XLogInsertRecord().

◆ XLogInsertRecord()

XLogRecPtr XLogInsertRecord ( XLogRecData rdata,
XLogRecPtr  fpw_lsn,
uint8  flags,
int  num_fpi,
bool  topxid_included 
)

Definition at line 748 of file xlog.c.

753 {
755  pg_crc32c rdata_crc;
756  bool inserted;
757  XLogRecord *rechdr = (XLogRecord *) rdata->data;
758  uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
760  XLogRecPtr StartPos;
761  XLogRecPtr EndPos;
762  bool prevDoPageWrites = doPageWrites;
763  TimeLineID insertTLI;
764 
765  /* Does this record type require special handling? */
766  if (unlikely(rechdr->xl_rmid == RM_XLOG_ID))
767  {
768  if (info == XLOG_SWITCH)
769  class = WALINSERT_SPECIAL_SWITCH;
770  else if (info == XLOG_CHECKPOINT_REDO)
772  }
773 
774  /* we assume that all of the record header is in the first chunk */
775  Assert(rdata->len >= SizeOfXLogRecord);
776 
777  /* cross-check on whether we should be here or not */
778  if (!XLogInsertAllowed())
779  elog(ERROR, "cannot make new WAL entries during recovery");
780 
781  /*
782  * Given that we're not in recovery, InsertTimeLineID is set and can't
783  * change, so we can read it without a lock.
784  */
785  insertTLI = XLogCtl->InsertTimeLineID;
786 
787  /*----------
788  *
789  * We have now done all the preparatory work we can without holding a
790  * lock or modifying shared state. From here on, inserting the new WAL
791  * record to the shared WAL buffer cache is a two-step process:
792  *
793  * 1. Reserve the right amount of space from the WAL. The current head of
794  * reserved space is kept in Insert->CurrBytePos, and is protected by
795  * insertpos_lck.
796  *
797  * 2. Copy the record to the reserved WAL space. This involves finding the
798  * correct WAL buffer containing the reserved space, and copying the
799  * record in place. This can be done concurrently in multiple processes.
800  *
801  * To keep track of which insertions are still in-progress, each concurrent
802  * inserter acquires an insertion lock. In addition to just indicating that
803  * an insertion is in progress, the lock tells others how far the inserter
804  * has progressed. There is a small fixed number of insertion locks,
805  * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
806  * boundary, it updates the value stored in the lock to the how far it has
807  * inserted, to allow the previous buffer to be flushed.
808  *
809  * Holding onto an insertion lock also protects RedoRecPtr and
810  * fullPageWrites from changing until the insertion is finished.
811  *
812  * Step 2 can usually be done completely in parallel. If the required WAL
813  * page is not initialized yet, you have to grab WALBufMappingLock to
814  * initialize it, but the WAL writer tries to do that ahead of insertions
815  * to avoid that from happening in the critical path.
816  *
817  *----------
818  */
820 
821  if (likely(class == WALINSERT_NORMAL))
822  {
824 
825  /*
826  * Check to see if my copy of RedoRecPtr is out of date. If so, may
827  * have to go back and have the caller recompute everything. This can
828  * only happen just after a checkpoint, so it's better to be slow in
829  * this case and fast otherwise.
830  *
831  * Also check to see if fullPageWrites was just turned on or there's a
832  * running backup (which forces full-page writes); if we weren't
833  * already doing full-page writes then go back and recompute.
834  *
835  * If we aren't doing full-page writes then RedoRecPtr doesn't
836  * actually affect the contents of the XLOG record, so we'll update
837  * our local copy but not force a recomputation. (If doPageWrites was
838  * just turned off, we could recompute the record without full pages,
839  * but we choose not to bother.)
840  */
841  if (RedoRecPtr != Insert->RedoRecPtr)
842  {
843  Assert(RedoRecPtr < Insert->RedoRecPtr);
844  RedoRecPtr = Insert->RedoRecPtr;
845  }
846  doPageWrites = (Insert->fullPageWrites || Insert->runningBackups > 0);
847 
848  if (doPageWrites &&
849  (!prevDoPageWrites ||
850  (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr)))
851  {
852  /*
853  * Oops, some buffer now needs to be backed up that the caller
854  * didn't back up. Start over.
855  */
858  return InvalidXLogRecPtr;
859  }
860 
861  /*
862  * Reserve space for the record in the WAL. This also sets the xl_prev
863  * pointer.
864  */
865  ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
866  &rechdr->xl_prev);
867 
868  /* Normal records are always inserted. */
869  inserted = true;
870  }
871  else if (class == WALINSERT_SPECIAL_SWITCH)
872  {
873  /*
874  * In order to insert an XLOG_SWITCH record, we need to hold all of
875  * the WAL insertion locks, not just one, so that no one else can
876  * begin inserting a record until we've figured out how much space
877  * remains in the current WAL segment and claimed all of it.
878  *
879  * Nonetheless, this case is simpler than the normal cases handled
880  * below, which must check for changes in doPageWrites and RedoRecPtr.
881  * Those checks are only needed for records that can contain buffer
882  * references, and an XLOG_SWITCH record never does.
883  */
884  Assert(fpw_lsn == InvalidXLogRecPtr);
886  inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
887  }
888  else
889  {
891 
892  /*
893  * We need to update both the local and shared copies of RedoRecPtr,
894  * which means that we need to hold all the WAL insertion locks.
895  * However, there can't be any buffer references, so as above, we need
896  * not check RedoRecPtr before inserting the record; we just need to
897  * update it afterwards.
898  */
899  Assert(fpw_lsn == InvalidXLogRecPtr);
901  ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
902  &rechdr->xl_prev);
903  RedoRecPtr = Insert->RedoRecPtr = StartPos;
904  inserted = true;
905  }
906 
907  if (inserted)
908  {
909  /*
910  * Now that xl_prev has been filled in, calculate CRC of the record
911  * header.
912  */
913  rdata_crc = rechdr->xl_crc;
914  COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
915  FIN_CRC32C(rdata_crc);
916  rechdr->xl_crc = rdata_crc;
917 
918  /*
919  * All the record data, including the header, is now ready to be
920  * inserted. Copy the record in the space reserved.
921  */
923  class == WALINSERT_SPECIAL_SWITCH, rdata,
924  StartPos, EndPos, insertTLI);
925 
926  /*
927  * Unless record is flagged as not important, update LSN of last
928  * important record in the current slot. When holding all locks, just
929  * update the first one.
930  */
931  if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
932  {
933  int lockno = holdingAllLocks ? 0 : MyLockNo;
934 
935  WALInsertLocks[lockno].l.lastImportantAt = StartPos;
936  }
937  }
938  else
939  {
940  /*
941  * This was an xlog-switch record, but the current insert location was
942  * already exactly at the beginning of a segment, so there was no need
943  * to do anything.
944  */
945  }
946 
947  /*
948  * Done! Let others know that we're finished.
949  */
951 
953 
955 
956  /*
957  * Mark top transaction id is logged (if needed) so that we should not try
958  * to log it again with the next WAL record in the current subtransaction.
959  */
960  if (topxid_included)
962 
963  /*
964  * Update shared LogwrtRqst.Write, if we crossed page boundary.
965  */
966  if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
967  {
969  /* advance global request to include new block(s) */
970  if (XLogCtl->LogwrtRqst.Write < EndPos)
971  XLogCtl->LogwrtRqst.Write = EndPos;
974  }
975 
976  /*
977  * If this was an XLOG_SWITCH record, flush the record and the empty
978  * padding space that fills the rest of the segment, and perform
979  * end-of-segment actions (eg, notifying archiver).
980  */
981  if (class == WALINSERT_SPECIAL_SWITCH)
982  {
983  TRACE_POSTGRESQL_WAL_SWITCH();
984  XLogFlush(EndPos);
985 
986  /*
987  * Even though we reserved the rest of the segment for us, which is
988  * reflected in EndPos, we return a pointer to just the end of the
989  * xlog-switch record.
990  */
991  if (inserted)
992  {
993  EndPos = StartPos + SizeOfXLogRecord;
994  if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
995  {
996  uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
997 
998  if (offset == EndPos % XLOG_BLCKSZ)
999  EndPos += SizeOfXLogLongPHD;
1000  else
1001  EndPos += SizeOfXLogShortPHD;
1002  }
1003  }
1004  }
1005 
1006 #ifdef WAL_DEBUG
1007  if (XLOG_DEBUG)
1008  {
1009  static XLogReaderState *debug_reader = NULL;
1010  XLogRecord *record;
1011  DecodedXLogRecord *decoded;
1013  StringInfoData recordBuf;
1014  char *errormsg = NULL;
1015  MemoryContext oldCxt;
1016 
1017  oldCxt = MemoryContextSwitchTo(walDebugCxt);
1018 
1019  initStringInfo(&buf);
1020  appendStringInfo(&buf, "INSERT @ %X/%X: ", LSN_FORMAT_ARGS(EndPos));
1021 
1022  /*
1023  * We have to piece together the WAL record data from the XLogRecData
1024  * entries, so that we can pass it to the rm_desc function as one
1025  * contiguous chunk.
1026  */
1027  initStringInfo(&recordBuf);
1028  for (; rdata != NULL; rdata = rdata->next)
1029  appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1030 
1031  /* We also need temporary space to decode the record. */
1032  record = (XLogRecord *) recordBuf.data;
1033  decoded = (DecodedXLogRecord *)
1035 
1036  if (!debug_reader)
1037  debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1038  XL_ROUTINE(.page_read = NULL,
1039  .segment_open = NULL,
1040  .segment_close = NULL),
1041  NULL);
1042  if (!debug_reader)
1043  {
1044  appendStringInfoString(&buf, "error decoding record: out of memory while allocating a WAL reading processor");
1045  }
1046  else if (!DecodeXLogRecord(debug_reader,
1047  decoded,
1048  record,
1049  EndPos,
1050  &errormsg))
1051  {
1052  appendStringInfo(&buf, "error decoding record: %s",
1053  errormsg ? errormsg : "no error message");
1054  }
1055  else
1056  {
1057  appendStringInfoString(&buf, " - ");
1058 
1059  debug_reader->record = decoded;
1060  xlog_outdesc(&buf, debug_reader);
1061  debug_reader->record = NULL;
1062  }
1063  elog(LOG, "%s", buf.data);
1064 
1065  pfree(decoded);
1066  pfree(buf.data);
1067  pfree(recordBuf.data);
1068  MemoryContextSwitchTo(oldCxt);
1069  }
1070 #endif
1071 
1072  /*
1073  * Update our global variables
1074  */
1075  ProcLastRecPtr = StartPos;
1076  XactLastRecEnd = EndPos;
1077 
1078  /* Report WAL traffic to the instrumentation. */
1079  if (inserted)
1080  {
1081  pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1083  pgWalUsage.wal_fpi += num_fpi;
1084  }
1085 
1086  return EndPos;
1087 }
#define likely(x)
Definition: c.h:325
#define unlikely(x)
Definition: c.h:326
WalUsage pgWalUsage
Definition: instrument.c:22
if(TABLE==NULL||TABLE_index==NULL)
Definition: isn.c:76
MemoryContextSwitchTo(old_ctx)
void appendBinaryStringInfo(StringInfo str, const void *data, int datalen)
Definition: stringinfo.c:230
void appendStringInfoString(StringInfo str, const char *s)
Definition: stringinfo.c:179
uint64 wal_bytes
Definition: instrument.h:55
int64 wal_fpi
Definition: instrument.h:54
int64 wal_records
Definition: instrument.h:53
DecodedXLogRecord * record
Definition: xlogreader.h:236
pg_crc32c xl_crc
Definition: xlogrecord.h:49
void MarkSubxactTopXidLogged(void)
Definition: xact.c:590
void MarkCurrentTransactionIdLoggedIfAny(void)
Definition: xact.c:540
XLogRecPtr XactLastRecEnd
Definition: xlog.c:254
static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata, XLogRecPtr StartPos, XLogRecPtr EndPos, TimeLineID tli)
Definition: xlog.c:1225
static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
Definition: xlog.c:1108
static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
Definition: xlog.c:1164
XLogReaderState * XLogReaderAllocate(int wal_segment_size, const char *waldir, XLogReaderRoutine *routine, void *private_data)
Definition: xlogreader.c:106
bool DecodeXLogRecord(XLogReaderState *state, DecodedXLogRecord *decoded, XLogRecord *record, XLogRecPtr lsn, char **errormsg)
Definition: xlogreader.c:1662
size_t DecodeXLogRecordRequiredSpace(size_t xl_tot_len)
Definition: xlogreader.c:1629
#define XL_ROUTINE(...)
Definition: xlogreader.h:117
void xlog_outdesc(StringInfo buf, XLogReaderState *record)

References appendBinaryStringInfo(), appendStringInfo(), appendStringInfoString(), Assert, buf, COMP_CRC32C, CopyXLogRecordToWAL(), XLogRecData::data, StringInfoData::data, DecodeXLogRecord(), DecodeXLogRecordRequiredSpace(), doPageWrites, elog, END_CRIT_SECTION, ERROR, FIN_CRC32C, holdingAllLocks, if(), XLogCtlData::info_lck, initStringInfo(), XLogCtlData::Insert, Insert(), XLogCtlData::InsertTimeLineID, InvalidXLogRecPtr, WALInsertLockPadded::l, WALInsertLock::lastImportantAt, XLogRecData::len, likely, LOG, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, MarkCurrentTransactionIdLoggedIfAny(), MarkSubxactTopXidLogged(), MemoryContextSwitchTo(), MyLockNo, XLogRecData::next, palloc(), pfree(), pgWalUsage, ProcLastRecPtr, XLogReaderState::record, RedoRecPtr, RefreshXLogWriteResult, ReserveXLogInsertLocation(), ReserveXLogSwitch(), SizeOfXLogLongPHD, SizeOfXLogRecord, SizeOfXLogShortPHD, SpinLockAcquire, SpinLockRelease, START_CRIT_SECTION, unlikely, WalUsage::wal_bytes, WalUsage::wal_fpi, WalUsage::wal_records, wal_segment_size, WALINSERT_NORMAL, WALINSERT_SPECIAL_CHECKPOINT, WALINSERT_SPECIAL_SWITCH, WALInsertLockAcquire(), WALInsertLockAcquireExclusive(), WALInsertLockRelease(), WALInsertLocks, XLogwrtRqst::Write, XactLastRecEnd, XLogRecord::xl_crc, XLogRecord::xl_info, XLogRecord::xl_prev, XLogRecord::xl_rmid, XL_ROUTINE, XLogRecord::xl_tot_len, XLOG_CHECKPOINT_REDO, XLOG_MARK_UNIMPORTANT, xlog_outdesc(), XLOG_SWITCH, XLogCtl, XLogFlush(), XLogInsertAllowed(), XLogReaderAllocate(), XLogSegmentOffset, and XLR_INFO_MASK.

Referenced by XLogInsert().

◆ XLogNeedsFlush()

bool XLogNeedsFlush ( XLogRecPtr  record)

Definition at line 3133 of file xlog.c.

3134 {
3135  /*
3136  * During recovery, we don't flush WAL but update minRecoveryPoint
3137  * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3138  * would need to be updated.
3139  */
3140  if (RecoveryInProgress())
3141  {
3142  /*
3143  * An invalid minRecoveryPoint means that we need to recover all the
3144  * WAL, i.e., we're doing crash recovery. We never modify the control
3145  * file's value in that case, so we can short-circuit future checks
3146  * here too. This triggers a quick exit path for the startup process,
3147  * which cannot update its local copy of minRecoveryPoint as long as
3148  * it has not replayed all WAL available when doing crash recovery.
3149  */
3151  updateMinRecoveryPoint = false;
3152 
3153  /* Quick exit if already known to be updated or cannot be updated */
3155  return false;
3156 
3157  /*
3158  * Update local copy of minRecoveryPoint. But if the lock is busy,
3159  * just return a conservative guess.
3160  */
3161  if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
3162  return true;
3165  LWLockRelease(ControlFileLock);
3166 
3167  /*
3168  * Check minRecoveryPoint for any other process than the startup
3169  * process doing crash recovery, which should not update the control
3170  * file value if crash recovery is still running.
3171  */
3173  updateMinRecoveryPoint = false;
3174 
3175  /* check again */
3177  return false;
3178  else
3179  return true;
3180  }
3181 
3182  /* Quick exit if already known flushed */
3183  if (record <= LogwrtResult.Flush)
3184  return false;
3185 
3186  /* read LogwrtResult and update local state */
3188 
3189  /* check again */
3190  if (record <= LogwrtResult.Flush)
3191  return false;
3192 
3193  return true;
3194 }
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1339

References ControlFile, XLogwrtResult::Flush, InRecovery, LocalMinRecoveryPoint, LocalMinRecoveryPointTLI, LogwrtResult, LW_SHARED, LWLockConditionalAcquire(), LWLockRelease(), ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, RecoveryInProgress(), RefreshXLogWriteResult, updateMinRecoveryPoint, and XLogRecPtrIsInvalid.

Referenced by GetVictimBuffer(), and SetHintBits().

◆ XLogPutNextOid()

void XLogPutNextOid ( Oid  nextOid)

Definition at line 8023 of file xlog.c.

8024 {
8025  XLogBeginInsert();
8026  XLogRegisterData((char *) (&nextOid), sizeof(Oid));
8027  (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
8028 
8029  /*
8030  * We need not flush the NEXTOID record immediately, because any of the
8031  * just-allocated OIDs could only reach disk as part of a tuple insert or
8032  * update that would have its own XLOG record that must follow the NEXTOID
8033  * record. Therefore, the standard buffer LSN interlock applied to those
8034  * records will ensure no such OID reaches disk before the NEXTOID record
8035  * does.
8036  *
8037  * Note, however, that the above statement only covers state "within" the
8038  * database. When we use a generated OID as a file or directory name, we
8039  * are in a sense violating the basic WAL rule, because that filesystem
8040  * change may reach disk before the NEXTOID WAL record does. The impact
8041  * of this is that if a database crash occurs immediately afterward, we
8042  * might after restart re-generate the same OID and find that it conflicts
8043  * with the leftover file or directory. But since for safety's sake we
8044  * always loop until finding a nonconflicting filename, this poses no real
8045  * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
8046  */
8047 }

References XLOG_NEXTOID, XLogBeginInsert(), XLogInsert(), and XLogRegisterData().

Referenced by GetNewObjectId().

◆ XLogRecPtrToBytePos()

static uint64 XLogRecPtrToBytePos ( XLogRecPtr  ptr)
static

Definition at line 1941 of file xlog.c.

1942 {
1943  uint64 fullsegs;
1944  uint32 fullpages;
1945  uint32 offset;
1946  uint64 result;
1947 
1948  XLByteToSeg(ptr, fullsegs, wal_segment_size);
1949 
1950  fullpages = (XLogSegmentOffset(ptr, wal_segment_size)) / XLOG_BLCKSZ;
1951  offset = ptr % XLOG_BLCKSZ;
1952 
1953  if (fullpages == 0)
1954  {
1955  result = fullsegs * UsableBytesInSegment;
1956  if (offset > 0)
1957  {
1958  Assert(offset >= SizeOfXLogLongPHD);
1959  result += offset - SizeOfXLogLongPHD;
1960  }
1961  }
1962  else
1963  {
1964  result = fullsegs * UsableBytesInSegment +
1965  (XLOG_BLCKSZ - SizeOfXLogLongPHD) + /* account for first page */
1966  (fullpages - 1) * UsableBytesInPage; /* full pages */
1967  if (offset > 0)
1968  {
1969  Assert(offset >= SizeOfXLogShortPHD);
1970  result += offset - SizeOfXLogShortPHD;
1971  }
1972  }
1973 
1974  return result;
1975 }

References Assert, SizeOfXLogLongPHD, SizeOfXLogShortPHD, UsableBytesInPage, UsableBytesInSegment, wal_segment_size, XLByteToSeg, and XLogSegmentOffset.

Referenced by ReserveXLogInsertLocation(), ReserveXLogSwitch(), and StartupXLOG().

◆ XLogReportParameters()

static void XLogReportParameters ( void  )
static

Definition at line 8103 of file xlog.c.

8104 {
8105  if (wal_level != ControlFile->wal_level ||
8113  {
8114  /*
8115  * The change in number of backend slots doesn't need to be WAL-logged
8116  * if archiving is not enabled, as you can't start archive recovery
8117  * with wal_level=minimal anyway. We don't really care about the
8118  * values in pg_control either if wal_level=minimal, but seems better
8119  * to keep them up-to-date to avoid confusion.
8120  */
8122  {
8123  xl_parameter_change xlrec;
8124  XLogRecPtr recptr;
8125 
8131  xlrec.wal_level = wal_level;
8132  xlrec.wal_log_hints = wal_log_hints;
8134 
8135  XLogBeginInsert();
8136  XLogRegisterData((char *) &xlrec, sizeof(xlrec));
8137 
8138  recptr = XLogInsert(RM_XLOG_ID, XLOG_PARAMETER_CHANGE);
8139  XLogFlush(recptr);
8140  }
8141 
8142  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8143 
8153 
8154  LWLockRelease(ControlFileLock);
8155  }
8156 }

References ControlFile, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), max_locks_per_xact, xl_parameter_change::max_locks_per_xact, ControlFileData::max_locks_per_xact, max_prepared_xacts, xl_parameter_change::max_prepared_xacts, ControlFileData::max_prepared_xacts, max_wal_senders, xl_parameter_change::max_wal_senders, ControlFileData::max_wal_senders, max_worker_processes, xl_parameter_change::max_worker_processes, ControlFileData::max_worker_processes, MaxConnections, xl_parameter_change::MaxConnections, ControlFileData::MaxConnections, track_commit_timestamp, xl_parameter_change::track_commit_timestamp, ControlFileData::track_commit_timestamp, UpdateControlFile(), wal_level, xl_parameter_change::wal_level, ControlFileData::wal_level, wal_log_hints, xl_parameter_change::wal_log_hints, ControlFileData::wal_log_hints, XLOG_PARAMETER_CHANGE, XLogBeginInsert(), XLogFlush(), XLogInsert(), XLogIsNeeded, and XLogRegisterData().

Referenced by StartupXLOG().

◆ XLogRestorePoint()

XLogRecPtr XLogRestorePoint ( const char *  rpName)

Definition at line 8078 of file xlog.c.

8079 {
8080  XLogRecPtr RecPtr;
8081  xl_restore_point xlrec;
8082 
8083  xlrec.rp_time = GetCurrentTimestamp();
8084  strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
8085 
8086  XLogBeginInsert();
8087  XLogRegisterData((char *) &xlrec, sizeof(xl_restore_point));
8088 
8089  RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
8090 
8091  ereport(LOG,
8092  (errmsg("restore point \"%s\" created at %X/%X",
8093  rpName, LSN_FORMAT_ARGS(RecPtr))));
8094 
8095  return RecPtr;
8096 }
char rp_name[MAXFNAMELEN]
TimestampTz rp_time

References ereport, errmsg(), GetCurrentTimestamp(), LOG, LSN_FORMAT_ARGS, MAXFNAMELEN, xl_restore_point::rp_name, xl_restore_point::rp_time, strlcpy(), XLOG_RESTORE_POINT, XLogBeginInsert(), XLogInsert(), and XLogRegisterData().

Referenced by pg_create_restore_point().

◆ XLogSetAsyncXactLSN()

void XLogSetAsyncXactLSN ( XLogRecPtr  asyncXactLSN)

Definition at line 2631 of file xlog.c.

2632 {
2633  XLogRecPtr WriteRqstPtr = asyncXactLSN;
2634  bool sleeping;
2635  bool wakeup = false;
2636  XLogRecPtr prevAsyncXactLSN;
2637 
2639  sleeping = XLogCtl->WalWriterSleeping;
2640  prevAsyncXactLSN = XLogCtl->asyncXactLSN;
2641  if (XLogCtl->asyncXactLSN < asyncXactLSN)
2642  XLogCtl->asyncXactLSN = asyncXactLSN;
2644 
2645  /*
2646  * If somebody else already called this function with a more aggressive
2647  * LSN, they will have done what we needed (and perhaps more).
2648  */
2649  if (asyncXactLSN <= prevAsyncXactLSN)
2650  return;
2651 
2652  /*
2653  * If the WALWriter is sleeping, kick it to make it come out of low-power
2654  * mode, so that this async commit will reach disk within the expected
2655  * amount of time. Otherwise, determine whether it has enough WAL
2656  * available to flush, the same way that XLogBackgroundFlush() does.
2657  */
2658  if (sleeping)
2659  wakeup = true;
2660  else
2661  {
2662  int flushblocks;
2663 
2665 
2666  flushblocks =
2667  WriteRqstPtr / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2668 
2669  if (WalWriterFlushAfter == 0 || flushblocks >= WalWriterFlushAfter)
2670  wakeup = true;
2671  }
2672 
2673  if (wakeup)
2674  {
2675  volatile PROC_HDR *procglobal = ProcGlobal;
2676  ProcNumber walwriterProc = procglobal->walwriterProc;
2677 
2678  if (walwriterProc != INVALID_PROC_NUMBER)
2679  SetLatch(&GetPGProcByNumber(walwriterProc)->procLatch);
2680  }
2681 }
void SetLatch(Latch *latch)
Definition: latch.c:632
#define GetPGProcByNumber(n)
Definition: proc.h:436
#define INVALID_PROC_NUMBER
Definition: procnumber.h:26
int ProcNumber
Definition: procnumber.h:24
PROC_HDR * ProcGlobal
Definition: proc.c:78
Definition: proc.h:382
ProcNumber walwriterProc
Definition: proc.h:420
static TimestampTz wakeup[NUM_WALRCV_WAKEUPS]
Definition: walreceiver.c:129

References XLogCtlData::asyncXactLSN, XLogwrtResult::Flush, GetPGProcByNumber, XLogCtlData::info_lck, INVALID_PROC_NUMBER, LogwrtResult, ProcGlobal, RefreshXLogWriteResult, SetLatch(), SpinLockAcquire, SpinLockRelease, wakeup, WalWriterFlushAfter, PROC_HDR::walwriterProc, XLogCtlData::WalWriterSleeping, and XLogCtl.

Referenced by AbortTransaction(), LogCurrentRunningXacts(), RecordTransactionAbort(), and RecordTransactionCommit().

◆ XLogSetReplicationSlotMinimumLSN()

void XLogSetReplicationSlotMinimumLSN ( XLogRecPtr  lsn)

◆ XLOGShmemInit()

void XLOGShmemInit ( void  )

Definition at line 4918 of file xlog.c.

4919 {
4920  bool foundCFile,
4921  foundXLog;
4922  char *allocptr;
4923  int i;
4924  ControlFileData *localControlFile;
4925 
4926 #ifdef WAL_DEBUG
4927 
4928  /*
4929  * Create a memory context for WAL debugging that's exempt from the normal
4930  * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4931  * an allocation fails, but wal_debug is not for production use anyway.
4932  */
4933  if (walDebugCxt == NULL)
4934  {
4936  "WAL Debug",
4938  MemoryContextAllowInCriticalSection(walDebugCxt, true);
4939  }
4940 #endif
4941 
4942 
4943  XLogCtl = (XLogCtlData *)
4944  ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4945 
4946  localControlFile = ControlFile;
4948  ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4949 
4950  if (foundCFile || foundXLog)
4951  {
4952  /* both should be present or neither */
4953  Assert(foundCFile && foundXLog);
4954 
4955  /* Initialize local copy of WALInsertLocks */
4957 
4958  if (localControlFile)
4959  pfree(localControlFile);
4960  return;
4961  }
4962  memset(XLogCtl, 0, sizeof(XLogCtlData));
4963 
4964  /*
4965  * Already have read control file locally, unless in bootstrap mode. Move
4966  * contents into shared memory.
4967  */
4968  if (localControlFile)
4969  {
4970  memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
4971  pfree(localControlFile);
4972  }
4973 
4974  /*
4975  * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
4976  * multiple of the alignment for same, so no extra alignment padding is
4977  * needed here.
4978  */
4979  allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
4980  XLogCtl->xlblocks = (pg_atomic_uint64 *) allocptr;
4981  allocptr += sizeof(pg_atomic_uint64) * XLOGbuffers;
4982 
4983  for (i = 0; i < XLOGbuffers; i++)
4984  {
4986  }
4987 
4988  /* WAL insertion locks. Ensure they're aligned to the full padded size */
4989  allocptr += sizeof(WALInsertLockPadded) -
4990  ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
4992  (WALInsertLockPadded *) allocptr;
4993  allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
4994 
4995  for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
4996  {
5000  }
5001 
5002  /*
5003  * Align the start of the page buffers to a full xlog block size boundary.
5004  * This simplifies some calculations in XLOG insertion. It is also
5005  * required for O_DIRECT.
5006  */
5007  allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
5008  XLogCtl->pages = allocptr;
5009  memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
5010 
5011  /*
5012  * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
5013  * in additional info.)
5014  */
5018  XLogCtl->WalWriterSleeping = false;
5019 
5026 }
static void pg_atomic_init_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:453
struct pg_atomic_uint64 pg_atomic_uint64
void LWLockInitialize(LWLock *lock, int tranche_id)
Definition: lwlock.c:707
@ LWTRANCHE_WAL_INSERT
Definition: lwlock.h:186
MemoryContext TopMemoryContext
Definition: mcxt.c:149
void MemoryContextAllowInCriticalSection(MemoryContext context, bool allow)
Definition: mcxt.c:694
#define AllocSetContextCreate
Definition: memutils.h:129
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:160
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:387
#define SpinLockInit(lock)
Definition: spin.h:57
int XLogCacheBlck
Definition: xlog.c:494
WALInsertLockPadded * WALInsertLocks
Definition: xlog.c:444
slock_t insertpos_lck
Definition: xlog.c:398
union WALInsertLockPadded WALInsertLockPadded
Size XLOGShmemSize(void)
Definition: xlog.c:4868
struct XLogCtlData XLogCtlData

References ALLOCSET_DEFAULT_SIZES, AllocSetContextCreate, Assert, ControlFile, i, XLogCtlData::info_lck, XLogCtlData::Insert, XLogCtlInsert::insertpos_lck, XLogCtlData::InstallXLogFileSegmentActive, InvalidXLogRecPtr, WALInsertLockPadded::l, WALInsertLock::lastImportantAt, XLogCtlData::logFlushResult, XLogCtlData::logInsertResult, XLogCtlData::logWriteResult, LWLockInitialize(), LWTRANCHE_WAL_INSERT, MemoryContextAllowInCriticalSection(), NUM_XLOGINSERT_LOCKS, XLogCtlData::pages, pfree(), pg_atomic_init_u64(), RECOVERY_STATE_CRASH, XLogCtlData::SharedRecoveryState, ShmemInitStruct(), SpinLockInit, TopMemoryContext, TYPEALIGN, XLogCtlData::unloggedLSN, XLogCtlInsert::WALInsertLocks, WALInsertLocks, XLogCtlData::WalWriterSleeping, XLogCtlData::xlblocks, XLOGbuffers, XLogCtlData::XLogCacheBlck, XLogCtl, and XLOGShmemSize().

Referenced by CreateOrAttachShmemStructs().

◆ XLOGShmemSize()

Size XLOGShmemSize ( void  )

Definition at line 4868 of file xlog.c.

4869 {
4870  Size size;
4871 
4872  /*
4873  * If the value of wal_buffers is -1, use the preferred auto-tune value.
4874  * This isn't an amazingly clean place to do this, but we must wait till
4875  * NBuffers has received its final value, and must do it before using the
4876  * value of XLOGbuffers to do anything important.
4877  *
4878  * We prefer to report this value's source as PGC_S_DYNAMIC_DEFAULT.
4879  * However, if the DBA explicitly set wal_buffers = -1 in the config file,
4880  * then PGC_S_DYNAMIC_DEFAULT will fail to override that and we must force
4881  * the matter with PGC_S_OVERRIDE.
4882  */
4883  if (XLOGbuffers == -1)
4884  {
4885  char buf[32];
4886 
4887  snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
4888  SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4890  if (XLOGbuffers == -1) /* failed to apply it? */
4891  SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4892  PGC_S_OVERRIDE);
4893  }
4894  Assert(XLOGbuffers > 0);
4895 
4896  /* XLogCtl */
4897  size = sizeof(XLogCtlData);
4898 
4899  /* WAL insertion locks, plus alignment */
4901  /* xlblocks array */
4903  /* extra alignment padding for XLOG I/O buffers */
4904  size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
4905  /* and the buffers themselves */
4906  size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4907 
4908  /*
4909  * Note: we don't count ControlFileData, it comes out of the "slop factor"
4910  * added by CreateSharedMemoryAndSemaphores. This lets us use this
4911  * routine again below to compute the actual allocation size.
4912  */
4913 
4914  return size;
4915 }
#define Max(x, y)
Definition: c.h:1001
@ PGC_S_OVERRIDE
Definition: guc.h:119
@ PGC_POSTMASTER
Definition: guc.h:70
#define PG_IO_ALIGN_SIZE
Size add_size(Size s1, Size s2)
Definition: shmem.c:493
Size mul_size(Size s1, Size s2)
Definition: shmem.c:510

References add_size(), Assert, buf, Max, mul_size(), NUM_XLOGINSERT_LOCKS, PG_IO_ALIGN_SIZE, PGC_POSTMASTER, PGC_S_DYNAMIC_DEFAULT, PGC_S_OVERRIDE, SetConfigOption(), size, snprintf, XLOGbuffers, and XLOGChooseNumBuffers().

Referenced by CalculateShmemSize(), and XLOGShmemInit().

◆ XLogShutdownWalRcv()

void XLogShutdownWalRcv ( void  )

Definition at line 9473 of file xlog.c.

9474 {
9475  ShutdownWalRcv();
9476 
9477  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9479  LWLockRelease(ControlFileLock);
9480 }
void ShutdownWalRcv(void)

References XLogCtlData::InstallXLogFileSegmentActive, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), ShutdownWalRcv(), and XLogCtl.

Referenced by FinishWalRecovery(), and WaitForWALToBecomeAvailable().

◆ XLogWrite()

static void XLogWrite ( XLogwrtRqst  WriteRqst,
TimeLineID  tli,
bool  flexible 
)
static

Definition at line 2314 of file xlog.c.

2315 {
2316  bool ispartialpage;
2317  bool last_iteration;
2318  bool finishing_seg;
2319  int curridx;
2320  int npages;
2321  int startidx;
2322  uint32 startoffset;
2323 
2324  /* We should always be inside a critical section here */
2325  Assert(CritSectionCount > 0);
2326 
2327  /*
2328  * Update local LogwrtResult (caller probably did this already, but...)
2329  */
2331 
2332  /*
2333  * Since successive pages in the xlog cache are consecutively allocated,
2334  * we can usually gather multiple pages together and issue just one
2335  * write() call. npages is the number of pages we have determined can be
2336  * written together; startidx is the cache block index of the first one,
2337  * and startoffset is the file offset at which it should go. The latter
2338  * two variables are only valid when npages > 0, but we must initialize
2339  * all of them to keep the compiler quiet.
2340  */
2341  npages = 0;
2342  startidx = 0;
2343  startoffset = 0;
2344 
2345  /*
2346  * Within the loop, curridx is the cache block index of the page to
2347  * consider writing. Begin at the buffer containing the next unwritten
2348  * page, or last partially written page.
2349  */
2351 
2352  while (LogwrtResult.Write < WriteRqst.Write)
2353  {
2354  /*
2355  * Make sure we're not ahead of the insert process. This could happen
2356  * if we're passed a bogus WriteRqst.Write that is past the end of the
2357  * last page that's been initialized by AdvanceXLInsertBuffer.
2358  */
2359  XLogRecPtr EndPtr = pg_atomic_read_u64(&XLogCtl->xlblocks[curridx]);
2360 
2361  if (LogwrtResult.Write >= EndPtr)
2362  elog(PANIC, "xlog write request %X/%X is past end of log %X/%X",
2364  LSN_FORMAT_ARGS(EndPtr));
2365 
2366  /* Advance LogwrtResult.Write to end of current buffer page */
2367  LogwrtResult.Write = EndPtr;
2368  ispartialpage = WriteRqst.Write < LogwrtResult.Write;
2369 
2372  {
2373  /*
2374  * Switch to new logfile segment. We cannot have any pending
2375  * pages here (since we dump what we have at segment end).
2376  */
2377  Assert(npages == 0);
2378  if (openLogFile >= 0)
2379  XLogFileClose();
2382  openLogTLI = tli;
2383 
2384  /* create/use new log file */
2387  }
2388 
2389  /* Make sure we have the current logfile open */
2390  if (openLogFile < 0)
2391  {
2394  openLogTLI = tli;
2397  }
2398 
2399  /* Add current page to the set of pending pages-to-dump */
2400  if (npages == 0)
2401  {
2402  /* first of group */
2403  startidx = curridx;
2404  startoffset = XLogSegmentOffset(LogwrtResult.Write - XLOG_BLCKSZ,
2406  }
2407  npages++;
2408 
2409  /*
2410  * Dump the set if this will be the last loop iteration, or if we are
2411  * at the last page of the cache area (since the next page won't be
2412  * contiguous in memory), or if we are at the end of the logfile
2413  * segment.
2414  */
2415  last_iteration = WriteRqst.Write <= LogwrtResult.Write;
2416 
2417  finishing_seg = !ispartialpage &&
2418  (startoffset + npages * XLOG_BLCKSZ) >= wal_segment_size;
2419 
2420  if (last_iteration ||
2421  curridx == XLogCtl->XLogCacheBlck ||
2422  finishing_seg)
2423  {
2424  char *from;
2425  Size nbytes;
2426  Size nleft;
2427  ssize_t written;
2428  instr_time start;
2429 
2430  /* OK to write the page(s) */
2431  from = XLogCtl->pages + startidx * (Size) XLOG_BLCKSZ;
2432  nbytes = npages * (Size) XLOG_BLCKSZ;
2433  nleft = nbytes;
2434  do
2435  {
2436  errno = 0;
2437 
2438  /* Measure I/O timing to write WAL data */
2439  if (track_wal_io_timing)
2441  else
2443 
2444  pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
2445  written = pg_pwrite(openLogFile, from, nleft, startoffset);
2447 
2448  /*
2449  * Increment the I/O timing and the number of times WAL data
2450  * were written out to disk.
2451  */
2452  if (track_wal_io_timing)
2453  {
2454  instr_time end;
2455 
2458  }
2459 
2461 
2462  if (written <= 0)
2463  {
2464  char xlogfname[MAXFNAMELEN];
2465  int save_errno;
2466 
2467  if (errno == EINTR)
2468  continue;
2469 
2470  save_errno = errno;
2471  XLogFileName(xlogfname, tli, openLogSegNo,
2473  errno = save_errno;
2474  ereport(PANIC,
2476  errmsg("could not write to log file \"%s\" at offset %u, length %zu: %m",
2477  xlogfname, startoffset, nleft)));
2478  }
2479  nleft -= written;
2480  from += written;
2481  startoffset += written;
2482  } while (nleft > 0);
2483 
2484  npages = 0;
2485 
2486  /*
2487  * If we just wrote the whole last page of a logfile segment,
2488  * fsync the segment immediately. This avoids having to go back
2489  * and re-open prior segments when an fsync request comes along
2490  * later. Doing it here ensures that one and only one backend will
2491  * perform this fsync.
2492  *
2493  * This is also the right place to notify the Archiver that the
2494  * segment is ready to copy to archival storage, and to update the
2495  * timer for archive_timeout, and to signal for a checkpoint if
2496  * too many logfile segments have been used since the last
2497  * checkpoint.
2498  */
2499  if (finishing_seg)
2500  {
2502 
2503  /* signal that we need to wakeup walsenders later */
2505 
2506  LogwrtResult.Flush = LogwrtResult.Write; /* end of page */
2507 
2508  if (XLogArchivingActive())
2510 
2511  XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
2513 
2514  /*
2515  * Request a checkpoint if we've consumed too much xlog since
2516  * the last one. For speed, we first check using the local
2517  * copy of RedoRecPtr, which might be out of date; if it looks
2518  * like a checkpoint is needed, forcibly update RedoRecPtr and
2519  * recheck.
2520  */
2522  {
2523  (void) GetRedoRecPtr();
2526  }
2527  }
2528  }
2529 
2530  if (ispartialpage)
2531  {
2532  /* Only asked to write a partial page */
2533  LogwrtResult.Write = WriteRqst.Write;
2534  break;
2535  }
2536  curridx = NextBufIdx(curridx);
2537 
2538  /* If flexible, break out of loop as soon as we wrote something */
2539  if (flexible && npages == 0)
2540  break;
2541  }
2542 
2543  Assert(npages == 0);
2544 
2545  /*
2546  * If asked to flush, do so
2547  */
2548  if (LogwrtResult.Flush < WriteRqst.Flush &&
2550  {
2551  /*
2552  * Could get here without iterating above loop, in which case we might
2553  * have no open file or the wrong one. However, we do not need to
2554  * fsync more than one file.
2555  */
2558  {
2559  if (openLogFile >= 0 &&
2562  XLogFileClose();
2563  if (openLogFile < 0)
2564  {
2567  openLogTLI = tli;
2570  }
2571 
2573  }
2574 
2575  /* signal that we need to wakeup walsenders later */
2577 
2579  }
2580 
2581  /*
2582  * Update shared-memory status
2583  *
2584  * We make sure that the shared 'request' values do not fall behind the
2585  * 'result' values. This is not absolutely essential, but it saves some
2586  * code in a couple of places.
2587  */
2594 
2595  /*
2596  * We write Write first, bar, then Flush. When reading, the opposite must
2597  * be done (with a matching barrier in between), so that we always see a
2598  * Flush value that trails behind the Write value seen.
2599  */
2601  pg_write_barrier();
2603 
2604 #ifdef USE_ASSERT_CHECKING
2605  {
2606  XLogRecPtr Flush;
2607  XLogRecPtr Write;
2609 
2611  pg_read_barrier();
2613  pg_read_barrier();
2615 
2616  /* WAL written to disk is always ahead of WAL flushed */
2617  Assert(Write >= Flush);
2618 
2619  /* WAL inserted to buffers is always ahead of WAL written */
2620  Assert(Insert >= Write);
2621  }
2622 #endif
2623 }
void ReserveExternalFD(void)
Definition: fd.c:1220
volatile uint32 CritSectionCount
Definition: globals.c:44
PgStat_Counter wal_write
Definition: pgstat.h:490
instr_time wal_write_time
Definition: pgstat.h:492
XLogRecPtr Flush
Definition: walreceiver.c:111
XLogRecPtr Write
Definition: walreceiver.c:110
#define WalSndWakeupRequest()
Definition: walsender.h:58
#define EINTR
Definition: win32_port.h:374
XLogRecPtr GetRedoRecPtr(void)
Definition: xlog.c:6437
int XLogFileOpen(XLogSegNo segno, TimeLineID tli)
Definition: xlog.c:3618
#define NextBufIdx(idx)
Definition: xlog.c:584
void issue_xlog_fsync(int fd, XLogSegNo segno, TimeLineID tli)
Definition: xlog.c:8683
bool XLogCheckpointNeeded(XLogSegNo new_segno)
Definition: xlog.c:2290
void XLogArchiveNotifySeg(XLogSegNo segno, TimeLineID tli)
Definition: xlogarchive.c:492

References Assert, CHECKPOINT_CAUSE_XLOG, CritSectionCount, EINTR, elog, ereport, errcode_for_file_access(), errmsg(), XLogwrtRqst::Flush, XLogwrtResult::Flush, Flush, GetRedoRecPtr(), XLogCtlData::info_lck, Insert(), INSTR_TIME_ACCUM_DIFF, INSTR_TIME_SET_CURRENT, INSTR_TIME_SET_ZERO, issue_xlog_fsync(), IsUnderPostmaster, XLogCtlData::lastSegSwitchLSN, XLogCtlData::lastSegSwitchTime, XLogCtlData::logFlushResult, XLogCtlData::logInsertResult, XLogCtlData::logWriteResult, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, MAXFNAMELEN, NextBufIdx, openLogFile, openLogSegNo, openLogTLI, XLogCtlData::pages, PANIC, PendingWalStats, pg_atomic_read_u64(), pg_atomic_write_u64(), pg_pwrite, pg_read_barrier, pg_write_barrier, pgstat_report_wait_end(), pgstat_report_wait_start(), RefreshXLogWriteResult, RequestCheckpoint(), ReserveExternalFD(), SpinLockAcquire, SpinLockRelease, start, track_wal_io_timing, wal_segment_size, wal_sync_method, WAL_SYNC_METHOD_OPEN, WAL_SYNC_METHOD_OPEN_DSYNC, PgStat_PendingWalStats::wal_write, PgStat_PendingWalStats::wal_write_time, WalSndWakeupRequest, XLogwrtRqst::Write, XLogwrtResult::Write, Write, XLogCtlData::xlblocks, XLByteInPrevSeg, XLByteToPrevSeg, XLogArchiveNotifySeg(), XLogArchivingActive, XLogCtlData::XLogCacheBlck, XLogCheckpointNeeded(), XLogCtl, XLogFileClose(), XLogFileInit(), XLogFileName(), XLogFileOpen(), XLogRecPtrToBufIdx, and XLogSegmentOffset.

Referenced by AdvanceXLInsertBuffer(), XLogBackgroundFlush(), and XLogFlush().

Variable Documentation

◆ archive_mode_options

const struct config_enum_entry archive_mode_options[]
Initial value:
= {
{"always", ARCHIVE_MODE_ALWAYS, false},
{"on", ARCHIVE_MODE_ON, false},
{"off", ARCHIVE_MODE_OFF, false},
{"true", ARCHIVE_MODE_ON, true},
{"false", ARCHIVE_MODE_OFF, true},
{"yes", ARCHIVE_MODE_ON, true},
{"no", ARCHIVE_MODE_OFF, true},
{"1", ARCHIVE_MODE_ON, true},
{"0", ARCHIVE_MODE_OFF, true},
{NULL, 0, false}
}
@ ARCHIVE_MODE_ALWAYS
Definition: xlog.h:67
@ ARCHIVE_MODE_OFF
Definition: xlog.h:65
@ ARCHIVE_MODE_ON
Definition: xlog.h:66

Definition at line 166 of file xlog.c.

◆ check_wal_consistency_checking_deferred

bool check_wal_consistency_checking_deferred = false
static

Definition at line 166 of file xlog.c.

Referenced by check_wal_consistency_checking(), and InitializeWalConsistencyChecking().

◆ CheckPointDistanceEstimate

double CheckPointDistanceEstimate = 0
static

Definition at line 159 of file xlog.c.

Referenced by LogCheckpointEnd(), UpdateCheckPointDistanceEstimate(), and XLOGfileslop().

◆ CheckPointSegments

int CheckPointSegments

◆ CheckpointStats

◆ CommitDelay

int CommitDelay = 0

Definition at line 132 of file xlog.c.

Referenced by XLogFlush().

◆ CommitSiblings

int CommitSiblings = 5

Definition at line 133 of file xlog.c.

Referenced by XLogFlush().

◆ ControlFile

◆ doPageWrites

◆ EnableHotStandby

◆ fullPageWrites

bool fullPageWrites = true

Definition at line 122 of file xlog.c.

Referenced by BootStrapXLOG(), and UpdateFullPageWrites().

◆ holdingAllLocks

bool holdingAllLocks = false
static

◆ lastFullPageWrites

bool lastFullPageWrites
static

Definition at line 217 of file xlog.c.

Referenced by StartupXLOG(), and xlog_redo().

◆ LocalMinRecoveryPoint

XLogRecPtr LocalMinRecoveryPoint
static

◆ LocalMinRecoveryPointTLI

TimeLineID LocalMinRecoveryPointTLI
static

◆ LocalRecoveryInProgress

bool LocalRecoveryInProgress = true
static

Definition at line 224 of file xlog.c.

Referenced by RecoveryInProgress().

◆ LocalXLogInsertAllowed

int LocalXLogInsertAllowed = -1
static

Definition at line 236 of file xlog.c.

Referenced by CreateCheckPoint(), LocalSetXLogInsertAllowed(), and XLogInsertAllowed().

◆ log_checkpoints

bool log_checkpoints = true

◆ LogwrtResult

◆ max_slot_wal_keep_size_mb

int max_slot_wal_keep_size_mb = -1

Definition at line 135 of file xlog.c.

Referenced by KeepLogSeg(), and pg_get_replication_slots().

◆ max_wal_size_mb

int max_wal_size_mb = 1024

◆ min_wal_size_mb

int min_wal_size_mb = 80

Definition at line 115 of file xlog.c.

Referenced by ReadControlFile(), and XLOGfileslop().

◆ MyLockNo

int MyLockNo = 0
static

◆ openLogFile

int openLogFile = -1
static

◆ openLogSegNo

XLogSegNo openLogSegNo = 0
static

Definition at line 636 of file xlog.c.

Referenced by assign_wal_sync_method(), XLogBackgroundFlush(), XLogFileClose(), and XLogWrite().

◆ openLogTLI

TimeLineID openLogTLI = 0
static

Definition at line 637 of file xlog.c.

Referenced by assign_wal_sync_method(), BootStrapXLOG(), XLogFileClose(), and XLogWrite().

◆ PrevCheckPointDistance

double PrevCheckPointDistance = 0
static

Definition at line 160 of file xlog.c.

Referenced by LogCheckpointEnd(), and UpdateCheckPointDistanceEstimate().

◆ ProcLastRecPtr

◆ RedoRecPtr

◆ sessionBackupState

SessionBackupState sessionBackupState = SESSION_BACKUP_NONE
static

◆ track_wal_io_timing

bool track_wal_io_timing = false

Definition at line 137 of file xlog.c.

Referenced by issue_xlog_fsync(), and XLogWrite().

◆ updateMinRecoveryPoint

bool updateMinRecoveryPoint = true
static

Definition at line 648 of file xlog.c.

Referenced by SwitchIntoArchiveRecovery(), UpdateMinRecoveryPoint(), and XLogNeedsFlush().

◆ UsableBytesInSegment

int UsableBytesInSegment
static

◆ wal_compression

int wal_compression = WAL_COMPRESSION_NONE

Definition at line 124 of file xlog.c.

Referenced by XLogCompressBackupBlock(), and XLogRecordAssemble().

◆ wal_consistency_checking

bool* wal_consistency_checking = NULL

Definition at line 126 of file xlog.c.

Referenced by assign_wal_consistency_checking(), and XLogRecordAssemble().

◆ wal_consistency_checking_string

char* wal_consistency_checking_string = NULL

Definition at line 125 of file xlog.c.

Referenced by InitializeWalConsistencyChecking().

◆ wal_decode_buffer_size

int wal_decode_buffer_size = 512 * 1024

Definition at line 136 of file xlog.c.

Referenced by InitWalRecovery().

◆ wal_init_zero

bool wal_init_zero = true

Definition at line 127 of file xlog.c.

Referenced by XLogFileInitInternal().

◆ wal_keep_size_mb

int wal_keep_size_mb = 0

Definition at line 116 of file xlog.c.

Referenced by KeepLogSeg(), and pg_get_replication_slots().

◆ wal_level

◆ wal_log_hints

bool wal_log_hints = false

Definition at line 123 of file xlog.c.

Referenced by InitControlFile(), and XLogReportParameters().

◆ wal_recycle

bool wal_recycle = true

Definition at line 128 of file xlog.c.

Referenced by RemoveXlogFile().

◆ wal_retrieve_retry_interval

int wal_retrieve_retry_interval = 5000

◆ wal_segment_size

int wal_segment_size = DEFAULT_XLOG_SEG_SIZE

Definition at line 143 of file xlog.c.

Referenced by AdvanceXLInsertBuffer(), assign_wal_sync_method(), BootStrapXLOG(), build_backup_content(), CalculateCheckpointSegments(), CheckArchiveTimeout(), CheckXLogRemoved(), CleanupAfterArchiveRecovery(), copy_replication_slot(), CopyXLogRecordToWAL(), CreateCheckPoint(), CreateOverwriteContrecordRecord(), CreateRestartPoint(), do_pg_backup_stop(), ExecuteRecoveryCommand(), FinishWalRecovery(), GetOldestUnsummarizedLSN(), GetWALAvailability(), GetXLogBuffer(), InitWalRecovery(), InitXLogReaderState(), InstallXLogFileSegment(), InvalidateObsoleteReplicationSlots(), IsCheckpointOnSchedule(), issue_xlog_fsync(), KeepLogSeg(), MaybeRemoveOldWalSummaries(), perform_base_backup(), pg_control_checkpoint(), pg_get_replication_slots(), pg_split_walfile_name(), pg_walfile_name(), pg_walfile_name_offset(), PreallocXlogFiles(), ReadControlFile(), ReadRecord(), RemoveNonParentXlogFiles(), RemoveOldXlogFiles(), ReorderBufferRestoreChanges(), ReorderBufferRestoreCleanup(), ReorderBufferSerializedPath(), ReorderBufferSerializeTXN(), ReplicationSlotReserveWal(), RequestXLogStreaming(), reserve_wal_for_local_slot(), ReserveXLogSwitch(), RestoreArchivedFile(), StartReplication(), StartupDecodingContext(), SummarizeWAL(), UpdateLastRemovedPtr(), WALReadRaiseError(), WalReceiverMain(), WalSndSegmentOpen(), WriteControlFile(), XLogArchiveNotifySeg(), XLogBackgroundFlush(), XLogBytePosToEndRecPtr(), XLogBytePosToRecPtr(), XLogCheckpointNeeded(), XLOGChooseNumBuffers(), XLogFileClose(), XLogFileCopy(), XLogFileInitInternal(), XLogFileOpen(), XLogFileRead(), XLogFileReadAnyTLI(), XLOGfileslop(), XLogGetOldestSegno(), XLogInitNewTimeline(), XLogInsertRecord(), XLogPageRead(), XLogReaderAllocate(), XlogReadTwoPhaseData(), XLogRecPtrToBytePos(), XLogWalRcvClose(), XLogWalRcvWrite(), and XLogWrite().

◆ wal_sync_method

◆ wal_sync_method_options

const struct config_enum_entry wal_sync_method_options[]
Initial value:
= {
{"fsync", WAL_SYNC_METHOD_FSYNC, false},
{"fdatasync", WAL_SYNC_METHOD_FDATASYNC, false},
{NULL, 0, false}
}

Definition at line 166 of file xlog.c.

◆ WALInsertLocks

◆ XactLastCommitEnd

◆ XactLastRecEnd

◆ XLogArchiveCommand

char* XLogArchiveCommand = NULL

◆ XLogArchiveMode

◆ XLogArchiveTimeout

int XLogArchiveTimeout = 0

Definition at line 118 of file xlog.c.

Referenced by CheckArchiveTimeout(), and CheckpointerMain().

◆ XLOGbuffers

int XLOGbuffers = -1

Definition at line 117 of file xlog.c.

Referenced by check_wal_buffers(), XLOGShmemInit(), and XLOGShmemSize().

◆ XLogCtl