PostgreSQL Source Code  git master
xlog.h File Reference
#include "access/xlogbackup.h"
#include "access/xlogdefs.h"
#include "datatype/timestamp.h"
#include "lib/stringinfo.h"
#include "nodes/pg_list.h"
Include dependency graph for xlog.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Data Structures

struct  CheckpointStatsData
 

Macros

#define XLogArchivingActive()    (AssertMacro(XLogArchiveMode == ARCHIVE_MODE_OFF || wal_level >= WAL_LEVEL_REPLICA), XLogArchiveMode > ARCHIVE_MODE_OFF)
 
#define XLogArchivingAlways()    (AssertMacro(XLogArchiveMode == ARCHIVE_MODE_OFF || wal_level >= WAL_LEVEL_REPLICA), XLogArchiveMode == ARCHIVE_MODE_ALWAYS)
 
#define XLogIsNeeded()   (wal_level >= WAL_LEVEL_REPLICA)
 
#define XLogHintBitIsNeeded()   (DataChecksumsEnabled() || wal_log_hints)
 
#define XLogStandbyInfoActive()   (wal_level >= WAL_LEVEL_REPLICA)
 
#define XLogLogicalInfoActive()   (wal_level >= WAL_LEVEL_LOGICAL)
 
#define CHECKPOINT_IS_SHUTDOWN   0x0001 /* Checkpoint is for shutdown */
 
#define CHECKPOINT_END_OF_RECOVERY
 
#define CHECKPOINT_IMMEDIATE   0x0004 /* Do it without delays */
 
#define CHECKPOINT_FORCE   0x0008 /* Force even if no activity */
 
#define CHECKPOINT_FLUSH_ALL
 
#define CHECKPOINT_WAIT   0x0020 /* Wait for completion */
 
#define CHECKPOINT_REQUESTED   0x0040 /* Checkpoint request has been made */
 
#define CHECKPOINT_CAUSE_XLOG   0x0080 /* XLOG consumption */
 
#define CHECKPOINT_CAUSE_TIME   0x0100 /* Elapsed time */
 
#define XLOG_INCLUDE_ORIGIN   0x01 /* include the replication origin */
 
#define XLOG_MARK_UNIMPORTANT   0x02 /* record not important for durability */
 
#define RECOVERY_SIGNAL_FILE   "recovery.signal"
 
#define STANDBY_SIGNAL_FILE   "standby.signal"
 
#define BACKUP_LABEL_FILE   "backup_label"
 
#define BACKUP_LABEL_OLD   "backup_label.old"
 
#define TABLESPACE_MAP   "tablespace_map"
 
#define TABLESPACE_MAP_OLD   "tablespace_map.old"
 
#define PROMOTE_SIGNAL_FILE   "promote"
 

Typedefs

typedef enum WalSyncMethod WalSyncMethod
 
typedef enum ArchiveMode ArchiveMode
 
typedef enum WalLevel WalLevel
 
typedef enum WalCompression WalCompression
 
typedef enum RecoveryState RecoveryState
 
typedef struct CheckpointStatsData CheckpointStatsData
 
typedef enum WALAvailability WALAvailability
 
typedef enum SessionBackupState SessionBackupState
 

Enumerations

enum  WalSyncMethod {
  WAL_SYNC_METHOD_FSYNC = 0 , WAL_SYNC_METHOD_FDATASYNC , WAL_SYNC_METHOD_OPEN , WAL_SYNC_METHOD_FSYNC_WRITETHROUGH ,
  WAL_SYNC_METHOD_OPEN_DSYNC
}
 
enum  ArchiveMode { ARCHIVE_MODE_OFF = 0 , ARCHIVE_MODE_ON , ARCHIVE_MODE_ALWAYS }
 
enum  WalLevel { WAL_LEVEL_MINIMAL = 0 , WAL_LEVEL_REPLICA , WAL_LEVEL_LOGICAL }
 
enum  WalCompression { WAL_COMPRESSION_NONE = 0 , WAL_COMPRESSION_PGLZ , WAL_COMPRESSION_LZ4 , WAL_COMPRESSION_ZSTD }
 
enum  RecoveryState { RECOVERY_STATE_CRASH = 0 , RECOVERY_STATE_ARCHIVE , RECOVERY_STATE_DONE }
 
enum  WALAvailability {
  WALAVAIL_INVALID_LSN , WALAVAIL_RESERVED , WALAVAIL_EXTENDED , WALAVAIL_UNRESERVED ,
  WALAVAIL_REMOVED
}
 
enum  SessionBackupState { SESSION_BACKUP_NONE , SESSION_BACKUP_RUNNING }
 

Functions

XLogRecPtr XLogInsertRecord (struct XLogRecData *rdata, XLogRecPtr fpw_lsn, uint8 flags, int num_fpi, bool topxid_included)
 
void XLogFlush (XLogRecPtr record)
 
bool XLogBackgroundFlush (void)
 
bool XLogNeedsFlush (XLogRecPtr record)
 
int XLogFileInit (XLogSegNo logsegno, TimeLineID logtli)
 
int XLogFileOpen (XLogSegNo segno, TimeLineID tli)
 
void CheckXLogRemoved (XLogSegNo segno, TimeLineID tli)
 
XLogSegNo XLogGetLastRemovedSegno (void)
 
XLogSegNo XLogGetOldestSegno (TimeLineID tli)
 
void XLogSetAsyncXactLSN (XLogRecPtr asyncXactLSN)
 
void XLogSetReplicationSlotMinimumLSN (XLogRecPtr lsn)
 
void xlog_redo (struct XLogReaderState *record)
 
void xlog_desc (StringInfo buf, struct XLogReaderState *record)
 
const char * xlog_identify (uint8 info)
 
void issue_xlog_fsync (int fd, XLogSegNo segno, TimeLineID tli)
 
bool RecoveryInProgress (void)
 
RecoveryState GetRecoveryState (void)
 
bool XLogInsertAllowed (void)
 
XLogRecPtr GetXLogInsertRecPtr (void)
 
XLogRecPtr GetXLogWriteRecPtr (void)
 
uint64 GetSystemIdentifier (void)
 
char * GetMockAuthenticationNonce (void)
 
bool DataChecksumsEnabled (void)
 
XLogRecPtr GetFakeLSNForUnloggedRel (void)
 
Size XLOGShmemSize (void)
 
void XLOGShmemInit (void)
 
void BootStrapXLOG (void)
 
void InitializeWalConsistencyChecking (void)
 
void LocalProcessControlFile (bool reset)
 
WalLevel GetActiveWalLevelOnStandby (void)
 
void StartupXLOG (void)
 
void ShutdownXLOG (int code, Datum arg)
 
void CreateCheckPoint (int flags)
 
bool CreateRestartPoint (int flags)
 
WALAvailability GetWALAvailability (XLogRecPtr targetLSN)
 
void XLogPutNextOid (Oid nextOid)
 
XLogRecPtr XLogRestorePoint (const char *rpName)
 
void UpdateFullPageWrites (void)
 
void GetFullPageWriteInfo (XLogRecPtr *RedoRecPtr_p, bool *doPageWrites_p)
 
XLogRecPtr GetRedoRecPtr (void)
 
XLogRecPtr GetInsertRecPtr (void)
 
XLogRecPtr GetFlushRecPtr (TimeLineID *insertTLI)
 
TimeLineID GetWALInsertionTimeLine (void)
 
XLogRecPtr GetLastImportantRecPtr (void)
 
void SetWalWriterSleeping (bool sleeping)
 
Size WALReadFromBuffers (char *dstbuf, XLogRecPtr startptr, Size count, TimeLineID tli)
 
void RemoveNonParentXlogFiles (XLogRecPtr switchpoint, TimeLineID newTLI)
 
bool XLogCheckpointNeeded (XLogSegNo new_segno)
 
void SwitchIntoArchiveRecovery (XLogRecPtr EndRecPtr, TimeLineID replayTLI)
 
void ReachedEndOfBackup (XLogRecPtr EndRecPtr, TimeLineID tli)
 
void SetInstallXLogFileSegmentActive (void)
 
bool IsInstallXLogFileSegmentActive (void)
 
void XLogShutdownWalRcv (void)
 
void do_pg_backup_start (const char *backupidstr, bool fast, List **tablespaces, BackupState *state, StringInfo tblspcmapfile)
 
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)
 
SessionBackupState get_backup_status (void)
 

Variables

PGDLLIMPORT int wal_sync_method
 
PGDLLIMPORT XLogRecPtr ProcLastRecPtr
 
PGDLLIMPORT XLogRecPtr XactLastRecEnd
 
PGDLLIMPORT XLogRecPtr XactLastCommitEnd
 
PGDLLIMPORT int wal_segment_size
 
PGDLLIMPORT int min_wal_size_mb
 
PGDLLIMPORT int max_wal_size_mb
 
PGDLLIMPORT int wal_keep_size_mb
 
PGDLLIMPORT int max_slot_wal_keep_size_mb
 
PGDLLIMPORT int XLOGbuffers
 
PGDLLIMPORT int XLogArchiveTimeout
 
PGDLLIMPORT int wal_retrieve_retry_interval
 
PGDLLIMPORT char * XLogArchiveCommand
 
PGDLLIMPORT bool EnableHotStandby
 
PGDLLIMPORT bool fullPageWrites
 
PGDLLIMPORT bool wal_log_hints
 
PGDLLIMPORT int wal_compression
 
PGDLLIMPORT bool wal_init_zero
 
PGDLLIMPORT bool wal_recycle
 
PGDLLIMPORT boolwal_consistency_checking
 
PGDLLIMPORT char * wal_consistency_checking_string
 
PGDLLIMPORT bool log_checkpoints
 
PGDLLIMPORT bool track_wal_io_timing
 
PGDLLIMPORT int wal_decode_buffer_size
 
PGDLLIMPORT int CheckPointSegments
 
PGDLLIMPORT int XLogArchiveMode
 
PGDLLIMPORT int wal_level
 
PGDLLIMPORT CheckpointStatsData CheckpointStats
 

Macro Definition Documentation

◆ BACKUP_LABEL_FILE

#define BACKUP_LABEL_FILE   "backup_label"

Definition at line 298 of file xlog.h.

◆ BACKUP_LABEL_OLD

#define BACKUP_LABEL_OLD   "backup_label.old"

Definition at line 299 of file xlog.h.

◆ CHECKPOINT_CAUSE_TIME

#define CHECKPOINT_CAUSE_TIME   0x0100 /* Elapsed time */

Definition at line 147 of file xlog.h.

◆ CHECKPOINT_CAUSE_XLOG

#define CHECKPOINT_CAUSE_XLOG   0x0080 /* XLOG consumption */

Definition at line 146 of file xlog.h.

◆ CHECKPOINT_END_OF_RECOVERY

#define CHECKPOINT_END_OF_RECOVERY
Value:
0x0002 /* Like shutdown checkpoint, but
* issued at end of WAL recovery */

Definition at line 138 of file xlog.h.

◆ CHECKPOINT_FLUSH_ALL

#define CHECKPOINT_FLUSH_ALL
Value:
0x0010 /* Flush all pages, including those
* belonging to unlogged tables */

Definition at line 141 of file xlog.h.

◆ CHECKPOINT_FORCE

#define CHECKPOINT_FORCE   0x0008 /* Force even if no activity */

Definition at line 140 of file xlog.h.

◆ CHECKPOINT_IMMEDIATE

#define CHECKPOINT_IMMEDIATE   0x0004 /* Do it without delays */

Definition at line 139 of file xlog.h.

◆ CHECKPOINT_IS_SHUTDOWN

#define CHECKPOINT_IS_SHUTDOWN   0x0001 /* Checkpoint is for shutdown */

Definition at line 137 of file xlog.h.

◆ CHECKPOINT_REQUESTED

#define CHECKPOINT_REQUESTED   0x0040 /* Checkpoint request has been made */

Definition at line 144 of file xlog.h.

◆ CHECKPOINT_WAIT

#define CHECKPOINT_WAIT   0x0020 /* Wait for completion */

Definition at line 143 of file xlog.h.

◆ PROMOTE_SIGNAL_FILE

#define PROMOTE_SIGNAL_FILE   "promote"

Definition at line 305 of file xlog.h.

◆ RECOVERY_SIGNAL_FILE

#define RECOVERY_SIGNAL_FILE   "recovery.signal"

Definition at line 296 of file xlog.h.

◆ STANDBY_SIGNAL_FILE

#define STANDBY_SIGNAL_FILE   "standby.signal"

Definition at line 297 of file xlog.h.

◆ TABLESPACE_MAP

#define TABLESPACE_MAP   "tablespace_map"

Definition at line 301 of file xlog.h.

◆ TABLESPACE_MAP_OLD

#define TABLESPACE_MAP_OLD   "tablespace_map.old"

Definition at line 302 of file xlog.h.

◆ XLOG_INCLUDE_ORIGIN

#define XLOG_INCLUDE_ORIGIN   0x01 /* include the replication origin */

Definition at line 152 of file xlog.h.

◆ XLOG_MARK_UNIMPORTANT

#define XLOG_MARK_UNIMPORTANT   0x02 /* record not important for durability */

Definition at line 153 of file xlog.h.

◆ XLogArchivingActive

Definition at line 97 of file xlog.h.

◆ XLogArchivingAlways

Definition at line 100 of file xlog.h.

◆ XLogHintBitIsNeeded

#define XLogHintBitIsNeeded ( )    (DataChecksumsEnabled() || wal_log_hints)

Definition at line 118 of file xlog.h.

◆ XLogIsNeeded

#define XLogIsNeeded ( )    (wal_level >= WAL_LEVEL_REPLICA)

Definition at line 107 of file xlog.h.

◆ XLogLogicalInfoActive

#define XLogLogicalInfoActive ( )    (wal_level >= WAL_LEVEL_LOGICAL)

Definition at line 124 of file xlog.h.

◆ XLogStandbyInfoActive

#define XLogStandbyInfoActive ( )    (wal_level >= WAL_LEVEL_REPLICA)

Definition at line 121 of file xlog.h.

Typedef Documentation

◆ ArchiveMode

typedef enum ArchiveMode ArchiveMode

◆ CheckpointStatsData

◆ RecoveryState

◆ SessionBackupState

◆ WALAvailability

◆ WalCompression

◆ WalLevel

typedef enum WalLevel WalLevel

◆ WalSyncMethod

Enumeration Type Documentation

◆ ArchiveMode

Enumerator
ARCHIVE_MODE_OFF 
ARCHIVE_MODE_ON 
ARCHIVE_MODE_ALWAYS 

Definition at line 61 of file xlog.h.

62 {
63  ARCHIVE_MODE_OFF = 0, /* disabled */
64  ARCHIVE_MODE_ON, /* enabled while server is running normally */
65  ARCHIVE_MODE_ALWAYS, /* enabled always (even during recovery) */
66 } ArchiveMode;
ArchiveMode
Definition: xlog.h:62
@ ARCHIVE_MODE_ALWAYS
Definition: xlog.h:65
@ ARCHIVE_MODE_OFF
Definition: xlog.h:63
@ ARCHIVE_MODE_ON
Definition: xlog.h:64

◆ RecoveryState

Enumerator
RECOVERY_STATE_CRASH 
RECOVERY_STATE_ARCHIVE 
RECOVERY_STATE_DONE 

Definition at line 87 of file xlog.h.

88 {
89  RECOVERY_STATE_CRASH = 0, /* crash recovery */
90  RECOVERY_STATE_ARCHIVE, /* archive recovery */
91  RECOVERY_STATE_DONE, /* currently in production */
RecoveryState
Definition: xlog.h:88
@ RECOVERY_STATE_CRASH
Definition: xlog.h:89
@ RECOVERY_STATE_DONE
Definition: xlog.h:91
@ RECOVERY_STATE_ARCHIVE
Definition: xlog.h:90

◆ SessionBackupState

Enumerator
SESSION_BACKUP_NONE 
SESSION_BACKUP_RUNNING 

Definition at line 281 of file xlog.h.

284 {
@ SESSION_BACKUP_NONE
Definition: xlog.h:283

◆ WALAvailability

Enumerator
WALAVAIL_INVALID_LSN 
WALAVAIL_RESERVED 
WALAVAIL_EXTENDED 
WALAVAIL_UNRESERVED 
WALAVAIL_REMOVED 

Definition at line 184 of file xlog.h.

187 {
188  WALAVAIL_INVALID_LSN, /* parameter error */
189  WALAVAIL_RESERVED, /* WAL segment is within max_wal_size */
190  WALAVAIL_EXTENDED, /* WAL segment is reserved by a slot or
191  * wal_keep_size */
192  WALAVAIL_UNRESERVED, /* no longer reserved, but not removed yet */
@ WALAVAIL_RESERVED
Definition: xlog.h:187
@ WALAVAIL_UNRESERVED
Definition: xlog.h:190
@ WALAVAIL_EXTENDED
Definition: xlog.h:188
@ WALAVAIL_INVALID_LSN
Definition: xlog.h:186

◆ WalCompression

Enumerator
WAL_COMPRESSION_NONE 
WAL_COMPRESSION_PGLZ 
WAL_COMPRESSION_LZ4 
WAL_COMPRESSION_ZSTD 

Definition at line 78 of file xlog.h.

79 {
WalCompression
Definition: xlog.h:79
@ WAL_COMPRESSION_NONE
Definition: xlog.h:80
@ WAL_COMPRESSION_LZ4
Definition: xlog.h:82
@ WAL_COMPRESSION_PGLZ
Definition: xlog.h:81
@ WAL_COMPRESSION_ZSTD
Definition: xlog.h:83

◆ WalLevel

enum WalLevel
Enumerator
WAL_LEVEL_MINIMAL 
WAL_LEVEL_REPLICA 
WAL_LEVEL_LOGICAL 

Definition at line 70 of file xlog.h.

71 {
75 } WalLevel;
WalLevel
Definition: xlog.h:71
@ WAL_LEVEL_REPLICA
Definition: xlog.h:73
@ WAL_LEVEL_LOGICAL
Definition: xlog.h:74
@ WAL_LEVEL_MINIMAL
Definition: xlog.h:72

◆ WalSyncMethod

Enumerator
WAL_SYNC_METHOD_FSYNC 
WAL_SYNC_METHOD_FDATASYNC 
WAL_SYNC_METHOD_OPEN 
WAL_SYNC_METHOD_FSYNC_WRITETHROUGH 
WAL_SYNC_METHOD_OPEN_DSYNC 

Definition at line 22 of file xlog.h.

23 {
26  WAL_SYNC_METHOD_OPEN, /* for O_SYNC */
28  WAL_SYNC_METHOD_OPEN_DSYNC /* for O_DSYNC */
WalSyncMethod
Definition: xlog.h:23
@ 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

Function Documentation

◆ BootStrapXLOG()

void BootStrapXLOG ( void  )

Definition at line 4920 of file xlog.c.

4921 {
4922  CheckPoint checkPoint;
4923  char *buffer;
4924  XLogPageHeader page;
4925  XLogLongPageHeader longpage;
4926  XLogRecord *record;
4927  char *recptr;
4928  uint64 sysidentifier;
4929  struct timeval tv;
4930  pg_crc32c crc;
4931 
4932  /* allow ordinary WAL segment creation, like StartupXLOG() would */
4934 
4935  /*
4936  * Select a hopefully-unique system identifier code for this installation.
4937  * We use the result of gettimeofday(), including the fractional seconds
4938  * field, as being about as unique as we can easily get. (Think not to
4939  * use random(), since it hasn't been seeded and there's no portable way
4940  * to seed it other than the system clock value...) The upper half of the
4941  * uint64 value is just the tv_sec part, while the lower half contains the
4942  * tv_usec part (which must fit in 20 bits), plus 12 bits from our current
4943  * PID for a little extra uniqueness. A person knowing this encoding can
4944  * determine the initialization time of the installation, which could
4945  * perhaps be useful sometimes.
4946  */
4947  gettimeofday(&tv, NULL);
4948  sysidentifier = ((uint64) tv.tv_sec) << 32;
4949  sysidentifier |= ((uint64) tv.tv_usec) << 12;
4950  sysidentifier |= getpid() & 0xFFF;
4951 
4952  /* page buffer must be aligned suitably for O_DIRECT */
4953  buffer = (char *) palloc(XLOG_BLCKSZ + XLOG_BLCKSZ);
4954  page = (XLogPageHeader) TYPEALIGN(XLOG_BLCKSZ, buffer);
4955  memset(page, 0, XLOG_BLCKSZ);
4956 
4957  /*
4958  * Set up information for the initial checkpoint record
4959  *
4960  * The initial checkpoint record is written to the beginning of the WAL
4961  * segment with logid=0 logseg=1. The very first WAL segment, 0/0, is not
4962  * used, so that we can use 0/0 to mean "before any valid WAL segment".
4963  */
4964  checkPoint.redo = wal_segment_size + SizeOfXLogLongPHD;
4965  checkPoint.ThisTimeLineID = BootstrapTimeLineID;
4966  checkPoint.PrevTimeLineID = BootstrapTimeLineID;
4967  checkPoint.fullPageWrites = fullPageWrites;
4968  checkPoint.nextXid =
4970  checkPoint.nextOid = FirstGenbkiObjectId;
4971  checkPoint.nextMulti = FirstMultiXactId;
4972  checkPoint.nextMultiOffset = 0;
4973  checkPoint.oldestXid = FirstNormalTransactionId;
4974  checkPoint.oldestXidDB = Template1DbOid;
4975  checkPoint.oldestMulti = FirstMultiXactId;
4976  checkPoint.oldestMultiDB = Template1DbOid;
4979  checkPoint.time = (pg_time_t) time(NULL);
4981 
4982  TransamVariables->nextXid = checkPoint.nextXid;
4983  TransamVariables->nextOid = checkPoint.nextOid;
4985  MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
4986  AdvanceOldestClogXid(checkPoint.oldestXid);
4987  SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
4988  SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
4990 
4991  /* Set up the XLOG page header */
4992  page->xlp_magic = XLOG_PAGE_MAGIC;
4993  page->xlp_info = XLP_LONG_HEADER;
4994  page->xlp_tli = BootstrapTimeLineID;
4996  longpage = (XLogLongPageHeader) page;
4997  longpage->xlp_sysid = sysidentifier;
4998  longpage->xlp_seg_size = wal_segment_size;
4999  longpage->xlp_xlog_blcksz = XLOG_BLCKSZ;
5000 
5001  /* Insert the initial checkpoint record */
5002  recptr = ((char *) page + SizeOfXLogLongPHD);
5003  record = (XLogRecord *) recptr;
5004  record->xl_prev = 0;
5005  record->xl_xid = InvalidTransactionId;
5006  record->xl_tot_len = SizeOfXLogRecord + SizeOfXLogRecordDataHeaderShort + sizeof(checkPoint);
5008  record->xl_rmid = RM_XLOG_ID;
5009  recptr += SizeOfXLogRecord;
5010  /* fill the XLogRecordDataHeaderShort struct */
5011  *(recptr++) = (char) XLR_BLOCK_ID_DATA_SHORT;
5012  *(recptr++) = sizeof(checkPoint);
5013  memcpy(recptr, &checkPoint, sizeof(checkPoint));
5014  recptr += sizeof(checkPoint);
5015  Assert(recptr - (char *) record == record->xl_tot_len);
5016 
5017  INIT_CRC32C(crc);
5018  COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
5019  COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
5020  FIN_CRC32C(crc);
5021  record->xl_crc = crc;
5022 
5023  /* Create first XLOG segment file */
5026 
5027  /*
5028  * We needn't bother with Reserve/ReleaseExternalFD here, since we'll
5029  * close the file again in a moment.
5030  */
5031 
5032  /* Write the first page with the initial record */
5033  errno = 0;
5034  pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_WRITE);
5035  if (write(openLogFile, page, XLOG_BLCKSZ) != XLOG_BLCKSZ)
5036  {
5037  /* if write didn't set errno, assume problem is no disk space */
5038  if (errno == 0)
5039  errno = ENOSPC;
5040  ereport(PANIC,
5042  errmsg("could not write bootstrap write-ahead log file: %m")));
5043  }
5045 
5046  pgstat_report_wait_start(WAIT_EVENT_WAL_BOOTSTRAP_SYNC);
5047  if (pg_fsync(openLogFile) != 0)
5048  ereport(PANIC,
5050  errmsg("could not fsync bootstrap write-ahead log file: %m")));
5052 
5053  if (close(openLogFile) != 0)
5054  ereport(PANIC,
5056  errmsg("could not close bootstrap write-ahead log file: %m")));
5057 
5058  openLogFile = -1;
5059 
5060  /* Now create pg_control */
5061  InitControlFile(sysidentifier);
5062  ControlFile->time = checkPoint.time;
5063  ControlFile->checkPoint = checkPoint.redo;
5064  ControlFile->checkPointCopy = checkPoint;
5065 
5066  /* some additional ControlFile fields are set in WriteControlFile() */
5067  WriteControlFile();
5068 
5069  /* Bootstrap the commit log, too */
5070  BootStrapCLOG();
5074 
5075  pfree(buffer);
5076 
5077  /*
5078  * Force control file to be read - in contrast to normal processing we'd
5079  * otherwise never run the checks and GUC related initializations therein.
5080  */
5081  ReadControlFile();
5082 }
#define TYPEALIGN(ALIGNVAL, LEN)
Definition: c.h:791
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
int errcode_for_file_access(void)
Definition: elog.c:882
int errmsg(const char *fmt,...)
Definition: elog.c:1072
#define PANIC
Definition: elog.h:42
#define ereport(elevel,...)
Definition: elog.h:149
int pg_fsync(int fd)
Definition: fd.c:386
#define close(a)
Definition: win32.h:12
#define write(a, b, c)
Definition: win32.h:14
Assert(fmt[strlen(fmt) - 1] !='\n')
void pfree(void *pointer)
Definition: mcxt.c:1508
void * palloc(Size size)
Definition: mcxt.c:1304
void MultiXactSetNextMXact(MultiXactId nextMulti, MultiXactOffset nextMultiOffset)
Definition: multixact.c:2253
void SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid, bool is_startup)
Definition: multixact.c:2287
void BootStrapMultiXact(void)
Definition: multixact.c:1959
#define FirstMultiXactId
Definition: multixact.h:25
#define XLOG_CHECKPOINT_SHUTDOWN
Definition: pg_control.h:67
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:50
MultiXactId oldestMulti
Definition: pg_control.h:49
MultiXactOffset nextMultiOffset
Definition: pg_control.h:46
TransactionId newestCommitTsXid
Definition: pg_control.h:54
TransactionId oldestXid
Definition: pg_control.h:47
TimeLineID PrevTimeLineID
Definition: pg_control.h:40
TimeLineID ThisTimeLineID
Definition: pg_control.h:39
Oid nextOid
Definition: pg_control.h:44
TransactionId oldestActiveXid
Definition: pg_control.h:63
bool fullPageWrites
Definition: pg_control.h:42
MultiXactId nextMulti
Definition: pg_control.h:45
FullTransactionId nextXid
Definition: pg_control.h:43
TransactionId oldestCommitTsXid
Definition: pg_control.h:52
pg_time_t time
Definition: pg_control.h:51
XLogRecPtr redo
Definition: pg_control.h:37
Oid oldestXidDB
Definition: pg_control.h:48
CheckPoint checkPointCopy
Definition: pg_control.h:134
pg_time_t time
Definition: pg_control.h:131
XLogRecPtr checkPoint
Definition: pg_control.h:132
FullTransactionId nextXid
Definition: transam.h:220
TimeLineID xlp_tli
Definition: xlog_internal.h:40
XLogRecPtr xlp_pageaddr
Definition: xlog_internal.h:41
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
static void pgstat_report_wait_start(uint32 wait_event_info)
Definition: wait_event.h:88
static void pgstat_report_wait_end(void)
Definition: wait_event.h:104
int gettimeofday(struct timeval *tp, void *tzp)
int XLogFileInit(XLogSegNo logsegno, TimeLineID logtli)
Definition: xlog.c:3310
bool fullPageWrites
Definition: xlog.c:122
static void InitControlFile(uint64 sysidentifier)
Definition: xlog.c:4131
void SetInstallXLogFileSegmentActive(void)
Definition: xlog.c:9317
static int openLogFile
Definition: xlog.c:624
static void WriteControlFile(void)
Definition: xlog.c:4166
int wal_segment_size
Definition: xlog.c:143
static TimeLineID openLogTLI
Definition: xlog.c:626
static ControlFileData * ControlFile
Definition: xlog.c:576
#define BootstrapTimeLineID
Definition: xlog.c:111
static void ReadControlFile(void)
Definition: xlog.c:4248
XLogLongPageHeaderData * XLogLongPageHeader
Definition: xlog_internal.h:71
XLogPageHeaderData * XLogPageHeader
Definition: xlog_internal.h:54
#define XLP_LONG_HEADER
Definition: xlog_internal.h:76
#define XLOG_PAGE_MAGIC
Definition: xlog_internal.h:34
#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_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().

◆ CheckXLogRemoved()

void CheckXLogRemoved ( XLogSegNo  segno,
TimeLineID  tli 
)

Definition at line 3657 of file xlog.c.

3658 {
3659  int save_errno = errno;
3660  XLogSegNo lastRemovedSegNo;
3661 
3663  lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3665 
3666  if (segno <= lastRemovedSegNo)
3667  {
3668  char filename[MAXFNAMELEN];
3669 
3670  XLogFileName(filename, tli, segno, wal_segment_size);
3671  errno = save_errno;
3672  ereport(ERROR,
3674  errmsg("requested WAL segment %s has already been removed",
3675  filename)));
3676  }
3677  errno = save_errno;
3678 }
#define ERROR
Definition: elog.h:39
static char * filename
Definition: pg_dumpall.c:121
#define SpinLockRelease(lock)
Definition: spin.h:64
#define SpinLockAcquire(lock)
Definition: spin.h:62
slock_t info_lck
Definition: xlog.c:555
XLogSegNo lastRemovedSegNo
Definition: xlog.c:467
static XLogCtlData * XLogCtl
Definition: xlog.c:568
#define MAXFNAMELEN
static void XLogFileName(char *fname, TimeLineID tli, XLogSegNo logSegNo, int wal_segsz_bytes)
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().

◆ CreateCheckPoint()

void CreateCheckPoint ( int  flags)

Definition at line 6734 of file xlog.c.

6735 {
6736  bool shutdown;
6737  CheckPoint checkPoint;
6738  XLogRecPtr recptr;
6739  XLogSegNo _logSegNo;
6741  uint32 freespace;
6742  XLogRecPtr PriorRedoPtr;
6743  XLogRecPtr last_important_lsn;
6744  VirtualTransactionId *vxids;
6745  int nvxids;
6746  int oldXLogAllowed = 0;
6747 
6748  /*
6749  * An end-of-recovery checkpoint is really a shutdown checkpoint, just
6750  * issued at a different time.
6751  */
6753  shutdown = true;
6754  else
6755  shutdown = false;
6756 
6757  /* sanity check */
6758  if (RecoveryInProgress() && (flags & CHECKPOINT_END_OF_RECOVERY) == 0)
6759  elog(ERROR, "can't create a checkpoint during recovery");
6760 
6761  /*
6762  * Prepare to accumulate statistics.
6763  *
6764  * Note: because it is possible for log_checkpoints to change while a
6765  * checkpoint proceeds, we always accumulate stats, even if
6766  * log_checkpoints is currently off.
6767  */
6768  MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
6770 
6771  /*
6772  * Let smgr prepare for checkpoint; this has to happen outside the
6773  * critical section and before we determine the REDO pointer. Note that
6774  * smgr must not do anything that'd have to be undone if we decide no
6775  * checkpoint is needed.
6776  */
6778 
6779  /*
6780  * Use a critical section to force system panic if we have trouble.
6781  */
6783 
6784  if (shutdown)
6785  {
6786  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6789  LWLockRelease(ControlFileLock);
6790  }
6791 
6792  /* Begin filling in the checkpoint WAL record */
6793  MemSet(&checkPoint, 0, sizeof(checkPoint));
6794  checkPoint.time = (pg_time_t) time(NULL);
6795 
6796  /*
6797  * For Hot Standby, derive the oldestActiveXid before we fix the redo
6798  * pointer. This allows us to begin accumulating changes to assemble our
6799  * starting snapshot of locks and transactions.
6800  */
6801  if (!shutdown && XLogStandbyInfoActive())
6803  else
6805 
6806  /*
6807  * Get location of last important record before acquiring insert locks (as
6808  * GetLastImportantRecPtr() also locks WAL locks).
6809  */
6810  last_important_lsn = GetLastImportantRecPtr();
6811 
6812  /*
6813  * If this isn't a shutdown or forced checkpoint, and if there has been no
6814  * WAL activity requiring a checkpoint, skip it. The idea here is to
6815  * avoid inserting duplicate checkpoints when the system is idle.
6816  */
6818  CHECKPOINT_FORCE)) == 0)
6819  {
6820  if (last_important_lsn == ControlFile->checkPoint)
6821  {
6822  END_CRIT_SECTION();
6823  ereport(DEBUG1,
6824  (errmsg_internal("checkpoint skipped because system is idle")));
6825  return;
6826  }
6827  }
6828 
6829  /*
6830  * An end-of-recovery checkpoint is created before anyone is allowed to
6831  * write WAL. To allow us to write the checkpoint record, temporarily
6832  * enable XLogInsertAllowed.
6833  */
6834  if (flags & CHECKPOINT_END_OF_RECOVERY)
6835  oldXLogAllowed = LocalSetXLogInsertAllowed();
6836 
6837  checkPoint.ThisTimeLineID = XLogCtl->InsertTimeLineID;
6838  if (flags & CHECKPOINT_END_OF_RECOVERY)
6839  checkPoint.PrevTimeLineID = XLogCtl->PrevTimeLineID;
6840  else
6841  checkPoint.PrevTimeLineID = checkPoint.ThisTimeLineID;
6842 
6843  /*
6844  * We must block concurrent insertions while examining insert state.
6845  */
6847 
6848  checkPoint.fullPageWrites = Insert->fullPageWrites;
6849 
6850  if (shutdown)
6851  {
6852  XLogRecPtr curInsert = XLogBytePosToRecPtr(Insert->CurrBytePos);
6853 
6854  /*
6855  * Compute new REDO record ptr = location of next XLOG record.
6856  *
6857  * Since this is a shutdown checkpoint, there can't be any concurrent
6858  * WAL insertion.
6859  */
6860  freespace = INSERT_FREESPACE(curInsert);
6861  if (freespace == 0)
6862  {
6863  if (XLogSegmentOffset(curInsert, wal_segment_size) == 0)
6864  curInsert += SizeOfXLogLongPHD;
6865  else
6866  curInsert += SizeOfXLogShortPHD;
6867  }
6868  checkPoint.redo = curInsert;
6869 
6870  /*
6871  * Here we update the shared RedoRecPtr for future XLogInsert calls;
6872  * this must be done while holding all the insertion locks.
6873  *
6874  * Note: if we fail to complete the checkpoint, RedoRecPtr will be
6875  * left pointing past where it really needs to point. This is okay;
6876  * the only consequence is that XLogInsert might back up whole buffers
6877  * that it didn't really need to. We can't postpone advancing
6878  * RedoRecPtr because XLogInserts that happen while we are dumping
6879  * buffers must assume that their buffer changes are not included in
6880  * the checkpoint.
6881  */
6882  RedoRecPtr = XLogCtl->Insert.RedoRecPtr = checkPoint.redo;
6883  }
6884 
6885  /*
6886  * Now we can release the WAL insertion locks, allowing other xacts to
6887  * proceed while we are flushing disk buffers.
6888  */
6890 
6891  /*
6892  * If this is an online checkpoint, we have not yet determined the redo
6893  * point. We do so now by inserting the special XLOG_CHECKPOINT_REDO
6894  * record; the LSN at which it starts becomes the new redo pointer. We
6895  * don't do this for a shutdown checkpoint, because in that case no WAL
6896  * can be written between the redo point and the insertion of the
6897  * checkpoint record itself, so the checkpoint record itself serves to
6898  * mark the redo point.
6899  */
6900  if (!shutdown)
6901  {
6902  int dummy = 0;
6903 
6904  /* Record must have payload to avoid assertion failure. */
6905  XLogBeginInsert();
6906  XLogRegisterData((char *) &dummy, sizeof(dummy));
6907  (void) XLogInsert(RM_XLOG_ID, XLOG_CHECKPOINT_REDO);
6908 
6909  /*
6910  * XLogInsertRecord will have updated XLogCtl->Insert.RedoRecPtr in
6911  * shared memory and RedoRecPtr in backend-local memory, but we need
6912  * to copy that into the record that will be inserted when the
6913  * checkpoint is complete.
6914  */
6915  checkPoint.redo = RedoRecPtr;
6916  }
6917 
6918  /* Update the info_lck-protected copy of RedoRecPtr as well */
6920  XLogCtl->RedoRecPtr = checkPoint.redo;
6922 
6923  /*
6924  * If enabled, log checkpoint start. We postpone this until now so as not
6925  * to log anything if we decided to skip the checkpoint.
6926  */
6927  if (log_checkpoints)
6928  LogCheckpointStart(flags, false);
6929 
6930  /* Update the process title */
6931  update_checkpoint_display(flags, false, false);
6932 
6933  TRACE_POSTGRESQL_CHECKPOINT_START(flags);
6934 
6935  /*
6936  * Get the other info we need for the checkpoint record.
6937  *
6938  * We don't need to save oldestClogXid in the checkpoint, it only matters
6939  * for the short period in which clog is being truncated, and if we crash
6940  * during that we'll redo the clog truncation and fix up oldestClogXid
6941  * there.
6942  */
6943  LWLockAcquire(XidGenLock, LW_SHARED);
6944  checkPoint.nextXid = TransamVariables->nextXid;
6945  checkPoint.oldestXid = TransamVariables->oldestXid;
6947  LWLockRelease(XidGenLock);
6948 
6949  LWLockAcquire(CommitTsLock, LW_SHARED);
6952  LWLockRelease(CommitTsLock);
6953 
6954  LWLockAcquire(OidGenLock, LW_SHARED);
6955  checkPoint.nextOid = TransamVariables->nextOid;
6956  if (!shutdown)
6957  checkPoint.nextOid += TransamVariables->oidCount;
6958  LWLockRelease(OidGenLock);
6959 
6960  MultiXactGetCheckptMulti(shutdown,
6961  &checkPoint.nextMulti,
6962  &checkPoint.nextMultiOffset,
6963  &checkPoint.oldestMulti,
6964  &checkPoint.oldestMultiDB);
6965 
6966  /*
6967  * Having constructed the checkpoint record, ensure all shmem disk buffers
6968  * and commit-log buffers are flushed to disk.
6969  *
6970  * This I/O could fail for various reasons. If so, we will fail to
6971  * complete the checkpoint, but there is no reason to force a system
6972  * panic. Accordingly, exit critical section while doing it.
6973  */
6974  END_CRIT_SECTION();
6975 
6976  /*
6977  * In some cases there are groups of actions that must all occur on one
6978  * side or the other of a checkpoint record. Before flushing the
6979  * checkpoint record we must explicitly wait for any backend currently
6980  * performing those groups of actions.
6981  *
6982  * One example is end of transaction, so we must wait for any transactions
6983  * that are currently in commit critical sections. If an xact inserted
6984  * its commit record into XLOG just before the REDO point, then a crash
6985  * restart from the REDO point would not replay that record, which means
6986  * that our flushing had better include the xact's update of pg_xact. So
6987  * we wait till he's out of his commit critical section before proceeding.
6988  * See notes in RecordTransactionCommit().
6989  *
6990  * Because we've already released the insertion locks, this test is a bit
6991  * fuzzy: it is possible that we will wait for xacts we didn't really need
6992  * to wait for. But the delay should be short and it seems better to make
6993  * checkpoint take a bit longer than to hold off insertions longer than
6994  * necessary. (In fact, the whole reason we have this issue is that xact.c
6995  * does commit record XLOG insertion and clog update as two separate steps
6996  * protected by different locks, but again that seems best on grounds of
6997  * minimizing lock contention.)
6998  *
6999  * A transaction that has not yet set delayChkptFlags when we look cannot
7000  * be at risk, since it has not inserted its commit record yet; and one
7001  * that's already cleared it is not at risk either, since it's done fixing
7002  * clog and we will correctly flush the update below. So we cannot miss
7003  * any xacts we need to wait for.
7004  */
7006  if (nvxids > 0)
7007  {
7008  do
7009  {
7010  pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_START);
7011  pg_usleep(10000L); /* wait for 10 msec */
7013  } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7015  }
7016  pfree(vxids);
7017 
7018  CheckPointGuts(checkPoint.redo, flags);
7019 
7021  if (nvxids > 0)
7022  {
7023  do
7024  {
7025  pgstat_report_wait_start(WAIT_EVENT_CHECKPOINT_DELAY_COMPLETE);
7026  pg_usleep(10000L); /* wait for 10 msec */
7028  } while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids,
7030  }
7031  pfree(vxids);
7032 
7033  /*
7034  * Take a snapshot of running transactions and write this to WAL. This
7035  * allows us to reconstruct the state of running transactions during
7036  * archive recovery, if required. Skip, if this info disabled.
7037  *
7038  * If we are shutting down, or Startup process is completing crash
7039  * recovery we don't need to write running xact data.
7040  */
7041  if (!shutdown && XLogStandbyInfoActive())
7043 
7045 
7046  /*
7047  * Now insert the checkpoint record into XLOG.
7048  */
7049  XLogBeginInsert();
7050  XLogRegisterData((char *) (&checkPoint), sizeof(checkPoint));
7051  recptr = XLogInsert(RM_XLOG_ID,
7052  shutdown ? XLOG_CHECKPOINT_SHUTDOWN :
7054 
7055  XLogFlush(recptr);
7056 
7057  /*
7058  * We mustn't write any new WAL after a shutdown checkpoint, or it will be
7059  * overwritten at next startup. No-one should even try, this just allows
7060  * sanity-checking. In the case of an end-of-recovery checkpoint, we want
7061  * to just temporarily disable writing until the system has exited
7062  * recovery.
7063  */
7064  if (shutdown)
7065  {
7066  if (flags & CHECKPOINT_END_OF_RECOVERY)
7067  LocalXLogInsertAllowed = oldXLogAllowed;
7068  else
7069  LocalXLogInsertAllowed = 0; /* never again write WAL */
7070  }
7071 
7072  /*
7073  * We now have ProcLastRecPtr = start of actual checkpoint record, recptr
7074  * = end of actual checkpoint record.
7075  */
7076  if (shutdown && checkPoint.redo != ProcLastRecPtr)
7077  ereport(PANIC,
7078  (errmsg("concurrent write-ahead log activity while database system is shutting down")));
7079 
7080  /*
7081  * Remember the prior checkpoint's redo ptr for
7082  * UpdateCheckPointDistanceEstimate()
7083  */
7084  PriorRedoPtr = ControlFile->checkPointCopy.redo;
7085 
7086  /*
7087  * Update the control file.
7088  */
7089  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7090  if (shutdown)
7093  ControlFile->checkPointCopy = checkPoint;
7094  /* crash recovery should always recover to the end of WAL */
7097 
7098  /*
7099  * Persist unloggedLSN value. It's reset on crash recovery, so this goes
7100  * unused on non-shutdown checkpoints, but seems useful to store it always
7101  * for debugging purposes.
7102  */
7104 
7106  LWLockRelease(ControlFileLock);
7107 
7108  /* Update shared-memory copy of checkpoint XID/epoch */
7110  XLogCtl->ckptFullXid = checkPoint.nextXid;
7112 
7113  /*
7114  * We are now done with critical updates; no need for system panic if we
7115  * have trouble while fooling with old log segments.
7116  */
7117  END_CRIT_SECTION();
7118 
7119  /*
7120  * WAL summaries end when the next XLOG_CHECKPOINT_REDO or
7121  * XLOG_CHECKPOINT_SHUTDOWN record is reached. This is the first point
7122  * where (a) we're not inside of a critical section and (b) we can be
7123  * certain that the relevant record has been flushed to disk, which must
7124  * happen before it can be summarized.
7125  *
7126  * If this is a shutdown checkpoint, then this happens reasonably
7127  * promptly: we've only just inserted and flushed the
7128  * XLOG_CHECKPOINT_SHUTDOWN record. If this is not a shutdown checkpoint,
7129  * then this might not be very prompt at all: the XLOG_CHECKPOINT_REDO
7130  * record was written before we began flushing data to disk, and that
7131  * could be many minutes ago at this point. However, we don't XLogFlush()
7132  * after inserting that record, so we're not guaranteed that it's on disk
7133  * until after the above call that flushes the XLOG_CHECKPOINT_ONLINE
7134  * record.
7135  */
7137 
7138  /*
7139  * Let smgr do post-checkpoint cleanup (eg, deleting old files).
7140  */
7142 
7143  /*
7144  * Update the average distance between checkpoints if the prior checkpoint
7145  * exists.
7146  */
7147  if (PriorRedoPtr != InvalidXLogRecPtr)
7149 
7150  /*
7151  * Delete old log files, those no longer needed for last checkpoint to
7152  * prevent the disk holding the xlog from growing full.
7153  */
7155  KeepLogSeg(recptr, &_logSegNo);
7157  _logSegNo, InvalidOid,
7159  {
7160  /*
7161  * Some slots have been invalidated; recalculate the old-segment
7162  * horizon, starting again from RedoRecPtr.
7163  */
7165  KeepLogSeg(recptr, &_logSegNo);
7166  }
7167  _logSegNo--;
7168  RemoveOldXlogFiles(_logSegNo, RedoRecPtr, recptr,
7169  checkPoint.ThisTimeLineID);
7170 
7171  /*
7172  * Make more log segments if needed. (Do this after recycling old log
7173  * segments, since that may supply some of the needed files.)
7174  */
7175  if (!shutdown)
7176  PreallocXlogFiles(recptr, checkPoint.ThisTimeLineID);
7177 
7178  /*
7179  * Truncate pg_subtrans if possible. We can throw away all data before
7180  * the oldest XMIN of any running transaction. No future transaction will
7181  * attempt to reference any pg_subtrans entry older than that (see Asserts
7182  * in subtrans.c). During recovery, though, we mustn't do this because
7183  * StartupSUBTRANS hasn't been called yet.
7184  */
7185  if (!RecoveryInProgress())
7187 
7188  /* Real work is done; log and update stats. */
7189  LogCheckpointEnd(false);
7190 
7191  /* Reset the process title */
7192  update_checkpoint_display(flags, false, true);
7193 
7194  TRACE_POSTGRESQL_CHECKPOINT_DONE(CheckpointStats.ckpt_bufs_written,
7195  NBuffers,
7199 }
static uint64 pg_atomic_read_membarrier_u64(volatile pg_atomic_uint64 *ptr)
Definition: atomics.h:471
TimestampTz GetCurrentTimestamp(void)
Definition: timestamp.c:1654
unsigned int uint32
Definition: c.h:493
#define MemSet(start, val, len)
Definition: c.h:1007
int errmsg_internal(const char *fmt,...)
Definition: elog.c:1159
#define DEBUG1
Definition: elog.h:30
#define elog(elevel,...)
Definition: elog.h:224
static void Insert(File file)
Definition: fd.c:1313
int NBuffers
Definition: globals.c:139
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1172
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1785
@ LW_SHARED
Definition: lwlock.h:117
@ LW_EXCLUSIVE
Definition: lwlock.h:116
#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:2207
#define XLOG_CHECKPOINT_REDO
Definition: pg_control.h:81
@ DB_SHUTDOWNING
Definition: pg_control.h:93
@ DB_SHUTDOWNED
Definition: pg_control.h:91
#define XLOG_CHECKPOINT_ONLINE
Definition: pg_control.h:68
#define InvalidOid
Definition: postgres_ext.h:36
#define DELAY_CHKPT_START
Definition: proc.h:114
#define DELAY_CHKPT_COMPLETE
Definition: proc.h:115
VirtualTransactionId * GetVirtualXIDsDelayingChkpt(int *nvxids, int type)
Definition: procarray.c:3030
TransactionId GetOldestActiveTransactionId(void)
Definition: procarray.c:2867
TransactionId GetOldestTransactionIdConsideredRunning(void)
Definition: procarray.c:2022
bool HaveVirtualXIDsDelayingChkpt(VirtualTransactionId *vxids, int nvxids, int type)
Definition: procarray.c:3076
void pg_usleep(long microsec)
Definition: signal.c:53
bool InvalidateObsoleteReplicationSlots(ReplicationSlotInvalidationCause cause, XLogSegNo oldestSegno, Oid dboid, TransactionId snapshotConflictHorizon)
Definition: slot.c:1759
@ RS_INVAL_WAL_REMOVED
Definition: slot.h:51
XLogRecPtr LogStandbySnapshot(void)
Definition: standby.c:1285
TimestampTz ckpt_start_t
Definition: xlog.h:159
int ckpt_segs_removed
Definition: xlog.h:168
int ckpt_segs_added
Definition: xlog.h:167
int ckpt_bufs_written
Definition: xlog.h:165
int ckpt_segs_recycled
Definition: xlog.h:169
XLogRecPtr minRecoveryPoint
Definition: pg_control.h:167
XLogRecPtr unloggedLSN
Definition: pg_control.h:136
TimeLineID minRecoveryPointTLI
Definition: pg_control.h:168
TransactionId oldestCommitTsXid
Definition: transam.h:232
TransactionId newestCommitTsXid
Definition: transam.h:233
TransactionId oldestXid
Definition: transam.h:222
FullTransactionId ckptFullXid
Definition: xlog.c:463
TimeLineID InsertTimeLineID
Definition: xlog.c:511
XLogRecPtr RedoRecPtr
Definition: xlog.c:462
XLogCtlInsert Insert
Definition: xlog.c:458
TimeLineID PrevTimeLineID
Definition: xlog.c:512
pg_atomic_uint64 unloggedLSN
Definition: xlog.c:470
XLogRecPtr RedoRecPtr
Definition: xlog.c:436
void TruncateSUBTRANS(TransactionId oldestXact)
Definition: subtrans.c:411
void SyncPreCheckpoint(void)
Definition: sync.c:177
void SyncPostCheckpoint(void)
Definition: sync.c:202
void SetWalSummarizerLatch(void)
XLogRecPtr ProcLastRecPtr
Definition: xlog.c:253
bool RecoveryInProgress(void)
Definition: xlog.c:6201
static void WALInsertLockRelease(void)
Definition: xlog.c:1433
static XLogRecPtr XLogBytePosToRecPtr(uint64 bytepos)
Definition: xlog.c:1817
static void WALInsertLockAcquireExclusive(void)
Definition: xlog.c:1404
static void UpdateControlFile(void)
Definition: xlog.c:4449
static void RemoveOldXlogFiles(XLogSegNo segno, XLogRecPtr lastredoptr, XLogRecPtr endptr, TimeLineID insertTLI)
Definition: xlog.c:3795
static void LogCheckpointStart(int flags, bool restartpoint)
Definition: xlog.c:6499
static XLogRecPtr RedoRecPtr
Definition: xlog.c:273
static void LogCheckpointEnd(bool restartpoint)
Definition: xlog.c:6531
static void PreallocXlogFiles(XLogRecPtr endptr, TimeLineID tli)
Definition: xlog.c:3620
bool log_checkpoints
Definition: xlog.c:129
static void KeepLogSeg(XLogRecPtr recptr, XLogSegNo *logSegNo)
Definition: xlog.c:7787
static int LocalSetXLogInsertAllowed(void)
Definition: xlog.c:6289
XLogRecPtr GetLastImportantRecPtr(void)
Definition: xlog.c:6406
static void UpdateCheckPointDistanceEstimate(uint64 nbytes)
Definition: xlog.c:6634
#define INSERT_FREESPACE(endptr)
Definition: xlog.c:582
static int LocalXLogInsertAllowed
Definition: xlog.c:236
CheckpointStatsData CheckpointStats
Definition: xlog.c:209
void XLogFlush(XLogRecPtr record)
Definition: xlog.c:2728
static void CheckPointGuts(XLogRecPtr checkPointRedo, int flags)
Definition: xlog.c:7344
static void update_checkpoint_display(int flags, bool restartpoint, bool reset)
Definition: xlog.c:6672
#define CHECKPOINT_END_OF_RECOVERY
Definition: xlog.h:138
#define CHECKPOINT_FORCE
Definition: xlog.h:140
#define CHECKPOINT_IS_SHUTDOWN
Definition: xlog.h:137
#define XLogStandbyInfoActive()
Definition: xlog.h:121
#define XLogSegmentOffset(xlogptr, wal_segsz_bytes)
#define XLByteToSeg(xlrp, logSegNo, wal_segsz_bytes)
#define SizeOfXLogShortPHD
Definition: xlog_internal.h:52
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
void XLogRegisterData(char *data, uint32 len)
Definition: xloginsert.c:364
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:474
void XLogBeginInsert(void)
Definition: xloginsert.c:149

References 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, SetWalSummarizerLatch(), 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(), 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().

◆ CreateRestartPoint()

bool CreateRestartPoint ( int  flags)

Definition at line 7425 of file xlog.c.

7426 {
7427  XLogRecPtr lastCheckPointRecPtr;
7428  XLogRecPtr lastCheckPointEndPtr;
7429  CheckPoint lastCheckPoint;
7430  XLogRecPtr PriorRedoPtr;
7431  XLogRecPtr receivePtr;
7432  XLogRecPtr replayPtr;
7433  TimeLineID replayTLI;
7434  XLogRecPtr endptr;
7435  XLogSegNo _logSegNo;
7436  TimestampTz xtime;
7437 
7438  /* Concurrent checkpoint/restartpoint cannot happen */
7440 
7441  /* Get a local copy of the last safe checkpoint record. */
7443  lastCheckPointRecPtr = XLogCtl->lastCheckPointRecPtr;
7444  lastCheckPointEndPtr = XLogCtl->lastCheckPointEndPtr;
7445  lastCheckPoint = XLogCtl->lastCheckPoint;
7447 
7448  /*
7449  * Check that we're still in recovery mode. It's ok if we exit recovery
7450  * mode after this check, the restart point is valid anyway.
7451  */
7452  if (!RecoveryInProgress())
7453  {
7454  ereport(DEBUG2,
7455  (errmsg_internal("skipping restartpoint, recovery has already ended")));
7456  return false;
7457  }
7458 
7459  /*
7460  * If the last checkpoint record we've replayed is already our last
7461  * restartpoint, we can't perform a new restart point. We still update
7462  * minRecoveryPoint in that case, so that if this is a shutdown restart
7463  * point, we won't start up earlier than before. That's not strictly
7464  * necessary, but when hot standby is enabled, it would be rather weird if
7465  * the database opened up for read-only connections at a point-in-time
7466  * before the last shutdown. Such time travel is still possible in case of
7467  * immediate shutdown, though.
7468  *
7469  * We don't explicitly advance minRecoveryPoint when we do create a
7470  * restartpoint. It's assumed that flushing the buffers will do that as a
7471  * side-effect.
7472  */
7473  if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
7474  lastCheckPoint.redo <= ControlFile->checkPointCopy.redo)
7475  {
7476  ereport(DEBUG2,
7477  (errmsg_internal("skipping restartpoint, already performed at %X/%X",
7478  LSN_FORMAT_ARGS(lastCheckPoint.redo))));
7479 
7481  if (flags & CHECKPOINT_IS_SHUTDOWN)
7482  {
7483  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7486  LWLockRelease(ControlFileLock);
7487  }
7488  return false;
7489  }
7490 
7491  /*
7492  * Update the shared RedoRecPtr so that the startup process can calculate
7493  * the number of segments replayed since last restartpoint, and request a
7494  * restartpoint if it exceeds CheckPointSegments.
7495  *
7496  * Like in CreateCheckPoint(), hold off insertions to update it, although
7497  * during recovery this is just pro forma, because no WAL insertions are
7498  * happening.
7499  */
7501  RedoRecPtr = XLogCtl->Insert.RedoRecPtr = lastCheckPoint.redo;
7503 
7504  /* Also update the info_lck-protected copy */
7506  XLogCtl->RedoRecPtr = lastCheckPoint.redo;
7508 
7509  /*
7510  * Prepare to accumulate statistics.
7511  *
7512  * Note: because it is possible for log_checkpoints to change while a
7513  * checkpoint proceeds, we always accumulate stats, even if
7514  * log_checkpoints is currently off.
7515  */
7516  MemSet(&CheckpointStats, 0, sizeof(CheckpointStats));
7518 
7519  if (log_checkpoints)
7520  LogCheckpointStart(flags, true);
7521 
7522  /* Update the process title */
7523  update_checkpoint_display(flags, true, false);
7524 
7525  CheckPointGuts(lastCheckPoint.redo, flags);
7526 
7527  /*
7528  * This location needs to be after CheckPointGuts() to ensure that some
7529  * work has already happened during this checkpoint.
7530  */
7531  INJECTION_POINT("create-restart-point");
7532 
7533  /*
7534  * Remember the prior checkpoint's redo ptr for
7535  * UpdateCheckPointDistanceEstimate()
7536  */
7537  PriorRedoPtr = ControlFile->checkPointCopy.redo;
7538 
7539  /*
7540  * Update pg_control, using current time. Check that it still shows an
7541  * older checkpoint, else do nothing; this is a quick hack to make sure
7542  * nothing really bad happens if somehow we get here after the
7543  * end-of-recovery checkpoint.
7544  */
7545  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
7546  if (ControlFile->checkPointCopy.redo < lastCheckPoint.redo)
7547  {
7548  /*
7549  * Update the checkpoint information. We do this even if the cluster
7550  * does not show DB_IN_ARCHIVE_RECOVERY to match with the set of WAL
7551  * segments recycled below.
7552  */
7553  ControlFile->checkPoint = lastCheckPointRecPtr;
7554  ControlFile->checkPointCopy = lastCheckPoint;
7555 
7556  /*
7557  * Ensure minRecoveryPoint is past the checkpoint record and update it
7558  * if the control file still shows DB_IN_ARCHIVE_RECOVERY. Normally,
7559  * this will have happened already while writing out dirty buffers,
7560  * but not necessarily - e.g. because no buffers were dirtied. We do
7561  * this because a backup performed in recovery uses minRecoveryPoint
7562  * to determine which WAL files must be included in the backup, and
7563  * the file (or files) containing the checkpoint record must be
7564  * included, at a minimum. Note that for an ordinary restart of
7565  * recovery there's no value in having the minimum recovery point any
7566  * earlier than this anyway, because redo will begin just after the
7567  * checkpoint record.
7568  */
7570  {
7571  if (ControlFile->minRecoveryPoint < lastCheckPointEndPtr)
7572  {
7573  ControlFile->minRecoveryPoint = lastCheckPointEndPtr;
7575 
7576  /* update local copy */
7579  }
7580  if (flags & CHECKPOINT_IS_SHUTDOWN)
7582  }
7584  }
7585  LWLockRelease(ControlFileLock);
7586 
7587  /*
7588  * Update the average distance between checkpoints/restartpoints if the
7589  * prior checkpoint exists.
7590  */
7591  if (PriorRedoPtr != InvalidXLogRecPtr)
7593 
7594  /*
7595  * Delete old log files, those no longer needed for last restartpoint to
7596  * prevent the disk holding the xlog from growing full.
7597  */
7599 
7600  /*
7601  * Retreat _logSegNo using the current end of xlog replayed or received,
7602  * whichever is later.
7603  */
7604  receivePtr = GetWalRcvFlushRecPtr(NULL, NULL);
7605  replayPtr = GetXLogReplayRecPtr(&replayTLI);
7606  endptr = (receivePtr < replayPtr) ? replayPtr : receivePtr;
7607  KeepLogSeg(endptr, &_logSegNo);
7609  _logSegNo, InvalidOid,
7611  {
7612  /*
7613  * Some slots have been invalidated; recalculate the old-segment
7614  * horizon, starting again from RedoRecPtr.
7615  */
7617  KeepLogSeg(endptr, &_logSegNo);
7618  }
7619  _logSegNo--;
7620 
7621  /*
7622  * Try to recycle segments on a useful timeline. If we've been promoted
7623  * since the beginning of this restartpoint, use the new timeline chosen
7624  * at end of recovery. If we're still in recovery, use the timeline we're
7625  * currently replaying.
7626  *
7627  * There is no guarantee that the WAL segments will be useful on the
7628  * current timeline; if recovery proceeds to a new timeline right after
7629  * this, the pre-allocated WAL segments on this timeline will not be used,
7630  * and will go wasted until recycled on the next restartpoint. We'll live
7631  * with that.
7632  */
7633  if (!RecoveryInProgress())
7634  replayTLI = XLogCtl->InsertTimeLineID;
7635 
7636  RemoveOldXlogFiles(_logSegNo, RedoRecPtr, endptr, replayTLI);
7637 
7638  /*
7639  * Make more log segments if needed. (Do this after recycling old log
7640  * segments, since that may supply some of the needed files.)
7641  */
7642  PreallocXlogFiles(endptr, replayTLI);
7643 
7644  /*
7645  * Truncate pg_subtrans if possible. We can throw away all data before
7646  * the oldest XMIN of any running transaction. No future transaction will
7647  * attempt to reference any pg_subtrans entry older than that (see Asserts
7648  * in subtrans.c). When hot standby is disabled, though, we mustn't do
7649  * this because StartupSUBTRANS hasn't been called yet.
7650  */
7651  if (EnableHotStandby)
7653 
7654  /* Real work is done; log and update stats. */
7655  LogCheckpointEnd(true);
7656 
7657  /* Reset the process title */
7658  update_checkpoint_display(flags, true, true);
7659 
7660  xtime = GetLatestXTime();
7662  (errmsg("recovery restart point at %X/%X",
7663  LSN_FORMAT_ARGS(lastCheckPoint.redo)),
7664  xtime ? errdetail("Last completed transaction was at log time %s.",
7665  timestamptz_to_str(xtime)) : 0));
7666 
7667  /*
7668  * Finally, execute archive_cleanup_command, if any.
7669  */
7670  if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
7672  "archive_cleanup_command",
7673  false,
7674  WAIT_EVENT_ARCHIVE_CLEANUP_COMMAND);
7675 
7676  return true;
7677 }
const char * timestamptz_to_str(TimestampTz t)
Definition: timestamp.c:1853
int64 TimestampTz
Definition: timestamp.h:39
int errdetail(const char *fmt,...)
Definition: elog.c:1205
#define LOG
Definition: elog.h:31
#define DEBUG2
Definition: elog.h:29
bool IsUnderPostmaster
Definition: globals.c:117
#define INJECTION_POINT(name)
@ B_CHECKPOINTER
Definition: miscadmin.h:354
BackendType MyBackendType
Definition: miscinit.c:63
@ DB_IN_ARCHIVE_RECOVERY
Definition: pg_control.h:95
@ DB_SHUTDOWNED_IN_RECOVERY
Definition: pg_control.h:92
CheckPoint lastCheckPoint
Definition: xlog.c:547
XLogRecPtr lastCheckPointRecPtr
Definition: xlog.c:545
XLogRecPtr lastCheckPointEndPtr
Definition: xlog.c:546
XLogRecPtr GetWalRcvFlushRecPtr(XLogRecPtr *latestChunkStart, TimeLineID *receiveTLI)
bool EnableHotStandby
Definition: xlog.c:121
static void UpdateMinRecoveryPoint(XLogRecPtr lsn, bool force)
Definition: xlog.c:2648
static XLogRecPtr LocalMinRecoveryPoint
Definition: xlog.c:635
static TimeLineID LocalMinRecoveryPointTLI
Definition: xlog.c:636
void ExecuteRecoveryCommand(const char *command, const char *commandName, bool failOnSignal, uint32 wait_event_info)
Definition: xlogarchive.c:295
#define LSN_FORMAT_ARGS(lsn)
Definition: xlogdefs.h:43
#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 9225 of file xlog.c.

9226 {
9227  bool during_backup_start = DatumGetBool(arg);
9228 
9229  /* If called during backup start, there shouldn't be one already running */
9230  Assert(!during_backup_start || sessionBackupState == SESSION_BACKUP_NONE);
9231 
9232  if (during_backup_start || sessionBackupState != SESSION_BACKUP_NONE)
9233  {
9237 
9240 
9241  if (!during_backup_start)
9242  ereport(WARNING,
9243  errmsg("aborting backup due to backend exiting before pg_backup_stop was called"));
9244  }
9245 }
#define WARNING
Definition: elog.h:36
void * arg
static bool DatumGetBool(Datum X)
Definition: postgres.h:90
int runningBackups
Definition: xlog.c:444
static SessionBackupState sessionBackupState
Definition: xlog.c:397

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 8623 of file xlog.c.

8625 {
8627 
8628  Assert(state != NULL);
8630 
8631  /*
8632  * During recovery, we don't need to check WAL level. Because, if WAL
8633  * level is not sufficient, it's impossible to get here during recovery.
8634  */
8636  ereport(ERROR,
8637  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8638  errmsg("WAL level not sufficient for making an online backup"),
8639  errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
8640 
8641  if (strlen(backupidstr) > MAXPGPATH)
8642  ereport(ERROR,
8643  (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
8644  errmsg("backup label too long (max %d bytes)",
8645  MAXPGPATH)));
8646 
8647  memcpy(state->name, backupidstr, strlen(backupidstr));
8648 
8649  /*
8650  * Mark backup active in shared memory. We must do full-page WAL writes
8651  * during an on-line backup even if not doing so at other times, because
8652  * it's quite possible for the backup dump to obtain a "torn" (partially
8653  * written) copy of a database page if it reads the page concurrently with
8654  * our write to the same page. This can be fixed as long as the first
8655  * write to the page in the WAL sequence is a full-page write. Hence, we
8656  * increment runningBackups then force a CHECKPOINT, to ensure there are
8657  * no dirty pages in shared memory that might get dumped while the backup
8658  * is in progress without having a corresponding WAL record. (Once the
8659  * backup is complete, we need not force full-page writes anymore, since
8660  * we expect that any pages not modified during the backup interval must
8661  * have been correctly captured by the backup.)
8662  *
8663  * Note that forcing full-page writes has no effect during an online
8664  * backup from the standby.
8665  *
8666  * We must hold all the insertion locks to change the value of
8667  * runningBackups, to ensure adequate interlocking against
8668  * XLogInsertRecord().
8669  */
8673 
8674  /*
8675  * Ensure we decrement runningBackups if we fail below. NB -- for this to
8676  * work correctly, it is critical that sessionBackupState is only updated
8677  * after this block is over.
8678  */
8680  {
8681  bool gotUniqueStartpoint = false;
8682  DIR *tblspcdir;
8683  struct dirent *de;
8684  tablespaceinfo *ti;
8685  int datadirpathlen;
8686 
8687  /*
8688  * Force an XLOG file switch before the checkpoint, to ensure that the
8689  * WAL segment the checkpoint is written to doesn't contain pages with
8690  * old timeline IDs. That would otherwise happen if you called
8691  * pg_backup_start() right after restoring from a PITR archive: the
8692  * first WAL segment containing the startup checkpoint has pages in
8693  * the beginning with the old timeline ID. That can cause trouble at
8694  * recovery: we won't have a history file covering the old timeline if
8695  * pg_wal directory was not included in the base backup and the WAL
8696  * archive was cleared too before starting the backup.
8697  *
8698  * This also ensures that we have emitted a WAL page header that has
8699  * XLP_BKP_REMOVABLE off before we emit the checkpoint record.
8700  * Therefore, if a WAL archiver (such as pglesslog) is trying to
8701  * compress out removable backup blocks, it won't remove any that
8702  * occur after this point.
8703  *
8704  * During recovery, we skip forcing XLOG file switch, which means that
8705  * the backup taken during recovery is not available for the special
8706  * recovery case described above.
8707  */
8709  RequestXLogSwitch(false);
8710 
8711  do
8712  {
8713  bool checkpointfpw;
8714 
8715  /*
8716  * Force a CHECKPOINT. Aside from being necessary to prevent torn
8717  * page problems, this guarantees that two successive backup runs
8718  * will have different checkpoint positions and hence different
8719  * history file names, even if nothing happened in between.
8720  *
8721  * During recovery, establish a restartpoint if possible. We use
8722  * the last restartpoint as the backup starting checkpoint. This
8723  * means that two successive backup runs can have same checkpoint
8724  * positions.
8725  *
8726  * Since the fact that we are executing do_pg_backup_start()
8727  * during recovery means that checkpointer is running, we can use
8728  * RequestCheckpoint() to establish a restartpoint.
8729  *
8730  * We use CHECKPOINT_IMMEDIATE only if requested by user (via
8731  * passing fast = true). Otherwise this can take awhile.
8732  */
8734  (fast ? CHECKPOINT_IMMEDIATE : 0));
8735 
8736  /*
8737  * Now we need to fetch the checkpoint record location, and also
8738  * its REDO pointer. The oldest point in WAL that would be needed
8739  * to restore starting from the checkpoint is precisely the REDO
8740  * pointer.
8741  */
8742  LWLockAcquire(ControlFileLock, LW_SHARED);
8743  state->checkpointloc = ControlFile->checkPoint;
8744  state->startpoint = ControlFile->checkPointCopy.redo;
8746  checkpointfpw = ControlFile->checkPointCopy.fullPageWrites;
8747  LWLockRelease(ControlFileLock);
8748 
8750  {
8751  XLogRecPtr recptr;
8752 
8753  /*
8754  * Check to see if all WAL replayed during online backup
8755  * (i.e., since last restartpoint used as backup starting
8756  * checkpoint) contain full-page writes.
8757  */
8759  recptr = XLogCtl->lastFpwDisableRecPtr;
8761 
8762  if (!checkpointfpw || state->startpoint <= recptr)
8763  ereport(ERROR,
8764  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8765  errmsg("WAL generated with full_page_writes=off was replayed "
8766  "since last restartpoint"),
8767  errhint("This means that the backup being taken on the standby "
8768  "is corrupt and should not be used. "
8769  "Enable full_page_writes and run CHECKPOINT on the primary, "
8770  "and then try an online backup again.")));
8771 
8772  /*
8773  * During recovery, since we don't use the end-of-backup WAL
8774  * record and don't write the backup history file, the
8775  * starting WAL location doesn't need to be unique. This means
8776  * that two base backups started at the same time might use
8777  * the same checkpoint as starting locations.
8778  */
8779  gotUniqueStartpoint = true;
8780  }
8781 
8782  /*
8783  * If two base backups are started at the same time (in WAL sender
8784  * processes), we need to make sure that they use different
8785  * checkpoints as starting locations, because we use the starting
8786  * WAL location as a unique identifier for the base backup in the
8787  * end-of-backup WAL record and when we write the backup history
8788  * file. Perhaps it would be better generate a separate unique ID
8789  * for each backup instead of forcing another checkpoint, but
8790  * taking a checkpoint right after another is not that expensive
8791  * either because only few buffers have been dirtied yet.
8792  */
8794  if (XLogCtl->Insert.lastBackupStart < state->startpoint)
8795  {
8796  XLogCtl->Insert.lastBackupStart = state->startpoint;
8797  gotUniqueStartpoint = true;
8798  }
8800  } while (!gotUniqueStartpoint);
8801 
8802  /*
8803  * Construct tablespace_map file.
8804  */
8805  datadirpathlen = strlen(DataDir);
8806 
8807  /* Collect information about all tablespaces */
8808  tblspcdir = AllocateDir("pg_tblspc");
8809  while ((de = ReadDir(tblspcdir, "pg_tblspc")) != NULL)
8810  {
8811  char fullpath[MAXPGPATH + 10];
8812  char linkpath[MAXPGPATH];
8813  char *relpath = NULL;
8814  char *s;
8815  PGFileType de_type;
8816  char *badp;
8817  Oid tsoid;
8818 
8819  /*
8820  * Try to parse the directory name as an unsigned integer.
8821  *
8822  * Tablespace directories should be positive integers that can be
8823  * represented in 32 bits, with no leading zeroes or trailing
8824  * garbage. If we come across a name that doesn't meet those
8825  * criteria, skip it.
8826  */
8827  if (de->d_name[0] < '1' || de->d_name[1] > '9')
8828  continue;
8829  errno = 0;
8830  tsoid = strtoul(de->d_name, &badp, 10);
8831  if (*badp != '\0' || errno == EINVAL || errno == ERANGE)
8832  continue;
8833 
8834  snprintf(fullpath, sizeof(fullpath), "pg_tblspc/%s", de->d_name);
8835 
8836  de_type = get_dirent_type(fullpath, de, false, ERROR);
8837 
8838  if (de_type == PGFILETYPE_LNK)
8839  {
8840  StringInfoData escapedpath;
8841  int rllen;
8842 
8843  rllen = readlink(fullpath, linkpath, sizeof(linkpath));
8844  if (rllen < 0)
8845  {
8846  ereport(WARNING,
8847  (errmsg("could not read symbolic link \"%s\": %m",
8848  fullpath)));
8849  continue;
8850  }
8851  else if (rllen >= sizeof(linkpath))
8852  {
8853  ereport(WARNING,
8854  (errmsg("symbolic link \"%s\" target is too long",
8855  fullpath)));
8856  continue;
8857  }
8858  linkpath[rllen] = '\0';
8859 
8860  /*
8861  * Relpath holds the relative path of the tablespace directory
8862  * when it's located within PGDATA, or NULL if it's located
8863  * elsewhere.
8864  */
8865  if (rllen > datadirpathlen &&
8866  strncmp(linkpath, DataDir, datadirpathlen) == 0 &&
8867  IS_DIR_SEP(linkpath[datadirpathlen]))
8868  relpath = pstrdup(linkpath + datadirpathlen + 1);
8869 
8870  /*
8871  * Add a backslash-escaped version of the link path to the
8872  * tablespace map file.
8873  */
8874  initStringInfo(&escapedpath);
8875  for (s = linkpath; *s; s++)
8876  {
8877  if (*s == '\n' || *s == '\r' || *s == '\\')
8878  appendStringInfoChar(&escapedpath, '\\');
8879  appendStringInfoChar(&escapedpath, *s);
8880  }
8881  appendStringInfo(tblspcmapfile, "%s %s\n",
8882  de->d_name, escapedpath.data);
8883  pfree(escapedpath.data);
8884  }
8885  else if (de_type == PGFILETYPE_DIR)
8886  {
8887  /*
8888  * It's possible to use allow_in_place_tablespaces to create
8889  * directories directly under pg_tblspc, for testing purposes
8890  * only.
8891  *
8892  * In this case, we store a relative path rather than an
8893  * absolute path into the tablespaceinfo.
8894  */
8895  snprintf(linkpath, sizeof(linkpath), "pg_tblspc/%s",
8896  de->d_name);
8897  relpath = pstrdup(linkpath);
8898  }
8899  else
8900  {
8901  /* Skip any other file type that appears here. */
8902  continue;
8903  }
8904 
8905  ti = palloc(sizeof(tablespaceinfo));
8906  ti->oid = tsoid;
8907  ti->path = pstrdup(linkpath);
8908  ti->rpath = relpath;
8909  ti->size = -1;
8910 
8911  if (tablespaces)
8912  *tablespaces = lappend(*tablespaces, ti);
8913  }
8914  FreeDir(tblspcdir);
8915 
8916  state->starttime = (pg_time_t) time(NULL);
8917  }
8919 
8920  state->started_in_recovery = backup_started_in_recovery;
8921 
8922  /*
8923  * Mark that the start phase has correctly finished for the backup.
8924  */
8926 }
static bool backup_started_in_recovery
Definition: basebackup.c:123
void RequestCheckpoint(int flags)
Definition: checkpointer.c:941
int errhint(const char *fmt,...)
Definition: elog.c:1319
int errcode(int sqlerrcode)
Definition: elog.c:859
struct dirent * ReadDir(DIR *dir, const char *dirname)
Definition: fd.c:2909
int FreeDir(DIR *dir)
Definition: fd.c:2961
DIR * AllocateDir(const char *dirname)
Definition: fd.c:2843
PGFileType get_dirent_type(const char *path, const struct dirent *de, bool look_through_symlinks, int elevel)
Definition: file_utils.c:525
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:68
#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
char * pstrdup(const char *in)
Definition: mcxt.c:1683
#define MAXPGPATH
#define snprintf
Definition: port.h:238
#define IS_DIR_SEP(ch)
Definition: port.h:102
unsigned int Oid
Definition: postgres_ext.h:31
#define relpath(rlocator, forknum)
Definition: relpath.h:94
void appendStringInfo(StringInfo str, const char *fmt,...)
Definition: stringinfo.c:97
void appendStringInfoChar(StringInfo str, char ch)
Definition: stringinfo.c:194
void initStringInfo(StringInfo str)
Definition: stringinfo.c:59
Definition: dirent.c:26
XLogRecPtr lastFpwDisableRecPtr
Definition: xlog.c:553
XLogRecPtr lastBackupStart
Definition: xlog.c:445
Definition: dirent.h:10
char d_name[MAX_PATH]
Definition: dirent.h:15
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:7894
void do_pg_abort_backup(int code, Datum arg)
Definition: xlog.c:9225
@ SESSION_BACKUP_RUNNING
Definition: xlog.h:284
#define CHECKPOINT_WAIT
Definition: xlog.h:143
#define CHECKPOINT_IMMEDIATE
Definition: xlog.h:139
#define XLogIsNeeded()
Definition: xlog.h:107

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, 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, 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 8951 of file xlog.c.

8952 {
8953  bool backup_stopped_in_recovery = false;
8954  char histfilepath[MAXPGPATH];
8955  char lastxlogfilename[MAXFNAMELEN];
8956  char histfilename[MAXFNAMELEN];
8957  XLogSegNo _logSegNo;
8958  FILE *fp;
8959  int seconds_before_warning;
8960  int waits = 0;
8961  bool reported_waiting = false;
8962 
8963  Assert(state != NULL);
8964 
8965  backup_stopped_in_recovery = RecoveryInProgress();
8966 
8967  /*
8968  * During recovery, we don't need to check WAL level. Because, if WAL
8969  * level is not sufficient, it's impossible to get here during recovery.
8970  */
8971  if (!backup_stopped_in_recovery && !XLogIsNeeded())
8972  ereport(ERROR,
8973  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
8974  errmsg("WAL level not sufficient for making an online backup"),
8975  errhint("wal_level must be set to \"replica\" or \"logical\" at server start.")));
8976 
8977  /*
8978  * OK to update backup counter and session-level lock.
8979  *
8980  * Note that CHECK_FOR_INTERRUPTS() must not occur while updating them,
8981  * otherwise they can be updated inconsistently, which might cause
8982  * do_pg_abort_backup() to fail.
8983  */
8985 
8986  /*
8987  * It is expected that each do_pg_backup_start() call is matched by
8988  * exactly one do_pg_backup_stop() call.
8989  */
8992 
8993  /*
8994  * Clean up session-level lock.
8995  *
8996  * You might think that WALInsertLockRelease() can be called before
8997  * cleaning up session-level lock because session-level lock doesn't need
8998  * to be protected with WAL insertion lock. But since
8999  * CHECK_FOR_INTERRUPTS() can occur in it, session-level lock must be
9000  * cleaned up before it.
9001  */
9003 
9005 
9006  /*
9007  * If we are taking an online backup from the standby, we confirm that the
9008  * standby has not been promoted during the backup.
9009  */
9010  if (state->started_in_recovery && !backup_stopped_in_recovery)
9011  ereport(ERROR,
9012  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9013  errmsg("the standby was promoted during online backup"),
9014  errhint("This means that the backup being taken is corrupt "
9015  "and should not be used. "
9016  "Try taking another online backup.")));
9017 
9018  /*
9019  * During recovery, we don't write an end-of-backup record. We assume that
9020  * pg_control was backed up last and its minimum recovery point can be
9021  * available as the backup end location. Since we don't have an
9022  * end-of-backup record, we use the pg_control value to check whether
9023  * we've reached the end of backup when starting recovery from this
9024  * backup. We have no way of checking if pg_control wasn't backed up last
9025  * however.
9026  *
9027  * We don't force a switch to new WAL file but it is still possible to
9028  * wait for all the required files to be archived if waitforarchive is
9029  * true. This is okay if we use the backup to start a standby and fetch
9030  * the missing WAL using streaming replication. But in the case of an
9031  * archive recovery, a user should set waitforarchive to true and wait for
9032  * them to be archived to ensure that all the required files are
9033  * available.
9034  *
9035  * We return the current minimum recovery point as the backup end
9036  * location. Note that it can be greater than the exact backup end
9037  * location if the minimum recovery point is updated after the backup of
9038  * pg_control. This is harmless for current uses.
9039  *
9040  * XXX currently a backup history file is for informational and debug
9041  * purposes only. It's not essential for an online backup. Furthermore,
9042  * even if it's created, it will not be archived during recovery because
9043  * an archiver is not invoked. So it doesn't seem worthwhile to write a
9044  * backup history file during recovery.
9045  */
9046  if (backup_stopped_in_recovery)
9047  {
9048  XLogRecPtr recptr;
9049 
9050  /*
9051  * Check to see if all WAL replayed during online backup contain
9052  * full-page writes.
9053  */
9055  recptr = XLogCtl->lastFpwDisableRecPtr;
9057 
9058  if (state->startpoint <= recptr)
9059  ereport(ERROR,
9060  (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
9061  errmsg("WAL generated with full_page_writes=off was replayed "
9062  "during online backup"),
9063  errhint("This means that the backup being taken on the standby "
9064  "is corrupt and should not be used. "
9065  "Enable full_page_writes and run CHECKPOINT on the primary, "
9066  "and then try an online backup again.")));
9067 
9068 
9069  LWLockAcquire(ControlFileLock, LW_SHARED);
9070  state->stoppoint = ControlFile->minRecoveryPoint;
9072  LWLockRelease(ControlFileLock);
9073  }
9074  else
9075  {
9076  char *history_file;
9077 
9078  /*
9079  * Write the backup-end xlog record
9080  */
9081  XLogBeginInsert();
9082  XLogRegisterData((char *) (&state->startpoint),
9083  sizeof(state->startpoint));
9084  state->stoppoint = XLogInsert(RM_XLOG_ID, XLOG_BACKUP_END);
9085 
9086  /*
9087  * Given that we're not in recovery, InsertTimeLineID is set and can't
9088  * change, so we can read it without a lock.
9089  */
9090  state->stoptli = XLogCtl->InsertTimeLineID;
9091 
9092  /*
9093  * Force a switch to a new xlog segment file, so that the backup is
9094  * valid as soon as archiver moves out the current segment file.
9095  */
9096  RequestXLogSwitch(false);
9097 
9098  state->stoptime = (pg_time_t) time(NULL);
9099 
9100  /*
9101  * Write the backup history file
9102  */
9103  XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9104  BackupHistoryFilePath(histfilepath, state->stoptli, _logSegNo,
9105  state->startpoint, wal_segment_size);
9106  fp = AllocateFile(histfilepath, "w");
9107  if (!fp)
9108  ereport(ERROR,
9110  errmsg("could not create file \"%s\": %m",
9111  histfilepath)));
9112 
9113  /* Build and save the contents of the backup history file */
9114  history_file = build_backup_content(state, true);
9115  fprintf(fp, "%s", history_file);
9116  pfree(history_file);
9117 
9118  if (fflush(fp) || ferror(fp) || FreeFile(fp))
9119  ereport(ERROR,
9121  errmsg("could not write file \"%s\": %m",
9122  histfilepath)));
9123 
9124  /*
9125  * Clean out any no-longer-needed history files. As a side effect,
9126  * this will post a .ready file for the newly created history file,
9127  * notifying the archiver that history file may be archived
9128  * immediately.
9129  */
9131  }
9132 
9133  /*
9134  * If archiving is enabled, wait for all the required WAL files to be
9135  * archived before returning. If archiving isn't enabled, the required WAL
9136  * needs to be transported via streaming replication (hopefully with
9137  * wal_keep_size set high enough), or some more exotic mechanism like
9138  * polling and copying files from pg_wal with script. We have no knowledge
9139  * of those mechanisms, so it's up to the user to ensure that he gets all
9140  * the required WAL.
9141  *
9142  * We wait until both the last WAL file filled during backup and the
9143  * history file have been archived, and assume that the alphabetic sorting
9144  * property of the WAL files ensures any earlier WAL files are safely
9145  * archived as well.
9146  *
9147  * We wait forever, since archive_command is supposed to work and we
9148  * assume the admin wanted his backup to work completely. If you don't
9149  * wish to wait, then either waitforarchive should be passed in as false,
9150  * or you can set statement_timeout. Also, some notices are issued to
9151  * clue in anyone who might be doing this interactively.
9152  */
9153 
9154  if (waitforarchive &&
9155  ((!backup_stopped_in_recovery && XLogArchivingActive()) ||
9156  (backup_stopped_in_recovery && XLogArchivingAlways())))
9157  {
9158  XLByteToPrevSeg(state->stoppoint, _logSegNo, wal_segment_size);
9159  XLogFileName(lastxlogfilename, state->stoptli, _logSegNo,
9161 
9162  XLByteToSeg(state->startpoint, _logSegNo, wal_segment_size);
9163  BackupHistoryFileName(histfilename, state->stoptli, _logSegNo,
9164  state->startpoint, wal_segment_size);
9165 
9166  seconds_before_warning = 60;
9167  waits = 0;
9168 
9169  while (XLogArchiveIsBusy(lastxlogfilename) ||
9170  XLogArchiveIsBusy(histfilename))
9171  {
9173 
9174  if (!reported_waiting && waits > 5)
9175  {
9176  ereport(NOTICE,
9177  (errmsg("base backup done, waiting for required WAL segments to be archived")));
9178  reported_waiting = true;
9179  }
9180 
9181  (void) WaitLatch(MyLatch,
9183  1000L,
9184  WAIT_EVENT_BACKUP_WAIT_WAL_ARCHIVE);
9186 
9187  if (++waits >= seconds_before_warning)
9188  {
9189  seconds_before_warning *= 2; /* This wraps in >10 years... */
9190  ereport(WARNING,
9191  (errmsg("still waiting for all required WAL segments to be archived (%d seconds elapsed)",
9192  waits),
9193  errhint("Check that your archive_command is executing properly. "
9194  "You can safely cancel this backup, "
9195  "but the database backup will not be usable without all the WAL segments.")));
9196  }
9197  }
9198 
9199  ereport(NOTICE,
9200  (errmsg("all required WAL segments have been archived")));
9201  }
9202  else if (waitforarchive)
9203  ereport(NOTICE,
9204  (errmsg("WAL archiving is not enabled; you must ensure that all required WAL segments are copied through other means to complete the backup")));
9205 }
#define NOTICE
Definition: elog.h:35
FILE * AllocateFile(const char *name, const char *mode)
Definition: fd.c:2583
int FreeFile(FILE *file)
Definition: fd.c:2781
struct Latch * MyLatch
Definition: globals.c:60
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:72
#define fprintf
Definition: port.h:242
static void CleanupBackupHistory(void)
Definition: xlog.c:4088
#define XLogArchivingActive()
Definition: xlog.h:97
#define XLogArchivingAlways()
Definition: xlog.h:100
#define XLByteToPrevSeg(xlrp, logSegNo, wal_segsz_bytes)
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 8932 of file xlog.c.

8933 {
8934  return sessionBackupState;
8935 }

References sessionBackupState.

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

◆ GetActiveWalLevelOnStandby()

WalLevel GetActiveWalLevelOnStandby ( void  )

Definition at line 4749 of file xlog.c.

4750 {
4751  return ControlFile->wal_level;
4752 }

References ControlFile, and ControlFileData::wal_level.

Referenced by CheckLogicalDecodingRequirements().

◆ GetFakeLSNForUnloggedRel()

XLogRecPtr GetFakeLSNForUnloggedRel ( void  )

Definition at line 4494 of file xlog.c.

4495 {
4497 }
static uint64 pg_atomic_fetch_add_u64(volatile pg_atomic_uint64 *ptr, int64 add_)
Definition: atomics.h:518

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

Referenced by gistGetFakeLSN().

◆ GetFlushRecPtr()

XLogRecPtr GetFlushRecPtr ( TimeLineID insertTLI)

Definition at line 6366 of file xlog.c.

6367 {
6369 
6373 
6374  /*
6375  * If we're writing and flushing WAL, the time line can't be changing, so
6376  * no lock is required.
6377  */
6378  if (insertTLI)
6379  *insertTLI = XLogCtl->InsertTimeLineID;
6380 
6381  return LogwrtResult.Flush;
6382 }
RecoveryState SharedRecoveryState
Definition: xlog.c:518
XLogwrtResult LogwrtResult
Definition: xlog.c:480
XLogRecPtr Flush
Definition: xlog.c:334
static XLogwrtResult LogwrtResult
Definition: xlog.c:614

References Assert(), XLogwrtResult::Flush, XLogCtlData::info_lck, XLogCtlData::InsertTimeLineID, XLogCtlData::LogwrtResult, LogwrtResult, RECOVERY_STATE_DONE, XLogCtlData::SharedRecoveryState, SpinLockAcquire, SpinLockRelease, 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 6334 of file xlog.c.

6335 {
6336  *RedoRecPtr_p = RedoRecPtr;
6337  *doPageWrites_p = doPageWrites;
6338 }
static bool doPageWrites
Definition: xlog.c:286

References doPageWrites, and RedoRecPtr.

Referenced by XLogCheckBufferNeedsBackup(), and XLogInsert().

◆ GetInsertRecPtr()

XLogRecPtr GetInsertRecPtr ( void  )

Definition at line 6349 of file xlog.c.

6350 {
6351  XLogRecPtr recptr;
6352 
6354  recptr = XLogCtl->LogwrtRqst.Write;
6356 
6357  return recptr;
6358 }
XLogwrtRqst LogwrtRqst
Definition: xlog.c:461
XLogRecPtr Write
Definition: xlog.c:327

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 6406 of file xlog.c.

6407 {
6409  int i;
6410 
6411  for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
6412  {
6413  XLogRecPtr last_important;
6414 
6415  /*
6416  * Need to take a lock to prevent torn reads of the LSN, which are
6417  * possible on some of the supported platforms. WAL insert locks only
6418  * support exclusive mode, so we have to use that.
6419  */
6421  last_important = WALInsertLocks[i].l.lastImportantAt;
6422  LWLockRelease(&WALInsertLocks[i].l.lock);
6423 
6424  if (res < last_important)
6425  res = last_important;
6426  }
6427 
6428  return res;
6429 }
int i
Definition: isn.c:73
XLogRecPtr lastImportantAt
Definition: xlog.c:377
WALInsertLock l
Definition: xlog.c:389
static WALInsertLockPadded * WALInsertLocks
Definition: xlog.c:571
#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().

◆ GetMockAuthenticationNonce()

char* GetMockAuthenticationNonce ( void  )

Definition at line 4468 of file xlog.c.

4469 {
4470  Assert(ControlFile != NULL);
4472 }
char mock_authentication_nonce[MOCK_AUTH_NONCE_LEN]
Definition: pg_control.h:228

References Assert(), ControlFile, and ControlFileData::mock_authentication_nonce.

Referenced by scram_mock_salt().

◆ GetRecoveryState()

RecoveryState GetRecoveryState ( void  )

Definition at line 6237 of file xlog.c.

6238 {
6239  RecoveryState retval;
6240 
6242  retval = XLogCtl->SharedRecoveryState;
6244 
6245  return retval;
6246 }

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

Referenced by XLogArchiveCheckDone().

◆ GetRedoRecPtr()

XLogRecPtr GetRedoRecPtr ( void  )

Definition at line 6304 of file xlog.c.

6305 {
6306  XLogRecPtr ptr;
6307 
6308  /*
6309  * The possibly not up-to-date copy in XlogCtl is enough. Even if we
6310  * grabbed a WAL insertion lock to read the authoritative value in
6311  * Insert->RedoRecPtr, someone might update it just after we've released
6312  * the lock.
6313  */
6315  ptr = XLogCtl->RedoRecPtr;
6317 
6318  if (RedoRecPtr < ptr)
6319  RedoRecPtr = ptr;
6320 
6321  return RedoRecPtr;
6322 }

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

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

◆ GetSystemIdentifier()

uint64 GetSystemIdentifier ( void  )

◆ GetWALAvailability()

WALAvailability GetWALAvailability ( XLogRecPtr  targetLSN)

Definition at line 7703 of file xlog.c.

7704 {
7705  XLogRecPtr currpos; /* current write LSN */
7706  XLogSegNo currSeg; /* segid of currpos */
7707  XLogSegNo targetSeg; /* segid of targetLSN */
7708  XLogSegNo oldestSeg; /* actual oldest segid */
7709  XLogSegNo oldestSegMaxWalSize; /* oldest segid kept by max_wal_size */
7710  XLogSegNo oldestSlotSeg; /* oldest segid kept by slot */
7711  uint64 keepSegs;
7712 
7713  /*
7714  * slot does not reserve WAL. Either deactivated, or has never been active
7715  */
7716  if (XLogRecPtrIsInvalid(targetLSN))
7717  return WALAVAIL_INVALID_LSN;
7718 
7719  /*
7720  * Calculate the oldest segment currently reserved by all slots,
7721  * considering wal_keep_size and max_slot_wal_keep_size. Initialize
7722  * oldestSlotSeg to the current segment.
7723  */
7724  currpos = GetXLogWriteRecPtr();
7725  XLByteToSeg(currpos, oldestSlotSeg, wal_segment_size);
7726  KeepLogSeg(currpos, &oldestSlotSeg);
7727 
7728  /*
7729  * Find the oldest extant segment file. We get 1 until checkpoint removes
7730  * the first WAL segment file since startup, which causes the status being
7731  * wrong under certain abnormal conditions but that doesn't actually harm.
7732  */
7733  oldestSeg = XLogGetLastRemovedSegno() + 1;
7734 
7735  /* calculate oldest segment by max_wal_size */
7736  XLByteToSeg(currpos, currSeg, wal_segment_size);
7738 
7739  if (currSeg > keepSegs)
7740  oldestSegMaxWalSize = currSeg - keepSegs;
7741  else
7742  oldestSegMaxWalSize = 1;
7743 
7744  /* the segment we care about */
7745  XLByteToSeg(targetLSN, targetSeg, wal_segment_size);
7746 
7747  /*
7748  * No point in returning reserved or extended status values if the
7749  * targetSeg is known to be lost.
7750  */
7751  if (targetSeg >= oldestSlotSeg)
7752  {
7753  /* show "reserved" when targetSeg is within max_wal_size */
7754  if (targetSeg >= oldestSegMaxWalSize)
7755  return WALAVAIL_RESERVED;
7756 
7757  /* being retained by slots exceeding max_wal_size */
7758  return WALAVAIL_EXTENDED;
7759  }
7760 
7761  /* WAL segments are no longer retained but haven't been removed yet */
7762  if (targetSeg >= oldestSeg)
7763  return WALAVAIL_UNRESERVED;
7764 
7765  /* Definitely lost */
7766  return WALAVAIL_REMOVED;
7767 }
XLogSegNo XLogGetLastRemovedSegno(void)
Definition: xlog.c:3688
int max_wal_size_mb
Definition: xlog.c:114
#define ConvertToXSegs(x, segsize)
Definition: xlog.c:605
XLogRecPtr GetXLogWriteRecPtr(void)
Definition: xlog.c:9282
@ WALAVAIL_REMOVED
Definition: xlog.h:191

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 6389 of file xlog.c.

6390 {
6392 
6393  /* Since the value can't be changing, no lock is required. */
6394  return XLogCtl->InsertTimeLineID;
6395 }

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().

◆ GetXLogInsertRecPtr()

XLogRecPtr GetXLogInsertRecPtr ( void  )

Definition at line 9266 of file xlog.c.

9267 {
9269  uint64 current_bytepos;
9270 
9271  SpinLockAcquire(&Insert->insertpos_lck);
9272  current_bytepos = Insert->CurrBytePos;
9273  SpinLockRelease(&Insert->insertpos_lck);
9274 
9275  return XLogBytePosToRecPtr(current_bytepos);
9276 }

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()

◆ InitializeWalConsistencyChecking()

void InitializeWalConsistencyChecking ( void  )

Definition at line 4674 of file xlog.c.

4675 {
4677 
4679  {
4680  struct config_generic *guc;
4681 
4682  guc = find_option("wal_consistency_checking", false, false, ERROR);
4683 
4685 
4686  set_config_option_ext("wal_consistency_checking",
4688  guc->scontext, guc->source, guc->srole,
4689  GUC_ACTION_SET, true, ERROR, false);
4690 
4691  /* checking should not be deferred again */
4693  }
4694 }
struct config_generic * find_option(const char *name, bool create_placeholders, bool skip_errors, int elevel)
Definition: guc.c:1237
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:3373
@ GUC_ACTION_SET
Definition: guc.h:199
bool process_shared_preload_libraries_done
Definition: miscinit.c:1779
GucContext scontext
Definition: guc_tables.h:167
GucSource source
Definition: guc_tables.h:165
char * wal_consistency_checking_string
Definition: xlog.c:125
static bool check_wal_consistency_checking_deferred
Definition: xlog.c:166

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().

◆ IsInstallXLogFileSegmentActive()

bool IsInstallXLogFileSegmentActive ( void  )

Definition at line 9325 of file xlog.c.

9326 {
9327  bool result;
9328 
9329  LWLockAcquire(ControlFileLock, LW_SHARED);
9331  LWLockRelease(ControlFileLock);
9332 
9333  return result;
9334 }
bool InstallXLogFileSegmentActive
Definition: xlog.c:528

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 8516 of file xlog.c.

8517 {
8518  char *msg = NULL;
8519  instr_time start;
8520 
8521  Assert(tli != 0);
8522 
8523  /*
8524  * Quick exit if fsync is disabled or write() has already synced the WAL
8525  * file.
8526  */
8527  if (!enableFsync ||
8530  return;
8531 
8532  /* Measure I/O timing to sync the WAL file */
8533  if (track_wal_io_timing)
8534  INSTR_TIME_SET_CURRENT(start);
8535  else
8536  INSTR_TIME_SET_ZERO(start);
8537 
8538  pgstat_report_wait_start(WAIT_EVENT_WAL_SYNC);
8539  switch (wal_sync_method)
8540  {
8541  case WAL_SYNC_METHOD_FSYNC:
8542  if (pg_fsync_no_writethrough(fd) != 0)
8543  msg = _("could not fsync file \"%s\": %m");
8544  break;
8545 #ifdef HAVE_FSYNC_WRITETHROUGH
8547  if (pg_fsync_writethrough(fd) != 0)
8548  msg = _("could not fsync write-through file \"%s\": %m");
8549  break;
8550 #endif
8552  if (pg_fdatasync(fd) != 0)
8553  msg = _("could not fdatasync file \"%s\": %m");
8554  break;
8555  case WAL_SYNC_METHOD_OPEN:
8557  /* not reachable */
8558  Assert(false);
8559  break;
8560  default:
8561  elog(PANIC, "unrecognized wal_sync_method: %d", wal_sync_method);
8562  break;
8563  }
8564 
8565  /* PANIC if failed to fsync */
8566  if (msg)
8567  {
8568  char xlogfname[MAXFNAMELEN];
8569  int save_errno = errno;
8570 
8571  XLogFileName(xlogfname, tli, segno, wal_segment_size);
8572  errno = save_errno;
8573  ereport(PANIC,
8575  errmsg(msg, xlogfname)));
8576  }
8577 
8579 
8580  /*
8581  * Increment the I/O timing and the number of times WAL files were synced.
8582  */
8583  if (track_wal_io_timing)
8584  {
8585  instr_time end;
8586 
8589  }
8590 
8592 }
#define _(x)
Definition: elog.c:90
int pg_fsync_no_writethrough(int fd)
Definition: fd.c:441
int pg_fdatasync(int fd)
Definition: fd.c:480
int pg_fsync_writethrough(int fd)
Definition: fd.c:461
bool enableFsync
Definition: globals.c:126
#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
PgStat_PendingWalStats PendingWalStats
Definition: pgstat_wal.c:24
static int fd(const char *x, int i)
Definition: preproc-init.c:105
instr_time wal_sync_time
Definition: pgstat.h:456
PgStat_Counter wal_sync
Definition: pgstat.h:454
int wal_sync_method
Definition: xlog.c:130
bool track_wal_io_timing
Definition: xlog.c:137

References _, Assert(), elog, enableFsync, ereport, errcode_for_file_access(), errmsg(), 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(), 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().

◆ LocalProcessControlFile()

void LocalProcessControlFile ( bool  reset)

Definition at line 4736 of file xlog.c.

4737 {
4738  Assert(reset || ControlFile == NULL);
4739  ControlFile = palloc(sizeof(ControlFileData));
4740  ReadControlFile();
4741 }
void reset(void)
Definition: sql-declare.c:600

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

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

◆ ReachedEndOfBackup()

void ReachedEndOfBackup ( XLogRecPtr  EndRecPtr,
TimeLineID  tli 
)

Definition at line 6114 of file xlog.c.

6115 {
6116  /*
6117  * We have reached the end of base backup, as indicated by pg_control. The
6118  * data on disk is now consistent (unless minRecoveryPoint is further
6119  * ahead, which can happen if we crashed during previous recovery). Reset
6120  * backupStartPoint and backupEndPoint, and update minRecoveryPoint to
6121  * make sure we don't allow starting up at an earlier point even if
6122  * recovery is stopped and restarted soon after this.
6123  */
6124  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6125 
6126  if (ControlFile->minRecoveryPoint < EndRecPtr)
6127  {
6128  ControlFile->minRecoveryPoint = EndRecPtr;
6130  }
6131 
6134  ControlFile->backupEndRequired = false;
6136 
6137  LWLockRelease(ControlFileLock);
6138 }
XLogRecPtr backupStartPoint
Definition: pg_control.h:169
bool backupEndRequired
Definition: pg_control.h:171
XLogRecPtr backupEndPoint
Definition: pg_control.h:170

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

Referenced by CheckRecoveryConsistency().

◆ RecoveryInProgress()

bool RecoveryInProgress ( void  )

Definition at line 6201 of file xlog.c.

6202 {
6203  /*
6204  * We check shared state each time only until we leave recovery mode. We
6205  * can't re-enter recovery, so there's no need to keep checking after the
6206  * shared variable has once been seen false.
6207  */
6209  return false;
6210  else
6211  {
6212  /*
6213  * use volatile pointer to make sure we make a fresh read of the
6214  * shared variable.
6215  */
6216  volatile XLogCtlData *xlogctl = XLogCtl;
6217 
6219 
6220  /*
6221  * Note: We don't need a memory barrier when we're still in recovery.
6222  * We might exit recovery immediately after return, so the caller
6223  * can't rely on 'true' meaning that we're still in recovery anyway.
6224  */
6225 
6226  return LocalRecoveryInProgress;
6227  }
6228 }
static bool LocalRecoveryInProgress
Definition: xlog.c:224

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

Referenced by 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_create_restore_point(), pg_current_wal_flush_lsn(), pg_current_wal_insert_lsn(), pg_current_wal_lsn(), 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_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(), 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().

◆ register_persistent_abort_backup_handler()

void register_persistent_abort_backup_handler ( void  )

Definition at line 9252 of file xlog.c.

9253 {
9254  static bool already_done = false;
9255 
9256  if (already_done)
9257  return;
9259  already_done = true;
9260 }
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 3870 of file xlog.c.

3871 {
3872  DIR *xldir;
3873  struct dirent *xlde;
3874  char switchseg[MAXFNAMELEN];
3875  XLogSegNo endLogSegNo;
3876  XLogSegNo switchLogSegNo;
3877  XLogSegNo recycleSegNo;
3878 
3879  /*
3880  * Initialize info about where to begin the work. This will recycle,
3881  * somewhat arbitrarily, 10 future segments.
3882  */
3883  XLByteToPrevSeg(switchpoint, switchLogSegNo, wal_segment_size);
3884  XLByteToSeg(switchpoint, endLogSegNo, wal_segment_size);
3885  recycleSegNo = endLogSegNo + 10;
3886 
3887  /*
3888  * Construct a filename of the last segment to be kept.
3889  */
3890  XLogFileName(switchseg, newTLI, switchLogSegNo, wal_segment_size);
3891 
3892  elog(DEBUG2, "attempting to remove WAL segments newer than log file %s",
3893  switchseg);
3894 
3895  xldir = AllocateDir(XLOGDIR);
3896 
3897  while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3898  {
3899  /* Ignore files that are not XLOG segments */
3900  if (!IsXLogFileName(xlde->d_name))
3901  continue;
3902 
3903  /*
3904  * Remove files that are on a timeline older than the new one we're
3905  * switching to, but with a segment number >= the first segment on the
3906  * new timeline.
3907  */
3908  if (strncmp(xlde->d_name, switchseg, 8) < 0 &&
3909  strcmp(xlde->d_name + 8, switchseg + 8) > 0)
3910  {
3911  /*
3912  * If the file has already been marked as .ready, however, don't
3913  * remove it yet. It should be OK to remove it - files that are
3914  * not part of our timeline history are not required for recovery
3915  * - but seems safer to let them be archived and removed later.
3916  */
3917  if (!XLogArchiveIsReady(xlde->d_name))
3918  RemoveXlogFile(xlde, recycleSegNo, &endLogSegNo, newTLI);
3919  }
3920  }
3921 
3922  FreeDir(xldir);
3923 }
static void RemoveXlogFile(const struct dirent *segment_de, XLogSegNo recycleSegNo, XLogSegNo *endlogSegNo, TimeLineID insertTLI)
Definition: xlog.c:3939
static bool IsXLogFileName(const char *fname)
#define XLOGDIR
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().

◆ SetInstallXLogFileSegmentActive()

void SetInstallXLogFileSegmentActive ( void  )

Definition at line 9317 of file xlog.c.

9318 {
9319  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9321  LWLockRelease(ControlFileLock);
9322 }

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

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

◆ SetWalWriterSleeping()

void SetWalWriterSleeping ( bool  sleeping)

Definition at line 9340 of file xlog.c.

9341 {
9343  XLogCtl->WalWriterSleeping = sleeping;
9345 }
bool WalWriterSleeping
Definition: xlog.c:535

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

Referenced by WalWriterMain().

◆ ShutdownXLOG()

void ShutdownXLOG ( int  code,
Datum  arg 
)

Definition at line 6452 of file xlog.c.

6453 {
6454  /*
6455  * We should have an aux process resource owner to use, and we should not
6456  * be in a transaction that's installed some other resowner.
6457  */
6459  Assert(CurrentResourceOwner == NULL ||
6462 
6463  /* Don't be chatty in standalone mode */
6465  (errmsg("shutting down")));
6466 
6467  /*
6468  * Signal walsenders to move to stopping state.
6469  */
6471 
6472  /*
6473  * Wait for WAL senders to be in stopping state. This prevents commands
6474  * from writing new WAL.
6475  */
6477 
6478  if (RecoveryInProgress())
6480  else
6481  {
6482  /*
6483  * If archiving is enabled, rotate the last XLOG file so that all the
6484  * remaining records are archived (postmaster wakes up the archiver
6485  * process one more time at the end of shutdown). The checkpoint
6486  * record will go to the next XLOG file and won't be archived (yet).
6487  */
6488  if (XLogArchivingActive())
6489  RequestXLogSwitch(false);
6490 
6492  }
6493 }
bool IsPostmasterEnvironment
Definition: globals.c:116
ResourceOwner CurrentResourceOwner
Definition: resowner.c:165
ResourceOwner AuxProcessResourceOwner
Definition: resowner.c:168
void WalSndInitStopping(void)
Definition: walsender.c:3745
void WalSndWaitStopping(void)
Definition: walsender.c:3771
bool CreateRestartPoint(int flags)
Definition: xlog.c:7425
void CreateCheckPoint(int flags)
Definition: xlog.c:6734

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 5307 of file xlog.c.

5308 {
5310  CheckPoint checkPoint;
5311  bool wasShutdown;
5312  bool didCrash;
5313  bool haveTblspcMap;
5314  bool haveBackupLabel;
5315  XLogRecPtr EndOfLog;
5316  TimeLineID EndOfLogTLI;
5317  TimeLineID newTLI;
5318  bool performedWalRecovery;
5319  EndOfWalRecoveryInfo *endOfRecoveryInfo;
5322  TransactionId oldestActiveXID;
5323  bool promoted = false;
5324 
5325  /*
5326  * We should have an aux process resource owner to use, and we should not
5327  * be in a transaction that's installed some other resowner.
5328  */
5330  Assert(CurrentResourceOwner == NULL ||
5333 
5334  /*
5335  * Check that contents look valid.
5336  */
5338  ereport(FATAL,
5339  (errmsg("control file contains invalid checkpoint location")));
5340 
5341  switch (ControlFile->state)
5342  {
5343  case DB_SHUTDOWNED:
5344 
5345  /*
5346  * This is the expected case, so don't be chatty in standalone
5347  * mode
5348  */
5350  (errmsg("database system was shut down at %s",
5351  str_time(ControlFile->time))));
5352  break;
5353 
5355  ereport(LOG,
5356  (errmsg("database system was shut down in recovery at %s",
5357  str_time(ControlFile->time))));
5358  break;
5359 
5360  case DB_SHUTDOWNING:
5361  ereport(LOG,
5362  (errmsg("database system shutdown was interrupted; last known up at %s",
5363  str_time(ControlFile->time))));
5364  break;
5365 
5366  case DB_IN_CRASH_RECOVERY:
5367  ereport(LOG,
5368  (errmsg("database system was interrupted while in recovery at %s",
5370  errhint("This probably means that some data is corrupted and"
5371  " you will have to use the last backup for recovery.")));
5372  break;
5373 
5375  ereport(LOG,
5376  (errmsg("database system was interrupted while in recovery at log time %s",
5378  errhint("If this has occurred more than once some data might be corrupted"
5379  " and you might need to choose an earlier recovery target.")));
5380  break;
5381 
5382  case DB_IN_PRODUCTION:
5383  ereport(LOG,
5384  (errmsg("database system was interrupted; last known up at %s",
5385  str_time(ControlFile->time))));
5386  break;
5387 
5388  default:
5389  ereport(FATAL,
5390  (errmsg("control file contains invalid database cluster state")));
5391  }
5392 
5393  /* This is just to allow attaching to startup process with a debugger */
5394 #ifdef XLOG_REPLAY_DELAY
5396  pg_usleep(60000000L);
5397 #endif
5398 
5399  /*
5400  * Verify that pg_wal, pg_wal/archive_status, and pg_wal/summaries exist.
5401  * In cases where someone has performed a copy for PITR, these directories
5402  * may have been excluded and need to be re-created.
5403  */
5405 
5406  /* Set up timeout handler needed to report startup progress. */
5410 
5411  /*----------
5412  * If we previously crashed, perform a couple of actions:
5413  *
5414  * - The pg_wal directory may still include some temporary WAL segments
5415  * used when creating a new segment, so perform some clean up to not
5416  * bloat this path. This is done first as there is no point to sync
5417  * this temporary data.
5418  *
5419  * - There might be data which we had written, intending to fsync it, but
5420  * which we had not actually fsync'd yet. Therefore, a power failure in
5421  * the near future might cause earlier unflushed writes to be lost, even
5422  * though more recent data written to disk from here on would be
5423  * persisted. To avoid that, fsync the entire data directory.
5424  */
5425  if (ControlFile->state != DB_SHUTDOWNED &&
5427  {
5430  didCrash = true;
5431  }
5432  else
5433  didCrash = false;
5434 
5435  /*
5436  * Prepare for WAL recovery if needed.
5437  *
5438  * InitWalRecovery analyzes the control file and the backup label file, if
5439  * any. It updates the in-memory ControlFile buffer according to the
5440  * starting checkpoint, and sets InRecovery and ArchiveRecoveryRequested.
5441  * It also applies the tablespace map file, if any.
5442  */
5443  InitWalRecovery(ControlFile, &wasShutdown,
5444  &haveBackupLabel, &haveTblspcMap);
5445  checkPoint = ControlFile->checkPointCopy;
5446 
5447  /* initialize shared memory variables from the checkpoint record */
5448  TransamVariables->nextXid = checkPoint.nextXid;
5449  TransamVariables->nextOid = checkPoint.nextOid;
5451  MultiXactSetNextMXact(checkPoint.nextMulti, checkPoint.nextMultiOffset);
5452  AdvanceOldestClogXid(checkPoint.oldestXid);
5453  SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
5454  SetMultiXactIdLimit(checkPoint.oldestMulti, checkPoint.oldestMultiDB, true);
5456  checkPoint.newestCommitTsXid);
5457  XLogCtl->ckptFullXid = checkPoint.nextXid;
5458 
5459  /*
5460  * Clear out any old relcache cache files. This is *necessary* if we do
5461  * any WAL replay, since that would probably result in the cache files
5462  * being out of sync with database reality. In theory we could leave them
5463  * in place if the database had been cleanly shut down, but it seems
5464  * safest to just remove them always and let them be rebuilt during the
5465  * first backend startup. These files needs to be removed from all
5466  * directories including pg_tblspc, however the symlinks are created only
5467  * after reading tablespace_map file in case of archive recovery from
5468  * backup, so needs to clear old relcache files here after creating
5469  * symlinks.
5470  */
5472 
5473  /*
5474  * Initialize replication slots, before there's a chance to remove
5475  * required resources.
5476  */
5478 
5479  /*
5480  * Startup logical state, needs to be setup now so we have proper data
5481  * during crash recovery.
5482  */
5484 
5485  /*
5486  * Startup CLOG. This must be done after TransamVariables->nextXid has
5487  * been initialized and before we accept connections or begin WAL replay.
5488  */
5489  StartupCLOG();
5490 
5491  /*
5492  * Startup MultiXact. We need to do this early to be able to replay
5493  * truncations.
5494  */
5495  StartupMultiXact();
5496 
5497  /*
5498  * Ditto for commit timestamps. Activate the facility if the setting is
5499  * enabled in the control file, as there should be no tracking of commit
5500  * timestamps done when the setting was disabled. This facility can be
5501  * started or stopped when replaying a XLOG_PARAMETER_CHANGE record.
5502  */
5504  StartupCommitTs();
5505 
5506  /*
5507  * Recover knowledge about replay progress of known replication partners.
5508  */
5510 
5511  /*
5512  * Initialize unlogged LSN. On a clean shutdown, it's restored from the
5513  * control file. On recovery, all unlogged relations are blown away, so
5514  * the unlogged LSN counter can be reset too.
5515  */
5519  else
5522 
5523  /*
5524  * Copy any missing timeline history files between 'now' and the recovery
5525  * target timeline from archive to pg_wal. While we don't need those files
5526  * ourselves - the history file of the recovery target timeline covers all
5527  * the previous timelines in the history too - a cascading standby server
5528  * might be interested in them. Or, if you archive the WAL from this
5529  * server to a different archive than the primary, it'd be good for all
5530  * the history files to get archived there after failover, so that you can
5531  * use one of the old timelines as a PITR target. Timeline history files
5532  * are small, so it's better to copy them unnecessarily than not copy them
5533  * and regret later.
5534  */
5536 
5537  /*
5538  * Before running in recovery, scan pg_twophase and fill in its status to
5539  * be able to work on entries generated by redo. Doing a scan before
5540  * taking any recovery action has the merit to discard any 2PC files that
5541  * are newer than the first record to replay, saving from any conflicts at
5542  * replay. This avoids as well any subsequent scans when doing recovery
5543  * of the on-disk two-phase data.
5544  */
5546 
5547  /*
5548  * When starting with crash recovery, reset pgstat data - it might not be
5549  * valid. Otherwise restore pgstat data. It's safe to do this here,
5550  * because postmaster will not yet have started any other processes.
5551  *
5552  * NB: Restoring replication slot stats relies on slot state to have
5553  * already been restored from disk.
5554  *
5555  * TODO: With a bit of extra work we could just start with a pgstat file
5556  * associated with the checkpoint redo location we're starting from.
5557  */
5558  if (didCrash)
5560  else
5562 
5563  lastFullPageWrites = checkPoint.fullPageWrites;
5564 
5567 
5568  /* REDO */
5569  if (InRecovery)
5570  {
5571  /* Initialize state for RecoveryInProgress() */
5573  if (InArchiveRecovery)
5575  else
5578 
5579  /*
5580  * Update pg_control to show that we are recovering and to show the
5581  * selected checkpoint as the place we are starting from. We also mark
5582  * pg_control with any minimum recovery stop point obtained from a
5583  * backup history file.
5584  *
5585  * No need to hold ControlFileLock yet, we aren't up far enough.
5586  */
5588 
5589  /*
5590  * If there was a backup label file, it's done its job and the info
5591  * has now been propagated into pg_control. We must get rid of the
5592  * label file so that if we crash during recovery, we'll pick up at
5593  * the latest recovery restartpoint instead of going all the way back
5594  * to the backup start point. It seems prudent though to just rename
5595  * the file out of the way rather than delete it completely.
5596  */
5597  if (haveBackupLabel)
5598  {
5599  unlink(BACKUP_LABEL_OLD);
5601  }
5602 
5603  /*
5604  * If there was a tablespace_map file, it's done its job and the
5605  * symlinks have been created. We must get rid of the map file so
5606  * that if we crash during recovery, we don't create symlinks again.
5607  * It seems prudent though to just rename the file out of the way
5608  * rather than delete it completely.
5609  */
5610  if (haveTblspcMap)
5611  {
5612  unlink(TABLESPACE_MAP_OLD);
5614  }
5615 
5616  /*
5617  * Initialize our local copy of minRecoveryPoint. When doing crash
5618  * recovery we want to replay up to the end of WAL. Particularly, in
5619  * the case of a promoted standby minRecoveryPoint value in the
5620  * control file is only updated after the first checkpoint. However,
5621  * if the instance crashes before the first post-recovery checkpoint
5622  * is completed then recovery will use a stale location causing the
5623  * startup process to think that there are still invalid page
5624  * references when checking for data consistency.
5625  */
5626  if (InArchiveRecovery)
5627  {
5630  }
5631  else
5632  {
5635  }
5636 
5637  /* Check that the GUCs used to generate the WAL allow recovery */
5639 
5640  /*
5641  * We're in recovery, so unlogged relations may be trashed and must be
5642  * reset. This should be done BEFORE allowing Hot Standby
5643  * connections, so that read-only backends don't try to read whatever
5644  * garbage is left over from before.
5645  */
5647 
5648  /*
5649  * Likewise, delete any saved transaction snapshot files that got left
5650  * behind by crashed backends.
5651  */
5653 
5654  /*
5655  * Initialize for Hot Standby, if enabled. We won't let backends in
5656  * yet, not until we've reached the min recovery point specified in
5657  * control file and we've established a recovery snapshot from a
5658  * running-xacts WAL record.
5659  */
5661  {
5662  TransactionId *xids;
5663  int nxids;
5664 
5665  ereport(DEBUG1,
5666  (errmsg_internal("initializing for hot standby")));
5667 
5669 
5670  if (wasShutdown)
5671  oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
5672  else
5673  oldestActiveXID = checkPoint.oldestActiveXid;
5674  Assert(TransactionIdIsValid(oldestActiveXID));
5675 
5676  /* Tell procarray about the range of xids it has to deal with */
5678 
5679  /*
5680  * Startup subtrans only. CLOG, MultiXact and commit timestamp
5681  * have already been started up and other SLRUs are not maintained
5682  * during recovery and need not be started yet.
5683  */
5684  StartupSUBTRANS(oldestActiveXID);
5685 
5686  /*
5687  * If we're beginning at a shutdown checkpoint, we know that
5688  * nothing was running on the primary at this point. So fake-up an
5689  * empty running-xacts record and use that here and now. Recover
5690  * additional standby state for prepared transactions.
5691  */
5692  if (wasShutdown)
5693  {
5694  RunningTransactionsData running;
5695  TransactionId latestCompletedXid;
5696 
5697  /*
5698  * Construct a RunningTransactions snapshot representing a
5699  * shut down server, with only prepared transactions still
5700  * alive. We're never overflowed at this point because all
5701  * subxids are listed with their parent prepared transactions.
5702  */
5703  running.xcnt = nxids;
5704  running.subxcnt = 0;
5705  running.subxid_overflow = false;
5706  running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
5707  running.oldestRunningXid = oldestActiveXID;
5708  latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
5709  TransactionIdRetreat(latestCompletedXid);
5710  Assert(TransactionIdIsNormal(latestCompletedXid));
5711  running.latestCompletedXid = latestCompletedXid;
5712  running.xids = xids;
5713 
5714  ProcArrayApplyRecoveryInfo(&running);
5715 
5717  }
5718  }
5719 
5720  /*
5721  * We're all set for replaying the WAL now. Do it.
5722  */
5724  performedWalRecovery = true;
5725  }
5726  else
5727  performedWalRecovery = false;
5728 
5729  /*
5730  * Finish WAL recovery.
5731  */
5732  endOfRecoveryInfo = FinishWalRecovery();
5733  EndOfLog = endOfRecoveryInfo->endOfLog;
5734  EndOfLogTLI = endOfRecoveryInfo->endOfLogTLI;
5735  abortedRecPtr = endOfRecoveryInfo->abortedRecPtr;
5736  missingContrecPtr = endOfRecoveryInfo->missingContrecPtr;
5737 
5738  /*
5739  * Reset ps status display, so as no information related to recovery shows
5740  * up.
5741  */
5742  set_ps_display("");
5743 
5744  /*
5745  * When recovering from a backup (we are in recovery, and archive recovery
5746  * was requested), complain if we did not roll forward far enough to reach
5747  * the point where the database is consistent. For regular online
5748  * backup-from-primary, that means reaching the end-of-backup WAL record
5749  * (at which point we reset backupStartPoint to be Invalid), for
5750  * backup-from-replica (which can't inject records into the WAL stream),
5751  * that point is when we reach the minRecoveryPoint in pg_control (which
5752  * we purposefully copy last when backing up from a replica). For
5753  * pg_rewind (which creates a backup_label with a method of "pg_rewind")
5754  * or snapshot-style backups (which don't), backupEndRequired will be set
5755  * to false.
5756  *
5757  * Note: it is indeed okay to look at the local variable
5758  * LocalMinRecoveryPoint here, even though ControlFile->minRecoveryPoint
5759  * might be further ahead --- ControlFile->minRecoveryPoint cannot have
5760  * been advanced beyond the WAL we processed.
5761  */
5762  if (InRecovery &&
5763  (EndOfLog < LocalMinRecoveryPoint ||
5765  {
5766  /*
5767  * Ran off end of WAL before reaching end-of-backup WAL record, or
5768  * minRecoveryPoint. That's a bad sign, indicating that you tried to
5769  * recover from an online backup but never called pg_backup_stop(), or
5770  * you didn't archive all the WAL needed.
5771  */
5773  {
5775  ereport(FATAL,
5776  (errmsg("WAL ends before end of online backup"),
5777  errhint("All WAL generated while online backup was taken must be available at recovery.")));
5778  else
5779  ereport(FATAL,
5780  (errmsg("WAL ends before consistent recovery point")));
5781  }
5782  }
5783 
5784  /*
5785  * Reset unlogged relations to the contents of their INIT fork. This is
5786  * done AFTER recovery is complete so as to include any unlogged relations
5787  * created during recovery, but BEFORE recovery is marked as having
5788  * completed successfully. Otherwise we'd not retry if any of the post
5789  * end-of-recovery steps fail.
5790  */
5791  if (InRecovery)
5793 
5794  /*
5795  * Pre-scan prepared transactions to find out the range of XIDs present.
5796  * This information is not quite needed yet, but it is positioned here so
5797  * as potential problems are detected before any on-disk change is done.
5798  */
5799  oldestActiveXID = PrescanPreparedTransactions(NULL, NULL);
5800 
5801  /*
5802  * Allow ordinary WAL segment creation before possibly switching to a new
5803  * timeline, which creates a new segment, and after the last ReadRecord().
5804  */
5806 
5807  /*
5808  * Consider whether we need to assign a new timeline ID.
5809  *
5810  * If we did archive recovery, we always assign a new ID. This handles a
5811  * couple of issues. If we stopped short of the end of WAL during
5812  * recovery, then we are clearly generating a new timeline and must assign
5813  * it a unique new ID. Even if we ran to the end, modifying the current
5814  * last segment is problematic because it may result in trying to
5815  * overwrite an already-archived copy of that segment, and we encourage
5816  * DBAs to make their archive_commands reject that. We can dodge the
5817  * problem by making the new active segment have a new timeline ID.
5818  *
5819  * In a normal crash recovery, we can just extend the timeline we were in.
5820  */
5821  newTLI = endOfRecoveryInfo->lastRecTLI;
5823  {
5824  newTLI = findNewestTimeLine(recoveryTargetTLI) + 1;
5825  ereport(LOG,
5826  (errmsg("selected new timeline ID: %u", newTLI)));
5827 
5828  /*
5829  * Make a writable copy of the last WAL segment. (Note that we also
5830  * have a copy of the last block of the old WAL in
5831  * endOfRecovery->lastPage; we will use that below.)
5832  */
5833  XLogInitNewTimeline(EndOfLogTLI, EndOfLog, newTLI);
5834 
5835  /*
5836  * Remove the signal files out of the way, so that we don't
5837  * accidentally re-enter archive recovery mode in a subsequent crash.
5838  */
5839  if (endOfRecoveryInfo->standby_signal_file_found)
5841 
5842  if (endOfRecoveryInfo->recovery_signal_file_found)
5844 
5845  /*
5846  * Write the timeline history file, and have it archived. After this
5847  * point (or rather, as soon as the file is archived), the timeline
5848  * will appear as "taken" in the WAL archive and to any standby
5849  * servers. If we crash before actually switching to the new
5850  * timeline, standby servers will nevertheless think that we switched
5851  * to the new timeline, and will try to connect to the new timeline.
5852  * To minimize the window for that, try to do as little as possible
5853  * between here and writing the end-of-recovery record.
5854  */
5856  EndOfLog, endOfRecoveryInfo->recoveryStopReason);
5857 
5858  ereport(LOG,
5859  (errmsg("archive recovery complete")));
5860  }
5861 
5862  /* Save the selected TimeLineID in shared memory, too */
5863  XLogCtl->InsertTimeLineID = newTLI;
5864  XLogCtl->PrevTimeLineID = endOfRecoveryInfo->lastRecTLI;
5865 
5866  /*
5867  * Actually, if WAL ended in an incomplete record, skip the parts that
5868  * made it through and start writing after the portion that persisted.
5869  * (It's critical to first write an OVERWRITE_CONTRECORD message, which
5870  * we'll do as soon as we're open for writing new WAL.)
5871  */
5873  {
5874  /*
5875  * We should only have a missingContrecPtr if we're not switching to a
5876  * new timeline. When a timeline switch occurs, WAL is copied from the
5877  * old timeline to the new only up to the end of the last complete
5878  * record, so there can't be an incomplete WAL record that we need to
5879  * disregard.
5880  */
5881  Assert(newTLI == endOfRecoveryInfo->lastRecTLI);
5883  EndOfLog = missingContrecPtr;
5884  }
5885 
5886  /*
5887  * Prepare to write WAL starting at EndOfLog location, and init xlog
5888  * buffer cache using the block containing the last record from the
5889  * previous incarnation.
5890  */
5891  Insert = &XLogCtl->Insert;
5892  Insert->PrevBytePos = XLogRecPtrToBytePos(endOfRecoveryInfo->lastRec);
5893  Insert->CurrBytePos = XLogRecPtrToBytePos(EndOfLog);
5894 
5895  /*
5896  * Tricky point here: lastPage contains the *last* block that the LastRec
5897  * record spans, not the one it starts in. The last block is indeed the
5898  * one we want to use.
5899  */
5900  if (EndOfLog % XLOG_BLCKSZ != 0)
5901  {
5902  char *page;
5903  int len;
5904  int firstIdx;
5905 
5906  firstIdx = XLogRecPtrToBufIdx(EndOfLog);
5907  len = EndOfLog - endOfRecoveryInfo->lastPageBeginPtr;
5908  Assert(len < XLOG_BLCKSZ);
5909 
5910  /* Copy the valid part of the last block, and zero the rest */
5911  page = &XLogCtl->pages[firstIdx * XLOG_BLCKSZ];
5912  memcpy(page, endOfRecoveryInfo->lastPage, len);
5913  memset(page + len, 0, XLOG_BLCKSZ - len);
5914 
5915  pg_atomic_write_u64(&XLogCtl->xlblocks[firstIdx], endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ);
5916  XLogCtl->InitializedUpTo = endOfRecoveryInfo->lastPageBeginPtr + XLOG_BLCKSZ;
5917  }
5918  else
5919  {
5920  /*
5921  * There is no partial block to copy. Just set InitializedUpTo, and
5922  * let the first attempt to insert a log record to initialize the next
5923  * buffer.
5924  */
5925  XLogCtl->InitializedUpTo = EndOfLog;
5926  }
5927 
5928  LogwrtResult.Write = LogwrtResult.Flush = EndOfLog;
5929 
5931 
5932  XLogCtl->LogwrtRqst.Write = EndOfLog;
5933  XLogCtl->LogwrtRqst.Flush = EndOfLog;
5934 
5935  /*
5936  * Preallocate additional log files, if wanted.
5937  */
5938  PreallocXlogFiles(EndOfLog, newTLI);
5939 
5940  /*
5941  * Okay, we're officially UP.
5942  */
5943  InRecovery = false;
5944 
5945  /* start the archive_timeout timer and LSN running */
5946  XLogCtl->lastSegSwitchTime = (pg_time_t) time(NULL);
5947  XLogCtl->lastSegSwitchLSN = EndOfLog;
5948 
5949  /* also initialize latestCompletedXid, to nextXid - 1 */
5950  LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
5953  LWLockRelease(ProcArrayLock);
5954 
5955  /*
5956  * Start up subtrans, if not already done for hot standby. (commit
5957  * timestamps are started below, if necessary.)
5958  */
5960  StartupSUBTRANS(oldestActiveXID);
5961 
5962  /*
5963  * Perform end of recovery actions for any SLRUs that need it.
5964  */
5965  TrimCLOG();
5966  TrimMultiXact();
5967 
5968  /*
5969  * Reload shared-memory state for prepared transactions. This needs to
5970  * happen before renaming the last partial segment of the old timeline as
5971  * it may be possible that we have to recovery some transactions from it.
5972  */
5974 
5975  /* Shut down xlogreader */
5977 
5978  /* Enable WAL writes for this backend only. */
5980 
5981  /* If necessary, write overwrite-contrecord before doing anything else */
5983  {
5986  }
5987 
5988  /*
5989  * Update full_page_writes in shared memory and write an XLOG_FPW_CHANGE
5990  * record before resource manager writes cleanup WAL records or checkpoint
5991  * record is written.
5992  */
5993  Insert->fullPageWrites = lastFullPageWrites;
5995 
5996  /*
5997  * Emit checkpoint or end-of-recovery record in XLOG, if required.
5998  */
5999  if (performedWalRecovery)
6000  promoted = PerformRecoveryXLogAction();
6001 
6002  /*
6003  * If any of the critical GUCs have changed, log them before we allow
6004  * backends to write WAL.
6005  */
6007 
6008  /* If this is archive recovery, perform post-recovery cleanup actions. */
6010  CleanupAfterArchiveRecovery(EndOfLogTLI, EndOfLog, newTLI);
6011 
6012  /*
6013  * Local WAL inserts enabled, so it's time to finish initialization of
6014  * commit timestamp.
6015  */
6017 
6018  /*
6019  * All done with end-of-recovery actions.
6020  *
6021  * Now allow backends to write WAL and update the control file status in
6022  * consequence. SharedRecoveryState, that controls if backends can write
6023  * WAL, is updated while holding ControlFileLock to prevent other backends
6024  * to look at an inconsistent state of the control file in shared memory.
6025  * There is still a small window during which backends can write WAL and
6026  * the control file is still referring to a system not in DB_IN_PRODUCTION
6027  * state while looking at the on-disk control file.
6028  *
6029  * Also, we use info_lck to update SharedRecoveryState to ensure that
6030  * there are no race conditions concerning visibility of other recent
6031  * updates to shared memory.
6032  */
6033  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6035 
6039 
6041  LWLockRelease(ControlFileLock);
6042 
6043  /*
6044  * Shutdown the recovery environment. This must occur after
6045  * RecoverPreparedTransactions() (see notes in lock_twophase_recover())
6046  * and after switching SharedRecoveryState to RECOVERY_STATE_DONE so as
6047  * any session building a snapshot will not rely on KnownAssignedXids as
6048  * RecoveryInProgress() would return false at this stage. This is
6049  * particularly critical for prepared 2PC transactions, that would still
6050  * need to be included in snapshots once recovery has ended.
6051  */
6054 
6055  /*
6056  * If there were cascading standby servers connected to us, nudge any wal
6057  * sender processes to notice that we've been promoted.
6058  */
6059  WalSndWakeup(true, true);
6060 
6061  /*
6062  * If this was a promotion, request an (online) checkpoint now. This isn't
6063  * required for consistency, but the last restartpoint might be far back,
6064  * and in case of a crash, recovering from it might take a longer than is
6065  * appropriate now that we're not in standby mode anymore.
6066  */
6067  if (promoted)
6069 }
static void pg_atomic_write_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:480
static void pg_atomic_write_membarrier_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:489
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:639
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
#define FATAL
Definition: elog.h:41
int durable_rename(const char *oldfile, const char *newfile, int elevel)
Definition: fd.c:782
int durable_unlink(const char *fname, int elevel)
Definition: fd.c:872
void SyncDataDirectory(void)
Definition: fd.c:3544
#define IsBootstrapProcessingMode()
Definition: miscadmin.h:451
void TrimMultiXact(void)
Definition: multixact.c:2103
void StartupMultiXact(void)
Definition: multixact.c:2078
void StartupReplicationOrigin(void)
Definition: origin.c:699
@ DB_IN_PRODUCTION
Definition: pg_control.h:96
@ DB_IN_CRASH_RECOVERY
Definition: pg_control.h:94
const void size_t len
void pgstat_restore_stats(void)
Definition: pgstat.c:407
void pgstat_discard_stats(void)
Definition: pgstat.c:419
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:6775
void StartupReorderBuffer(void)
void StartupReplicationSlots(void)
Definition: slot.c:1874
void DeleteAllExportedSnapshotFiles(void)
Definition: snapmgr.c:1567
void InitRecoveryTransactionEnvironment(void)
Definition: standby.c:94
void ShutdownRecoveryTransactionEnvironment(void)
Definition: standby.c:160
bool track_commit_timestamp
Definition: pg_control.h:184
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:84
TransactionId nextXid
Definition: standby.h:83
TransactionId latestCompletedXid
Definition: standby.h:87
TransactionId * xids
Definition: standby.h:89
FullTransactionId latestCompletedXid
Definition: transam.h:238
XLogRecPtr InitializedUpTo
Definition: xlog.c:492
char * pages
Definition: xlog.c:499
pg_time_t lastSegSwitchTime
Definition: xlog.c:473
XLogRecPtr lastSegSwitchLSN
Definition: xlog.c:474
pg_atomic_uint64 * xlblocks
Definition: xlog.c:500
XLogRecPtr Flush
Definition: xlog.c:328
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:2084
void restoreTwoPhaseData(void)
Definition: twophase.c:1898
TransactionId PrescanPreparedTransactions(TransactionId **xids_p, int *nxids_p)
Definition: twophase.c:1962
void StandbyRecoverPreparedTransactions(void)
Definition: twophase.c:2043
void WalSndWakeup(bool physical, bool logical)
Definition: walsender.c:3666
void UpdateFullPageWrites(void)
Definition: xlog.c:8000
static char * str_time(pg_time_t tnow)
Definition: xlog.c:5085
static void ValidateXLOGDirectoryStructure(void)
Definition: xlog.c:4029
static XLogRecPtr CreateOverwriteContrecordRecord(XLogRecPtr aborted_lsn, XLogRecPtr pagePtr, TimeLineID newTLI)
Definition: xlog.c:7274
static void XLogReportParameters(void)
Definition: xlog.c:7937
static bool PerformRecoveryXLogAction(void)
Definition: xlog.c:6151
static void CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog, TimeLineID newTLI)
Definition: xlog.c:5175
static bool lastFullPageWrites
Definition: xlog.c:217
static uint64 XLogRecPtrToBytePos(XLogRecPtr ptr)
Definition: xlog.c:1900
static void XLogInitNewTimeline(TimeLineID endTLI, XLogRecPtr endOfLog, TimeLineID newTLI)
Definition: xlog.c:5100
static void CheckRequiredParameterValues(void)
Definition: xlog.c:5264
#define XLogRecPtrToBufIdx(recptr)
Definition: xlog.c:593
static void RemoveTempXlogFiles(void)
Definition: xlog.c:3762
#define TABLESPACE_MAP_OLD
Definition: xlog.h:302
#define TABLESPACE_MAP
Definition: xlog.h:301
#define STANDBY_SIGNAL_FILE
Definition: xlog.h:297
#define BACKUP_LABEL_OLD
Definition: xlog.h:299
#define BACKUP_LABEL_FILE
Definition: xlog.h:298
#define RECOVERY_SIGNAL_FILE
Definition: xlog.h:296
#define XRecOffIsValid(xlrp)
#define FirstNormalUnloggedLSN
Definition: xlogdefs.h:36
void ShutdownWalRecovery(void)
bool ArchiveRecoveryRequested
Definition: xlogrecovery.c:137
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:49

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, 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::LogwrtResult, 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_overflow, 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().

◆ SwitchIntoArchiveRecovery()

void SwitchIntoArchiveRecovery ( XLogRecPtr  EndRecPtr,
TimeLineID  replayTLI 
)

Definition at line 6076 of file xlog.c.

6077 {
6078  /* initialize minRecoveryPoint to this record */
6079  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
6081  if (ControlFile->minRecoveryPoint < EndRecPtr)
6082  {
6083  ControlFile->minRecoveryPoint = EndRecPtr;
6084  ControlFile->minRecoveryPointTLI = replayTLI;
6085  }
6086  /* update local copy */
6089 
6090  /*
6091  * The startup process can update its local copy of minRecoveryPoint from
6092  * this point.
6093  */
6094  updateMinRecoveryPoint = true;
6095 
6097 
6098  /*
6099  * We update SharedRecoveryState while holding the lock on ControlFileLock
6100  * so both states are consistent in shared memory.
6101  */
6105 
6106  LWLockRelease(ControlFileLock);
6107 }
static bool updateMinRecoveryPoint
Definition: xlog.c:637

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().

◆ UpdateFullPageWrites()

void UpdateFullPageWrites ( void  )

Definition at line 8000 of file xlog.c.

8001 {
8003  bool recoveryInProgress;
8004 
8005  /*
8006  * Do nothing if full_page_writes has not been changed.
8007  *
8008  * It's safe to check the shared full_page_writes without the lock,
8009  * because we assume that there is no concurrently running process which
8010  * can update it.
8011  */
8012  if (fullPageWrites == Insert->fullPageWrites)
8013  return;
8014 
8015  /*
8016  * Perform this outside critical section so that the WAL insert
8017  * initialization done by RecoveryInProgress() doesn't trigger an
8018  * assertion failure.
8019  */
8020  recoveryInProgress = RecoveryInProgress();
8021 
8023 
8024  /*
8025  * It's always safe to take full page images, even when not strictly
8026  * required, but not the other round. So if we're setting full_page_writes
8027  * to true, first set it true and then write the WAL record. If we're
8028  * setting it to false, first write the WAL record and then set the global
8029  * flag.
8030  */
8031  if (fullPageWrites)
8032  {
8034  Insert->fullPageWrites = true;
8036  }
8037 
8038  /*
8039  * Write an XLOG_FPW_CHANGE record. This allows us to keep track of
8040  * full_page_writes during archive recovery, if required.
8041  */
8042  if (XLogStandbyInfoActive() && !recoveryInProgress)
8043  {
8044  XLogBeginInsert();
8045  XLogRegisterData((char *) (&fullPageWrites), sizeof(bool));
8046 
8047  XLogInsert(RM_XLOG_ID, XLOG_FPW_CHANGE);
8048  }
8049 
8050  if (!fullPageWrites)
8051  {
8053  Insert->fullPageWrites = false;
8055  }
8056  END_CRIT_SECTION();
8057 }
#define XLOG_FPW_CHANGE
Definition: pg_control.h:75

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().

◆ WALReadFromBuffers()

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

Definition at line 1718 of file xlog.c.

1720 {
1721  char *pdst = dstbuf;
1722  XLogRecPtr recptr = startptr;
1723  Size nbytes = count;
1724 
1725  if (RecoveryInProgress() || tli != GetWALInsertionTimeLine())
1726  return 0;
1727 
1728  Assert(!XLogRecPtrIsInvalid(startptr));
1729  Assert(startptr + count <= LogwrtResult.Write);
1730 
1731  /*
1732  * Loop through the buffers without a lock. For each buffer, atomically
1733  * read and verify the end pointer, then copy the data out, and finally
1734  * re-read and re-verify the end pointer.
1735  *
1736  * Once a page is evicted, it never returns to the WAL buffers, so if the
1737  * end pointer matches the expected end pointer before and after we copy
1738  * the data, then the right page must have been present during the data
1739  * copy. Read barriers are necessary to ensure that the data copy actually
1740  * happens between the two verification steps.
1741  *
1742  * If either verification fails, we simply terminate the loop and return
1743  * with the data that had been already copied out successfully.
1744  */
1745  while (nbytes > 0)
1746  {
1747  uint32 offset = recptr % XLOG_BLCKSZ;
1748  int idx = XLogRecPtrToBufIdx(recptr);
1749  XLogRecPtr expectedEndPtr;
1750  XLogRecPtr endptr;
1751  const char *page;
1752  const char *psrc;
1753  Size npagebytes;
1754 
1755  /*
1756  * Calculate the end pointer we expect in the xlblocks array if the
1757  * correct page is present.
1758  */
1759  expectedEndPtr = recptr + (XLOG_BLCKSZ - offset);
1760 
1761  /*
1762  * First verification step: check that the correct page is present in
1763  * the WAL buffers.
1764  */
1765  endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1766  if (expectedEndPtr != endptr)
1767  break;
1768 
1769  /*
1770  * The correct page is present (or was at the time the endptr was
1771  * read; must re-verify later). Calculate pointer to source data and
1772  * determine how much data to read from this page.
1773  */
1774  page = XLogCtl->pages + idx * (Size) XLOG_BLCKSZ;
1775  psrc = page + offset;
1776  npagebytes = Min(nbytes, XLOG_BLCKSZ - offset);
1777 
1778  /*
1779  * Ensure that the data copy and the first verification step are not
1780  * reordered.
1781  */
1782  pg_read_barrier();
1783 
1784  /* data copy */
1785  memcpy(pdst, psrc, npagebytes);
1786 
1787  /*
1788  * Ensure that the data copy and the second verification step are not
1789  * reordered.
1790  */
1791  pg_read_barrier();
1792 
1793  /*
1794  * Second verification step: check that the page we read from wasn't
1795  * evicted while we were copying the data.
1796  */
1797  endptr = pg_atomic_read_u64(&XLogCtl->xlblocks[idx]);
1798  if (expectedEndPtr != endptr)
1799  break;
1800 
1801  pdst += npagebytes;
1802  recptr += npagebytes;
1803  nbytes -= npagebytes;
1804  }
1805 
1806  Assert(pdst - dstbuf <= count);
1807 
1808  return pdst - dstbuf;
1809 }
Datum idx(PG_FUNCTION_ARGS)
Definition: _int_op.c:259
#define pg_read_barrier()
Definition: atomics.h:151
static uint64 pg_atomic_read_u64(volatile pg_atomic_uint64 *ptr)
Definition: atomics.h:462
#define Min(x, y)
Definition: c.h:991
size_t Size
Definition: c.h:592
TimeLineID GetWALInsertionTimeLine(void)
Definition: xlog.c:6389

References Assert(), GetWALInsertionTimeLine(), idx(), LogwrtResult, Min, XLogCtlData::pages, pg_atomic_read_u64(), pg_read_barrier, RecoveryInProgress(), XLogwrtResult::Write, XLogCtlData::xlblocks, XLogCtl, XLogRecPtrIsInvalid, and XLogRecPtrToBufIdx.

Referenced by XLogSendPhysical().

◆ xlog_desc()

void xlog_desc ( StringInfo  buf,
struct XLogReaderState record 
)

Definition at line 37 of file xlogdesc.c.

38 {
39  char *rec = XLogRecGetData(record);
40  uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
41 
42  if (info == XLOG_CHECKPOINT_SHUTDOWN ||
43  info == XLOG_CHECKPOINT_ONLINE)
44  {
45  CheckPoint *checkpoint = (CheckPoint *) rec;
46 
47  appendStringInfo(buf, "redo %X/%X; "
48  "tli %u; prev tli %u; fpw %s; xid %u:%u; oid %u; multi %u; offset %u; "
49  "oldest xid %u in DB %u; oldest multi %u in DB %u; "
50  "oldest/newest commit timestamp xid: %u/%u; "
51  "oldest running xid %u; %s",
52  LSN_FORMAT_ARGS(checkpoint->redo),
53  checkpoint->ThisTimeLineID,
54  checkpoint->PrevTimeLineID,
55  checkpoint->fullPageWrites ? "true" : "false",
57  XidFromFullTransactionId(checkpoint->nextXid),
58  checkpoint->nextOid,
59  checkpoint->nextMulti,
60  checkpoint->nextMultiOffset,
61  checkpoint->oldestXid,
62  checkpoint->oldestXidDB,
63  checkpoint->oldestMulti,
64  checkpoint->oldestMultiDB,
65  checkpoint->oldestCommitTsXid,
66  checkpoint->newestCommitTsXid,
67  checkpoint->oldestActiveXid,
68  (info == XLOG_CHECKPOINT_SHUTDOWN) ? "shutdown" : "online");
69  }
70  else if (info == XLOG_NEXTOID)
71  {
72  Oid nextOid;
73 
74  memcpy(&nextOid, rec, sizeof(Oid));
75  appendStringInfo(buf, "%u", nextOid);
76  }
77  else if (info == XLOG_RESTORE_POINT)
78  {
79  xl_restore_point *xlrec = (xl_restore_point *) rec;
80 
82  }
83  else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
84  {
85  /* no further information to print */
86  }
87  else if (info == XLOG_BACKUP_END)
88  {
89  XLogRecPtr startpoint;
90 
91  memcpy(&startpoint, rec, sizeof(XLogRecPtr));
92  appendStringInfo(buf, "%X/%X", LSN_FORMAT_ARGS(startpoint));
93  }
94  else if (info == XLOG_PARAMETER_CHANGE)
95  {
96  xl_parameter_change xlrec;
97  const char *wal_level_str;
98  const struct config_enum_entry *entry;
99 
100  memcpy(&xlrec, rec, sizeof(xl_parameter_change));
101 
102  /* Find a string representation for wal_level */
103  wal_level_str = "?";
104  for (entry = wal_level_options; entry->name; entry++)
105  {
106  if (entry->val == xlrec.wal_level)
107  {
108  wal_level_str = entry->name;
109  break;
110  }
111  }
112 
113  appendStringInfo(buf, "max_connections=%d max_worker_processes=%d "
114  "max_wal_senders=%d max_prepared_xacts=%d "
115  "max_locks_per_xact=%d wal_level=%s "
116  "wal_log_hints=%s track_commit_timestamp=%s",
117  xlrec.MaxConnections,
118  xlrec.max_worker_processes,
119  xlrec.max_wal_senders,
120  xlrec.max_prepared_xacts,
121  xlrec.max_locks_per_xact,
123  xlrec.wal_log_hints ? "on" : "off",
124  xlrec.track_commit_timestamp ? "on" : "off");
125  }
126  else if (info == XLOG_FPW_CHANGE)
127  {
128  bool fpw;
129 
130  memcpy(&fpw, rec, sizeof(bool));
131  appendStringInfoString(buf, fpw ? "true" : "false");
132  }
133  else if (info == XLOG_END_OF_RECOVERY)
134  {
135  xl_end_of_recovery xlrec;
136 
137  memcpy(&xlrec, rec, sizeof(xl_end_of_recovery));
138  appendStringInfo(buf, "tli %u; prev tli %u; time %s",
139  xlrec.ThisTimeLineID, xlrec.PrevTimeLineID,
141  }
142  else if (info == XLOG_OVERWRITE_CONTRECORD)
143  {
145 
146  memcpy(&xlrec, rec, sizeof(xl_overwrite_contrecord));
147  appendStringInfo(buf, "lsn %X/%X; time %s",
150  }
151  else if (info == XLOG_CHECKPOINT_REDO)
152  {
153  /* No details to write out */
154  }
155 }
static const char * wal_level_str(WalLevel wal_level)
unsigned char uint8
Definition: c.h:491
#define XLOG_RESTORE_POINT
Definition: pg_control.h:74
#define XLOG_OVERWRITE_CONTRECORD
Definition: pg_control.h:80
#define XLOG_FPI
Definition: pg_control.h:78
#define XLOG_FPI_FOR_HINT
Definition: pg_control.h:77
#define XLOG_NEXTOID
Definition: pg_control.h:70
#define XLOG_PARAMETER_CHANGE
Definition: pg_control.h:73
#define XLOG_END_OF_RECOVERY
Definition: pg_control.h:76
static char * buf
Definition: pg_test_fsync.c:73
void appendStringInfoString(StringInfo str, const char *s)
Definition: stringinfo.c:182
Definition: guc.h:170
const char * name
Definition: guc.h:171
int val
Definition: guc.h:172
TimeLineID PrevTimeLineID
TimestampTz end_time
TimeLineID ThisTimeLineID
char rp_name[MAXFNAMELEN]
#define EpochFromFullTransactionId(x)
Definition: transam.h:47
const struct config_enum_entry wal_level_options[]
Definition: xlogdesc.c:27
#define XLogRecGetInfo(decoder)
Definition: xlogreader.h:410
#define XLogRecGetData(decoder)
Definition: xlogreader.h:415
#define XLR_INFO_MASK
Definition: xlogrecord.h:62

References appendStringInfo(), appendStringInfoString(), buf, xl_end_of_recovery::end_time, EpochFromFullTransactionId, CheckPoint::fullPageWrites, LSN_FORMAT_ARGS, xl_parameter_change::max_locks_per_xact, xl_parameter_change::max_prepared_xacts, xl_parameter_change::max_wal_senders, xl_parameter_change::max_worker_processes, xl_parameter_change::MaxConnections, config_enum_entry::name, CheckPoint::newestCommitTsXid, CheckPoint::nextMulti, CheckPoint::nextMultiOffset, CheckPoint::nextOid, CheckPoint::nextXid, CheckPoint::oldestActiveXid, CheckPoint::oldestCommitTsXid, CheckPoint::oldestMulti, CheckPoint::oldestMultiDB, CheckPoint::oldestXid, CheckPoint::oldestXidDB, xl_overwrite_contrecord::overwrite_time, xl_overwrite_contrecord::overwritten_lsn, xl_end_of_recovery::PrevTimeLineID, CheckPoint::PrevTimeLineID, CheckPoint::redo, xl_restore_point::rp_name, xl_end_of_recovery::ThisTimeLineID, CheckPoint::ThisTimeLineID, timestamptz_to_str(), xl_parameter_change::track_commit_timestamp, config_enum_entry::val, xl_parameter_change::wal_level, wal_level_options, wal_level_str(), xl_parameter_change::wal_log_hints, XidFromFullTransactionId, 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_OVERWRITE_CONTRECORD, XLOG_PARAMETER_CHANGE, XLOG_RESTORE_POINT, XLogRecGetData, XLogRecGetInfo, and XLR_INFO_MASK.

◆ xlog_identify()

const char* xlog_identify ( uint8  info)

Definition at line 158 of file xlogdesc.c.

159 {
160  const char *id = NULL;
161 
162  switch (info & ~XLR_INFO_MASK)
163  {
165  id = "CHECKPOINT_SHUTDOWN";
166  break;
168  id = "CHECKPOINT_ONLINE";
169  break;
170  case XLOG_NOOP:
171  id = "NOOP";
172  break;
173  case XLOG_NEXTOID:
174  id = "NEXTOID";
175  break;
176  case XLOG_SWITCH:
177  id = "SWITCH";
178  break;
179  case XLOG_BACKUP_END:
180  id = "BACKUP_END";
181  break;
183  id = "PARAMETER_CHANGE";
184  break;
185  case XLOG_RESTORE_POINT:
186  id = "RESTORE_POINT";
187  break;
188  case XLOG_FPW_CHANGE:
189  id = "FPW_CHANGE";
190  break;
192  id = "END_OF_RECOVERY";
193  break;
195  id = "OVERWRITE_CONTRECORD";
196  break;
197  case XLOG_FPI:
198  id = "FPI";
199  break;
200  case XLOG_FPI_FOR_HINT:
201  id = "FPI_FOR_HINT";
202  break;
204  id = "CHECKPOINT_REDO";
205  break;
206  }
207 
208  return id;
209 }
#define XLOG_NOOP
Definition: pg_control.h:69
#define XLOG_SWITCH
Definition: pg_control.h:71

References 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, and XLR_INFO_MASK.

◆ xlog_redo()

void xlog_redo ( struct XLogReaderState record)

Definition at line 8069 of file xlog.c.

8070 {
8071  uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
8072  XLogRecPtr lsn = record->EndRecPtr;
8073 
8074  /*
8075  * In XLOG rmgr, backup blocks are only used by XLOG_FPI and
8076  * XLOG_FPI_FOR_HINT records.
8077  */
8078  Assert(info == XLOG_FPI || info == XLOG_FPI_FOR_HINT ||
8079  !XLogRecHasAnyBlockRefs(record));
8080 
8081  if (info == XLOG_NEXTOID)
8082  {
8083  Oid nextOid;
8084 
8085  /*
8086  * We used to try to take the maximum of TransamVariables->nextOid and
8087  * the recorded nextOid, but that fails if the OID counter wraps
8088  * around. Since no OID allocation should be happening during replay
8089  * anyway, better to just believe the record exactly. We still take
8090  * OidGenLock while setting the variable, just in case.
8091  */
8092  memcpy(&nextOid, XLogRecGetData(record), sizeof(Oid));
8093  LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8094  TransamVariables->nextOid = nextOid;
8096  LWLockRelease(OidGenLock);
8097  }
8098  else if (info == XLOG_CHECKPOINT_SHUTDOWN)
8099  {
8100  CheckPoint checkPoint;
8101  TimeLineID replayTLI;
8102 
8103  memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8104  /* In a SHUTDOWN checkpoint, believe the counters exactly */
8105  LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8106  TransamVariables->nextXid = checkPoint.nextXid;
8107  LWLockRelease(XidGenLock);
8108  LWLockAcquire(OidGenLock, LW_EXCLUSIVE);
8109  TransamVariables->nextOid = checkPoint.nextOid;
8111  LWLockRelease(OidGenLock);
8112  MultiXactSetNextMXact(checkPoint.nextMulti,
8113  checkPoint.nextMultiOffset);
8114 
8116  checkPoint.oldestMultiDB);
8117 
8118  /*
8119  * No need to set oldestClogXid here as well; it'll be set when we
8120  * redo an xl_clog_truncate if it changed since initialization.
8121  */
8122  SetTransactionIdLimit(checkPoint.oldestXid, checkPoint.oldestXidDB);
8123 
8124  /*
8125  * If we see a shutdown checkpoint while waiting for an end-of-backup
8126  * record, the backup was canceled and the end-of-backup record will
8127  * never arrive.
8128  */
8132  ereport(PANIC,
8133  (errmsg("online backup was canceled, recovery cannot continue")));
8134 
8135  /*
8136  * If we see a shutdown checkpoint, we know that nothing was running
8137  * on the primary at this point. So fake-up an empty running-xacts
8138  * record and use that here and now. Recover additional standby state
8139  * for prepared transactions.
8140  */
8142  {
8143  TransactionId *xids;
8144  int nxids;
8145  TransactionId oldestActiveXID;
8146  TransactionId latestCompletedXid;
8147  RunningTransactionsData running;
8148 
8149  oldestActiveXID = PrescanPreparedTransactions(&xids, &nxids);
8150 
8151  /*
8152  * Construct a RunningTransactions snapshot representing a shut
8153  * down server, with only prepared transactions still alive. We're
8154  * never overflowed at this point because all subxids are listed
8155  * with their parent prepared transactions.
8156  */
8157  running.xcnt = nxids;
8158  running.subxcnt = 0;
8159  running.subxid_overflow = false;
8160  running.nextXid = XidFromFullTransactionId(checkPoint.nextXid);
8161  running.oldestRunningXid = oldestActiveXID;
8162  latestCompletedXid = XidFromFullTransactionId(checkPoint.nextXid);
8163  TransactionIdRetreat(latestCompletedXid);
8164  Assert(TransactionIdIsNormal(latestCompletedXid));
8165  running.latestCompletedXid = latestCompletedXid;
8166  running.xids = xids;
8167 
8168  ProcArrayApplyRecoveryInfo(&running);
8169 
8171  }
8172 
8173  /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8174  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8175  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8176  LWLockRelease(ControlFileLock);
8177 
8178  /* Update shared-memory copy of checkpoint XID/epoch */
8180  XLogCtl->ckptFullXid = checkPoint.nextXid;
8182 
8183  /*
8184  * We should've already switched to the new TLI before replaying this
8185  * record.
8186  */
8187  (void) GetCurrentReplayRecPtr(&replayTLI);
8188  if (checkPoint.ThisTimeLineID != replayTLI)
8189  ereport(PANIC,
8190  (errmsg("unexpected timeline ID %u (should be %u) in shutdown checkpoint record",
8191  checkPoint.ThisTimeLineID, replayTLI)));
8192 
8193  RecoveryRestartPoint(&checkPoint, record);
8194  }
8195  else if (info == XLOG_CHECKPOINT_ONLINE)
8196  {
8197  CheckPoint checkPoint;
8198  TimeLineID replayTLI;
8199 
8200  memcpy(&checkPoint, XLogRecGetData(record), sizeof(CheckPoint));
8201  /* In an ONLINE checkpoint, treat the XID counter as a minimum */
8202  LWLockAcquire(XidGenLock, LW_EXCLUSIVE);
8204  checkPoint.nextXid))
8205  TransamVariables->nextXid = checkPoint.nextXid;
8206  LWLockRelease(XidGenLock);
8207 
8208  /*
8209  * We ignore the nextOid counter in an ONLINE checkpoint, preferring
8210  * to track OID assignment through XLOG_NEXTOID records. The nextOid
8211  * counter is from the start of the checkpoint and might well be stale
8212  * compared to later XLOG_NEXTOID records. We could try to take the
8213  * maximum of the nextOid counter and our latest value, but since
8214  * there's no particular guarantee about the speed with which the OID
8215  * counter wraps around, that's a risky thing to do. In any case,
8216  * users of the nextOid counter are required to avoid assignment of
8217  * duplicates, so that a somewhat out-of-date value should be safe.
8218  */
8219 
8220  /* Handle multixact */
8222  checkPoint.nextMultiOffset);
8223 
8224  /*
8225  * NB: This may perform multixact truncation when replaying WAL
8226  * generated by an older primary.
8227  */
8229  checkPoint.oldestMultiDB);
8231  checkPoint.oldestXid))
8232  SetTransactionIdLimit(checkPoint.oldestXid,
8233  checkPoint.oldestXidDB);
8234  /* ControlFile->checkPointCopy always tracks the latest ckpt XID */
8235  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8236  ControlFile->checkPointCopy.nextXid = checkPoint.nextXid;
8237  LWLockRelease(ControlFileLock);
8238 
8239  /* Update shared-memory copy of checkpoint XID/epoch */
8241  XLogCtl->ckptFullXid = checkPoint.nextXid;
8243 
8244  /* TLI should not change in an on-line checkpoint */
8245  (void) GetCurrentReplayRecPtr(&replayTLI);
8246  if (checkPoint.ThisTimeLineID != replayTLI)
8247  ereport(PANIC,
8248  (errmsg("unexpected timeline ID %u (should be %u) in online checkpoint record",
8249  checkPoint.ThisTimeLineID, replayTLI)));
8250 
8251  RecoveryRestartPoint(&checkPoint, record);
8252  }
8253  else if (info == XLOG_OVERWRITE_CONTRECORD)
8254  {
8255  /* nothing to do here, handled in xlogrecovery_redo() */
8256  }
8257  else if (info == XLOG_END_OF_RECOVERY)
8258  {
8259  xl_end_of_recovery xlrec;
8260  TimeLineID replayTLI;
8261 
8262  memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_end_of_recovery));
8263 
8264  /*
8265  * For Hot Standby, we could treat this like a Shutdown Checkpoint,
8266  * but this case is rarer and harder to test, so the benefit doesn't
8267  * outweigh the potential extra cost of maintenance.
8268  */
8269 
8270  /*
8271  * We should've already switched to the new TLI before replaying this
8272  * record.
8273  */
8274  (void) GetCurrentReplayRecPtr(&replayTLI);
8275  if (xlrec.ThisTimeLineID != replayTLI)
8276  ereport(PANIC,
8277  (errmsg("unexpected timeline ID %u (should be %u) in end-of-recovery record",
8278  xlrec.ThisTimeLineID, replayTLI)));
8279  }
8280  else if (info == XLOG_NOOP)
8281  {
8282  /* nothing to do here */
8283  }
8284  else if (info == XLOG_SWITCH)
8285  {
8286  /* nothing to do here */
8287  }
8288  else if (info == XLOG_RESTORE_POINT)
8289  {
8290  /* nothing to do here, handled in xlogrecovery.c */
8291  }
8292  else if (info == XLOG_FPI || info == XLOG_FPI_FOR_HINT)
8293  {
8294  /*
8295  * XLOG_FPI records contain nothing else but one or more block
8296  * references. Every block reference must include a full-page image
8297  * even if full_page_writes was disabled when the record was generated
8298  * - otherwise there would be no point in this record.
8299  *
8300  * XLOG_FPI_FOR_HINT records are generated when a page needs to be
8301  * WAL-logged because of a hint bit update. They are only generated
8302  * when checksums and/or wal_log_hints are enabled. They may include
8303  * no full-page images if full_page_writes was disabled when they were
8304  * generated. In this case there is nothing to do here.
8305  *
8306  * No recovery conflicts are generated by these generic records - if a
8307  * resource manager needs to generate conflicts, it has to define a
8308  * separate WAL record type and redo routine.
8309  */
8310  for (uint8 block_id = 0; block_id <= XLogRecMaxBlockId(record); block_id++)
8311  {
8312  Buffer buffer;
8313 
8314  if (!XLogRecHasBlockImage(record, block_id))
8315  {
8316  if (info == XLOG_FPI)
8317  elog(ERROR, "XLOG_FPI record did not contain a full-page image");
8318  continue;
8319  }
8320 
8321  if (XLogReadBufferForRedo(record, block_id, &buffer) != BLK_RESTORED)
8322  elog(ERROR, "unexpected XLogReadBufferForRedo result when restoring backup block");
8323  UnlockReleaseBuffer(buffer);
8324  }
8325  }
8326  else if (info == XLOG_BACKUP_END)
8327  {
8328  /* nothing to do here, handled in xlogrecovery_redo() */
8329  }
8330  else if (info == XLOG_PARAMETER_CHANGE)
8331  {
8332  xl_parameter_change xlrec;
8333 
8334  /* Update our copy of the parameters in pg_control */
8335  memcpy(&xlrec, XLogRecGetData(record), sizeof(xl_parameter_change));
8336 
8337  /*
8338  * Invalidate logical slots if we are in hot standby and the primary
8339  * does not have a WAL level sufficient for logical decoding. No need
8340  * to search for potentially conflicting logically slots if standby is
8341  * running with wal_level lower than logical, because in that case, we
8342  * would have either disallowed creation of logical slots or
8343  * invalidated existing ones.
8344  */
8345  if (InRecovery && InHotStandby &&
8346  xlrec.wal_level < WAL_LEVEL_LOGICAL &&
8349  0, InvalidOid,
8351 
8352  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
8358  ControlFile->wal_level = xlrec.wal_level;
8360 
8361  /*
8362  * Update minRecoveryPoint to ensure that if recovery is aborted, we
8363  * recover back up to this point before allowing hot standby again.
8364  * This is important if the max_* settings are decreased, to ensure
8365  * you don't run queries against the WAL preceding the change. The
8366  * local copies cannot be updated as long as crash recovery is
8367  * happening and we expect all the WAL to be replayed.
8368  */
8369  if (InArchiveRecovery)
8370  {
8373  }
8375  {
8376  TimeLineID replayTLI;
8377 
8378  (void) GetCurrentReplayRecPtr(&replayTLI);
8380  ControlFile->minRecoveryPointTLI = replayTLI;
8381  }
8382 
8386 
8388  LWLockRelease(ControlFileLock);
8389 
8390  /* Check to see if any parameter change gives a problem on recovery */
8392  }
8393  else if (info == XLOG_FPW_CHANGE)
8394  {
8395  bool fpw;
8396 
8397  memcpy(&fpw, XLogRecGetData(record), sizeof(bool));
8398 
8399  /*
8400  * Update the LSN of the last replayed XLOG_FPW_CHANGE record so that
8401  * do_pg_backup_start() and do_pg_backup_stop() can check whether
8402  * full_page_writes has been disabled during online backup.
8403  */
8404  if (!fpw)
8405  {
8407  if (XLogCtl->lastFpwDisableRecPtr < record->ReadRecPtr)
8410  }
8411 
8412  /* Keep track of full_page_writes */
8413  lastFullPageWrites = fpw;
8414  }
8415  else if (info == XLOG_CHECKPOINT_REDO)
8416  {
8417  /* nothing to do here, just for informational purposes */
8418  }
8419 }
int Buffer
Definition: buf.h:23
void UnlockReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:4577
void CommitTsParameterChange(bool newvalue, bool oldvalue)
Definition: commit_ts.c:664
void MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
Definition: multixact.c:2461
void MultiXactAdvanceNextMXact(MultiXactId minMulti, MultiXactOffset minMultiOffset)
Definition: multixact.c:2436
@ RS_INVAL_WAL_LEVEL
Definition: slot.h:55
int max_worker_processes
Definition: pg_control.h:180
int max_locks_per_xact
Definition: pg_control.h:183
int max_prepared_xacts
Definition: pg_control.h:182
XLogRecPtr EndRecPtr
Definition: xlogreader.h:207
XLogRecPtr ReadRecPtr
Definition: xlogreader.h:206
bool TransactionIdPrecedes(TransactionId id1, TransactionId id2)
Definition: transam.c:280
#define FullTransactionIdPrecedes(a, b)
Definition: transam.h:51
int wal_level
Definition: xlog.c:131
static void RecoveryRestartPoint(const CheckPoint *checkPoint, XLogReaderState *record)
Definition: xlog.c:7384
#define XLogRecMaxBlockId(decoder)
Definition: xlogreader.h:418
#define XLogRecHasBlockImage(decoder, block_id)
Definition: xlogreader.h:423
#define XLogRecHasAnyBlockRefs(decoder)
Definition: xlogreader.h:417
XLogRecPtr GetCurrentReplayRecPtr(TimeLineID *replayEndTLI)
XLogRedoAction XLogReadBufferForRedo(XLogReaderState *record, uint8 block_id, Buffer *buf)
Definition: xlogutils.c:314
@ STANDBY_INITIALIZED
Definition: xlogutils.h:50
#define InHotStandby
Definition: xlogutils.h:57
@ BLK_RESTORED
Definition: xlogutils.h:73

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_overflow, 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 2918 of file xlog.c.

2919 {
2920  XLogwrtRqst WriteRqst;
2921  bool flexible = true;
2922  static TimestampTz lastflush;
2923  TimestampTz now;
2924  int flushblocks;
2925  TimeLineID insertTLI;
2926 
2927  /* XLOG doesn't need flushing during recovery */
2928  if (RecoveryInProgress())
2929  return false;
2930 
2931  /*
2932  * Since we're not in recovery, InsertTimeLineID is set and can't change,
2933  * so we can read it without a lock.
2934  */
2935  insertTLI = XLogCtl->InsertTimeLineID;
2936 
2937  /* read LogwrtResult and update local state */
2940  WriteRqst = XLogCtl->LogwrtRqst;
2942 
2943  /* back off to last completed page boundary */
2944  WriteRqst.Write -= WriteRqst.Write % XLOG_BLCKSZ;
2945 
2946  /* if we have already flushed that far, consider async commit records */
2947  if (WriteRqst.Write <= LogwrtResult.Flush)
2948  {
2950  WriteRqst.Write = XLogCtl->asyncXactLSN;
2952  flexible = false; /* ensure it all gets written */
2953  }
2954 
2955  /*
2956  * If already known flushed, we're done. Just need to check if we are
2957  * holding an open file handle to a logfile that's no longer in use,
2958  * preventing the file from being deleted.
2959  */
2960  if (WriteRqst.Write <= LogwrtResult.Flush)
2961  {
2962  if (openLogFile >= 0)
2963  {
2966  {
2967  XLogFileClose();
2968  }
2969  }
2970  return false;
2971  }
2972 
2973  /*
2974  * Determine how far to flush WAL, based on the wal_writer_delay and
2975  * wal_writer_flush_after GUCs.
2976  *
2977  * Note that XLogSetAsyncXactLSN() performs similar calculation based on
2978  * wal_writer_flush_after, to decide when to wake us up. Make sure the
2979  * logic is the same in both places if you change this.
2980  */
2982  flushblocks =
2983  WriteRqst.Write / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2984 
2985  if (WalWriterFlushAfter == 0 || lastflush == 0)
2986  {
2987  /* first call, or block based limits disabled */
2988  WriteRqst.Flush = WriteRqst.Write;
2989  lastflush = now;
2990  }
2991  else if (TimestampDifferenceExceeds(lastflush, now, WalWriterDelay))
2992  {
2993  /*
2994  * Flush the writes at least every WalWriterDelay ms. This is
2995  * important to bound the amount of time it takes for an asynchronous
2996  * commit to hit disk.
2997  */
2998  WriteRqst.Flush = WriteRqst.Write;
2999  lastflush = now;
3000  }
3001  else if (flushblocks >= WalWriterFlushAfter)
3002  {
3003  /* exceeded wal_writer_flush_after blocks, flush */
3004  WriteRqst.Flush = WriteRqst.Write;
3005  lastflush = now;
3006  }
3007  else
3008  {
3009  /* no flushing, this time round */
3010  WriteRqst.Flush = 0;
3011  }
3012 
3013 #ifdef WAL_DEBUG
3014  if (XLOG_DEBUG)
3015  elog(LOG, "xlog bg flush request write %X/%X; flush: %X/%X, current is write %X/%X; flush %X/%X",
3016  LSN_FORMAT_ARGS(WriteRqst.Write),
3017  LSN_FORMAT_ARGS(WriteRqst.Flush),
3020 #endif
3021 
3023 
3024  /* now wait for any in-progress insertions to finish and get write lock */
3025  WaitXLogInsertionsToFinish(WriteRqst.Write);
3026  LWLockAcquire(WALWriteLock, LW_EXCLUSIVE);
3028  if (WriteRqst.Write > LogwrtResult.Write ||
3029  WriteRqst.Flush > LogwrtResult.Flush)
3030  {
3031  XLogWrite(WriteRqst, insertTLI, flexible);
3032  }
3033  LWLockRelease(WALWriteLock);
3034 
3035  END_CRIT_SECTION();
3036 
3037  /* wake up walsenders now that we've released heavily contended locks */
3039 
3040  /*
3041  * Great, done. To take some work off the critical path, try to initialize
3042  * as many of the no-longer-needed WAL buffers for future use as we can.
3043  */
3044  AdvanceXLInsertBuffer(InvalidXLogRecPtr, insertTLI, true);
3045 
3046  /*
3047  * If we determined that we need to write data, but somebody else
3048  * wrote/flushed already, it should be considered as being active, to
3049  * avoid hibernating too early.
3050  */
3051  return true;
3052 }
bool TimestampDifferenceExceeds(TimestampTz start_time, TimestampTz stop_time, int msec)
Definition: timestamp.c:1790
Datum now(PG_FUNCTION_ARGS)
Definition: timestamp.c:1618
XLogRecPtr asyncXactLSN
Definition: xlog.c:464
static void WalSndWakeupProcessRequests(bool physical, bool logical)
Definition: walsender.h:66
int WalWriterFlushAfter
Definition: walwriter.c:72
int WalWriterDelay
Definition: walwriter.c:71
static XLogRecPtr WaitXLogInsertionsToFinish(XLogRecPtr upto)
Definition: xlog.c:1492
static void AdvanceXLInsertBuffer(XLogRecPtr upto, TimeLineID tli, bool opportunistic)
Definition: xlog.c:1944
static void XLogWrite(XLogwrtRqst WriteRqst, TimeLineID tli, bool flexible)
Definition: xlog.c:2273
static void XLogFileClose(void)
Definition: xlog.c:3569
static XLogSegNo openLogSegNo
Definition: xlog.c:625
#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, XLogCtlData::LogwrtResult, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, LW_EXCLUSIVE, LWLockAcquire(), LWLockRelease(), now(), openLogFile, openLogSegNo, RecoveryInProgress(), 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().

◆ XLogCheckpointNeeded()

bool XLogCheckpointNeeded ( XLogSegNo  new_segno)

Definition at line 2249 of file xlog.c.

2250 {
2251  XLogSegNo old_segno;
2252 
2254 
2255  if (new_segno >= old_segno + (uint64) (CheckPointSegments - 1))
2256  return true;
2257  return false;
2258 }
int CheckPointSegments
Definition: xlog.c:156

References CheckPointSegments, RedoRecPtr, wal_segment_size, and XLByteToSeg.

Referenced by XLogPageRead(), and XLogWrite().

◆ XLogFileInit()

int XLogFileInit ( XLogSegNo  logsegno,
TimeLineID  logtli 
)

Definition at line 3310 of file xlog.c.

3311 {
3312  bool ignore_added;
3313  char path[MAXPGPATH];
3314  int fd;
3315 
3316  Assert(logtli != 0);
3317 
3318  fd = XLogFileInitInternal(logsegno, logtli, &ignore_added, path);
3319  if (fd >= 0)
3320  return fd;
3321 
3322  /* Now open original target segment (might not be file I just made) */
3323  fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3325  if (fd < 0)
3326  ereport(ERROR,
3328  errmsg("could not open file \"%s\": %m", path)));
3329  return fd;
3330 }
#define PG_BINARY
Definition: c.h:1260
int BasicOpenFile(const char *fileName, int fileFlags)
Definition: fd.c:1087
#define O_CLOEXEC
Definition: win32_port.h:359
static int get_sync_bit(int method)
Definition: xlog.c:8426
static int XLogFileInitInternal(XLogSegNo logsegno, TimeLineID logtli, bool *added, char *path)
Definition: xlog.c:3140

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().

◆ XLogFileOpen()

int XLogFileOpen ( XLogSegNo  segno,
TimeLineID  tli 
)

Definition at line 3548 of file xlog.c.

3549 {
3550  char path[MAXPGPATH];
3551  int fd;
3552 
3553  XLogFilePath(path, tli, segno, wal_segment_size);
3554 
3555  fd = BasicOpenFile(path, O_RDWR | PG_BINARY | O_CLOEXEC |
3557  if (fd < 0)
3558  ereport(PANIC,
3560  errmsg("could not open file \"%s\": %m", path)));
3561 
3562  return fd;
3563 }
static void XLogFilePath(char *path, TimeLineID tli, XLogSegNo logSegNo, int wal_segsz_bytes)

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().

◆ XLogFlush()

void XLogFlush ( XLogRecPtr  record)

Definition at line 2728 of file xlog.c.

2729 {
2730  XLogRecPtr WriteRqstPtr;
2731  XLogwrtRqst WriteRqst;
2732  TimeLineID insertTLI = XLogCtl->InsertTimeLineID;
2733 
2734  /*
2735  * During REDO, we are reading not writing WAL. Therefore, instead of
2736  * trying to flush the WAL, we should update minRecoveryPoint instead. We
2737  * test XLogInsertAllowed(), not InRecovery, because we need checkpointer
2738  * to act this way too, and because when it tries to write the
2739  * end-of-recovery checkpoint, it should indeed flush.
2740  */
2741  if (!XLogInsertAllowed())
2742  {
2743  UpdateMinRecoveryPoint(record, false);
2744  return;
2745  }
2746 
2747  /* Quick exit if already known flushed */
2748  if (record <= LogwrtResult.Flush)
2749  return;
2750 
2751 #ifdef WAL_DEBUG
2752  if (XLOG_DEBUG)
2753  elog(LOG, "xlog flush request %X/%X; write %X/%X; flush %X/%X",
2754  LSN_FORMAT_ARGS(record),
2757 #endif
2758 
2760 
2761  /*
2762  * Since fsync is usually a horribly expensive operation, we try to
2763  * piggyback as much data as we can on each fsync: if we see any more data
2764  * entered into the xlog buffer, we'll write and fsync that too, so that
2765  * the final value of LogwrtResult.Flush is as large as possible. This
2766  * gives us some chance of avoiding another fsync immediately after.
2767  */
2768 
2769  /* initialize to given target; may increase below */
2770  WriteRqstPtr = record;
2771 
2772  /*
2773  * Now wait until we get the write lock, or someone else does the flush
2774  * for us.
2775  */
2776  for (;;)
2777  {
2778  XLogRecPtr insertpos;
2779 
2780  /* read LogwrtResult and update local state */
2782  if (WriteRqstPtr < XLogCtl->LogwrtRqst.Write)
2783  WriteRqstPtr = XLogCtl->LogwrtRqst.Write;
2786 
2787  /* done already? */
2788  if (record <= LogwrtResult.Flush)
2789  break;
2790 
2791  /*
2792  * Before actually performing the write, wait for all in-flight
2793  * insertions to the pages we're about to write to finish.
2794  */
2795  insertpos = WaitXLogInsertionsToFinish(WriteRqstPtr);
2796 
2797  /*
2798  * Try to get the write lock. If we can't get it immediately, wait
2799  * until it's released, and recheck if we still need to do the flush
2800  * or if the backend that held the lock did it for us already. This
2801  * helps to maintain a good rate of group committing when the system
2802  * is bottlenecked by the speed of fsyncing.
2803  */
2804  if (!LWLockAcquireOrWait(WALWriteLock, LW_EXCLUSIVE))
2805  {
2806  /*
2807  * The lock is now free, but we didn't acquire it yet. Before we
2808  * do, loop back to check if someone else flushed the record for
2809  * us already.
2810  */
2811  continue;
2812  }
2813 
2814  /* Got the lock; recheck whether request is satisfied */
2816  if (record <= LogwrtResult.Flush)
2817  {
2818  LWLockRelease(WALWriteLock);
2819  break;
2820  }
2821 
2822  /*
2823  * Sleep before flush! By adding a delay here, we may give further
2824  * backends the opportunity to join the backlog of group commit
2825  * followers; this can significantly improve transaction throughput,
2826  * at the risk of increasing transaction latency.
2827  *
2828  * We do not sleep if enableFsync is not turned on, nor if there are
2829  * fewer than CommitSiblings other backends with active transactions.
2830  */
2831  if (CommitDelay > 0 && enableFsync &&
2833  {
2835 
2836  /*
2837  * Re-check how far we can now flush the WAL. It's generally not
2838  * safe to call WaitXLogInsertionsToFinish while holding
2839  * WALWriteLock, because an in-progress insertion might need to
2840  * also grab WALWriteLock to make progress. But we know that all
2841  * the insertions up to insertpos have already finished, because
2842  * that's what the earlier WaitXLogInsertionsToFinish() returned.
2843  * We're only calling it again to allow insertpos to be moved
2844  * further forward, not to actually wait for anyone.
2845  */
2846  insertpos = WaitXLogInsertionsToFinish(insertpos);
2847  }
2848 
2849  /* try to write/flush later additions to XLOG as well */
2850  WriteRqst.Write = insertpos;
2851  WriteRqst.Flush = insertpos;
2852 
2853  XLogWrite(WriteRqst, insertTLI, false);
2854 
2855  LWLockRelease(WALWriteLock);
2856  /* done */
2857  break;
2858  }
2859 
2860  END_CRIT_SECTION();
2861 
2862  /* wake up walsenders now that we've released heavily contended locks */
2864 
2865  /*
2866  * If we still haven't flushed to the request point then we have a
2867  * problem; most likely, the requested flush point is past end of XLOG.
2868  * This has been seen to occur when a disk page has a corrupted LSN.
2869  *
2870  * Formerly we treated this as a PANIC condition, but that hurts the
2871  * system's robustness rather than helping it: we do not want to take down
2872  * the whole system due to corruption on one data page. In particular, if
2873  * the bad page is encountered again during recovery then we would be
2874  * unable to restart the database at all! (This scenario actually
2875  * happened in the field several times with 7.1 releases.) As of 8.4, bad
2876  * LSNs encountered during recovery are UpdateMinRecoveryPoint's problem;
2877  * the only time we can reach here during recovery is while flushing the
2878  * end-of-recovery checkpoint record, and we don't expect that to have a
2879  * bad LSN.
2880  *
2881  * Note that for calls from xact.c, the ERROR will be promoted to PANIC
2882  * since xact.c calls this routine inside a critical section. However,
2883  * calls from bufmgr.c are not within critical sections and so we will not
2884  * force a restart for a bad LSN on a data page.
2885  */
2886  if (LogwrtResult.Flush < record)
2887  elog(ERROR,
2888  "xlog flush request %X/%X is not satisfied --- flushed only to %X/%X",
2889  LSN_FORMAT_ARGS(record),
2891 }
bool LWLockAcquireOrWait(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1400
bool MinimumActiveBackends(int min)
Definition: procarray.c:3533
int CommitDelay
Definition: xlog.c:132
int CommitSiblings
Definition: xlog.c:133
bool XLogInsertAllowed(void)
Definition: xlog.c:6256

References CommitDelay, CommitSiblings, elog, enableFsync, END_CRIT_SECTION, ERROR, XLogwrtRqst::Flush, XLogwrtResult::Flush, XLogCtlData::info_lck, XLogCtlData::InsertTimeLineID, LOG, XLogCtlData::LogwrtResult, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, LW_EXCLUSIVE, LWLockAcquireOrWait(), LWLockRelease(), MinimumActiveBackends(), pg_usleep(), RecoveryInProgress(), 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 3688 of file xlog.c.

3689 {
3690  XLogSegNo lastRemovedSegNo;
3691 
3693  lastRemovedSegNo = XLogCtl->lastRemovedSegNo;
3695 
3696  return lastRemovedSegNo;
3697 }

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 3704 of file xlog.c.

3705 {
3706  DIR *xldir;
3707  struct dirent *xlde;
3708  XLogSegNo oldest_segno = 0;
3709 
3710  xldir = AllocateDir(XLOGDIR);
3711  while ((xlde = ReadDir(xldir, XLOGDIR)) != NULL)
3712  {
3713  TimeLineID file_tli;
3714  XLogSegNo file_segno;
3715 
3716  /* Ignore files that are not XLOG segments. */
3717  if (!IsXLogFileName(xlde->d_name))
3718  continue;
3719 
3720  /* Parse filename to get TLI and segno. */
3721  XLogFromFileName(xlde->d_name, &file_tli, &file_segno,
3723 
3724  /* Ignore anything that's not from the TLI of interest. */
3725  if (tli != file_tli)
3726  continue;
3727 
3728  /* If it's the oldest so far, update oldest_segno. */
3729  if (oldest_segno == 0 || file_segno < oldest_segno)
3730  oldest_segno = file_segno;
3731  }
3732 
3733  FreeDir(xldir);
3734  return oldest_segno;
3735 }
static void XLogFromFileName(const char *fname, TimeLineID *tli, XLogSegNo *logSegNo, int wal_segsz_bytes)

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

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

◆ XLogInsertAllowed()

bool XLogInsertAllowed ( void  )

Definition at line 6256 of file xlog.c.

6257 {
6258  /*
6259  * If value is "unconditionally true" or "unconditionally false", just
6260  * return it. This provides the normal fast path once recovery is known
6261  * done.
6262  */
6263  if (LocalXLogInsertAllowed >= 0)
6264  return (bool) LocalXLogInsertAllowed;
6265 
6266  /*
6267  * Else, must check to see if we're still in recovery.
6268  */
6269  if (RecoveryInProgress())
6270  return false;
6271 
6272  /*
6273  * On exit from recovery, reset to "unconditionally true", since there is
6274  * no need to keep checking.
6275  */
6277  return true;
6278 }

References LocalXLogInsertAllowed, and RecoveryInProgress().

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

◆ XLogInsertRecord()

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

Definition at line 737 of file xlog.c.

742 {
744  pg_crc32c rdata_crc;
745  bool inserted;
746  XLogRecord *rechdr = (XLogRecord *) rdata->data;
747  uint8 info = rechdr->xl_info & ~XLR_INFO_MASK;
749  XLogRecPtr StartPos;
750  XLogRecPtr EndPos;
751  bool prevDoPageWrites = doPageWrites;
752  TimeLineID insertTLI;
753 
754  /* Does this record type require special handling? */
755  if (unlikely(rechdr->xl_rmid == RM_XLOG_ID))
756  {
757  if (info == XLOG_SWITCH)
758  class = WALINSERT_SPECIAL_SWITCH;
759  else if (info == XLOG_CHECKPOINT_REDO)
761  }
762 
763  /* we assume that all of the record header is in the first chunk */
764  Assert(rdata->len >= SizeOfXLogRecord);
765 
766  /* cross-check on whether we should be here or not */
767  if (!XLogInsertAllowed())
768  elog(ERROR, "cannot make new WAL entries during recovery");
769 
770  /*
771  * Given that we're not in recovery, InsertTimeLineID is set and can't
772  * change, so we can read it without a lock.
773  */
774  insertTLI = XLogCtl->InsertTimeLineID;
775 
776  /*----------
777  *
778  * We have now done all the preparatory work we can without holding a
779  * lock or modifying shared state. From here on, inserting the new WAL
780  * record to the shared WAL buffer cache is a two-step process:
781  *
782  * 1. Reserve the right amount of space from the WAL. The current head of
783  * reserved space is kept in Insert->CurrBytePos, and is protected by
784  * insertpos_lck.
785  *
786  * 2. Copy the record to the reserved WAL space. This involves finding the
787  * correct WAL buffer containing the reserved space, and copying the
788  * record in place. This can be done concurrently in multiple processes.
789  *
790  * To keep track of which insertions are still in-progress, each concurrent
791  * inserter acquires an insertion lock. In addition to just indicating that
792  * an insertion is in progress, the lock tells others how far the inserter
793  * has progressed. There is a small fixed number of insertion locks,
794  * determined by NUM_XLOGINSERT_LOCKS. When an inserter crosses a page
795  * boundary, it updates the value stored in the lock to the how far it has
796  * inserted, to allow the previous buffer to be flushed.
797  *
798  * Holding onto an insertion lock also protects RedoRecPtr and
799  * fullPageWrites from changing until the insertion is finished.
800  *
801  * Step 2 can usually be done completely in parallel. If the required WAL
802  * page is not initialized yet, you have to grab WALBufMappingLock to
803  * initialize it, but the WAL writer tries to do that ahead of insertions
804  * to avoid that from happening in the critical path.
805  *
806  *----------
807  */
809 
810  if (likely(class == WALINSERT_NORMAL))
811  {
813 
814  /*
815  * Check to see if my copy of RedoRecPtr is out of date. If so, may
816  * have to go back and have the caller recompute everything. This can
817  * only happen just after a checkpoint, so it's better to be slow in
818  * this case and fast otherwise.
819  *
820  * Also check to see if fullPageWrites was just turned on or there's a
821  * running backup (which forces full-page writes); if we weren't
822  * already doing full-page writes then go back and recompute.
823  *
824  * If we aren't doing full-page writes then RedoRecPtr doesn't
825  * actually affect the contents of the XLOG record, so we'll update
826  * our local copy but not force a recomputation. (If doPageWrites was
827  * just turned off, we could recompute the record without full pages,
828  * but we choose not to bother.)
829  */
830  if (RedoRecPtr != Insert->RedoRecPtr)
831  {
832  Assert(RedoRecPtr < Insert->RedoRecPtr);
833  RedoRecPtr = Insert->RedoRecPtr;
834  }
835  doPageWrites = (Insert->fullPageWrites || Insert->runningBackups > 0);
836 
837  if (doPageWrites &&
838  (!prevDoPageWrites ||
839  (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr)))
840  {
841  /*
842  * Oops, some buffer now needs to be backed up that the caller
843  * didn't back up. Start over.
844  */
847  return InvalidXLogRecPtr;
848  }
849 
850  /*
851  * Reserve space for the record in the WAL. This also sets the xl_prev
852  * pointer.
853  */
854  ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
855  &rechdr->xl_prev);
856 
857  /* Normal records are always inserted. */
858  inserted = true;
859  }
860  else if (class == WALINSERT_SPECIAL_SWITCH)
861  {
862  /*
863  * In order to insert an XLOG_SWITCH record, we need to hold all of
864  * the WAL insertion locks, not just one, so that no one else can
865  * begin inserting a record until we've figured out how much space
866  * remains in the current WAL segment and claimed all of it.
867  *
868  * Nonetheless, this case is simpler than the normal cases handled
869  * below, which must check for changes in doPageWrites and RedoRecPtr.
870  * Those checks are only needed for records that can contain buffer
871  * references, and an XLOG_SWITCH record never does.
872  */
873  Assert(fpw_lsn == InvalidXLogRecPtr);
875  inserted = ReserveXLogSwitch(&StartPos, &EndPos, &rechdr->xl_prev);
876  }
877  else
878  {
880 
881  /*
882  * We need to update both the local and shared copies of RedoRecPtr,
883  * which means that we need to hold all the WAL insertion locks.
884  * However, there can't be any buffer references, so as above, we need
885  * not check RedoRecPtr before inserting the record; we just need to
886  * update it afterwards.
887  */
888  Assert(fpw_lsn == InvalidXLogRecPtr);
890  ReserveXLogInsertLocation(rechdr->xl_tot_len, &StartPos, &EndPos,
891  &rechdr->xl_prev);
892  RedoRecPtr = Insert->RedoRecPtr = StartPos;
893  inserted = true;
894  }
895 
896  if (inserted)
897  {
898  /*
899  * Now that xl_prev has been filled in, calculate CRC of the record
900  * header.
901  */
902  rdata_crc = rechdr->xl_crc;
903  COMP_CRC32C(rdata_crc, rechdr, offsetof(XLogRecord, xl_crc));
904  FIN_CRC32C(rdata_crc);
905  rechdr->xl_crc = rdata_crc;
906 
907  /*
908  * All the record data, including the header, is now ready to be
909  * inserted. Copy the record in the space reserved.
910  */
912  class == WALINSERT_SPECIAL_SWITCH, rdata,
913  StartPos, EndPos, insertTLI);
914 
915  /*
916  * Unless record is flagged as not important, update LSN of last
917  * important record in the current slot. When holding all locks, just
918  * update the first one.
919  */
920  if ((flags & XLOG_MARK_UNIMPORTANT) == 0)
921  {
922  int lockno = holdingAllLocks ? 0 : MyLockNo;
923 
924  WALInsertLocks[lockno].l.lastImportantAt = StartPos;
925  }
926  }
927  else
928  {
929  /*
930  * This was an xlog-switch record, but the current insert location was
931  * already exactly at the beginning of a segment, so there was no need
932  * to do anything.
933  */
934  }
935 
936  /*
937  * Done! Let others know that we're finished.
938  */
940 
942 
944 
945  /*
946  * Mark top transaction id is logged (if needed) so that we should not try
947  * to log it again with the next WAL record in the current subtransaction.
948  */
949  if (topxid_included)
951 
952  /*
953  * Update shared LogwrtRqst.Write, if we crossed page boundary.
954  */
955  if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
956  {
958  /* advance global request to include new block(s) */
959  if (XLogCtl->LogwrtRqst.Write < EndPos)
960  XLogCtl->LogwrtRqst.Write = EndPos;
961  /* update local result copy while I have the chance */
964  }
965 
966  /*
967  * If this was an XLOG_SWITCH record, flush the record and the empty
968  * padding space that fills the rest of the segment, and perform
969  * end-of-segment actions (eg, notifying archiver).
970  */
971  if (class == WALINSERT_SPECIAL_SWITCH)
972  {
973  TRACE_POSTGRESQL_WAL_SWITCH();
974  XLogFlush(EndPos);
975 
976  /*
977  * Even though we reserved the rest of the segment for us, which is
978  * reflected in EndPos, we return a pointer to just the end of the
979  * xlog-switch record.
980  */
981  if (inserted)
982  {
983  EndPos = StartPos + SizeOfXLogRecord;
984  if (StartPos / XLOG_BLCKSZ != EndPos / XLOG_BLCKSZ)
985  {
986  uint64 offset = XLogSegmentOffset(EndPos, wal_segment_size);
987 
988  if (offset == EndPos % XLOG_BLCKSZ)
989  EndPos += SizeOfXLogLongPHD;
990  else
991  EndPos += SizeOfXLogShortPHD;
992  }
993  }
994  }
995 
996 #ifdef WAL_DEBUG
997  if (XLOG_DEBUG)
998  {
999  static XLogReaderState *debug_reader = NULL;
1000  XLogRecord *record;
1001  DecodedXLogRecord *decoded;
1003  StringInfoData recordBuf;
1004  char *errormsg = NULL;
1005  MemoryContext oldCxt;
1006 
1007  oldCxt = MemoryContextSwitchTo(walDebugCxt);
1008 
1009  initStringInfo(&buf);
1010  appendStringInfo(&buf, "INSERT @ %X/%X: ", LSN_FORMAT_ARGS(EndPos));
1011 
1012  /*
1013  * We have to piece together the WAL record data from the XLogRecData
1014  * entries, so that we can pass it to the rm_desc function as one
1015  * contiguous chunk.
1016  */
1017  initStringInfo(&recordBuf);
1018  for (; rdata != NULL; rdata = rdata->next)
1019  appendBinaryStringInfo(&recordBuf, rdata->data, rdata->len);
1020 
1021  /* We also need temporary space to decode the record. */
1022  record = (XLogRecord *) recordBuf.data;
1023  decoded = (DecodedXLogRecord *)
1025 
1026  if (!debug_reader)
1027  debug_reader = XLogReaderAllocate(wal_segment_size, NULL,
1028  XL_ROUTINE(.page_read = NULL,
1029  .segment_open = NULL,
1030  .segment_close = NULL),
1031  NULL);
1032  if (!debug_reader)
1033  {
1034  appendStringInfoString(&buf, "error decoding record: out of memory while allocating a WAL reading processor");
1035  }
1036  else if (!DecodeXLogRecord(debug_reader,
1037  decoded,
1038  record,
1039  EndPos,
1040  &errormsg))
1041  {
1042  appendStringInfo(&buf, "error decoding record: %s",
1043  errormsg ? errormsg : "no error message");
1044  }
1045  else
1046  {
1047  appendStringInfoString(&buf, " - ");
1048 
1049  debug_reader->record = decoded;
1050  xlog_outdesc(&buf, debug_reader);
1051  debug_reader->record = NULL;
1052  }
1053  elog(LOG, "%s", buf.data);
1054 
1055  pfree(decoded);
1056  pfree(buf.data);
1057  pfree(recordBuf.data);
1058  MemoryContextSwitchTo(oldCxt);
1059  }
1060 #endif
1061 
1062  /*
1063  * Update our global variables
1064  */
1065  ProcLastRecPtr = StartPos;
1066  XactLastRecEnd = EndPos;
1067 
1068  /* Report WAL traffic to the instrumentation. */
1069  if (inserted)
1070  {
1071  pgWalUsage.wal_bytes += rechdr->xl_tot_len;
1073  pgWalUsage.wal_fpi += num_fpi;
1074  }
1075 
1076  return EndPos;
1077 }
#define likely(x)
Definition: c.h:297
#define unlikely(x)
Definition: c.h:298
WalUsage pgWalUsage
Definition: instrument.c:22
if(TABLE==NULL||TABLE_index==NULL)
Definition: isn.c:77
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:124
void appendBinaryStringInfo(StringInfo str, const void *data, int datalen)
Definition: stringinfo.c:233
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
struct XLogRecData * next
pg_crc32c xl_crc
Definition: xlogrecord.h:49
void MarkSubxactTopXidLogged(void)
Definition: xact.c:583
void MarkCurrentTransactionIdLoggedIfAny(void)
Definition: xact.c:533
XLogRecPtr XactLastRecEnd
Definition: xlog.c:254
static void WALInsertLockAcquire(void)
Definition: xlog.c:1359
static void CopyXLogRecordToWAL(int write_len, bool isLogSwitch, XLogRecData *rdata, XLogRecPtr StartPos, XLogRecPtr EndPos, TimeLineID tli)
Definition: xlog.c:1215
static bool holdingAllLocks
Definition: xlog.c:641
static int MyLockNo
Definition: xlog.c:640
WalInsertClass
Definition: xlog.c:562
@ WALINSERT_SPECIAL_SWITCH
Definition: xlog.c:564
@ WALINSERT_NORMAL
Definition: xlog.c:563
@ WALINSERT_SPECIAL_CHECKPOINT
Definition: xlog.c:565
static void ReserveXLogInsertLocation(int size, XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
Definition: xlog.c:1098
static bool ReserveXLogSwitch(XLogRecPtr *StartPos, XLogRecPtr *EndPos, XLogRecPtr *PrevPtr)
Definition: xlog.c:1154
#define XLOG_MARK_UNIMPORTANT
Definition: xlog.h:153
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, XLogCtlData::LogwrtResult, LogwrtResult, XLogCtlData::LogwrtRqst, LSN_FORMAT_ARGS, MarkCurrentTransactionIdLoggedIfAny(), MarkSubxactTopXidLogged(), MemoryContextSwitchTo(), MyLockNo, XLogRecData::next, palloc(), pfree(), pgWalUsage, ProcLastRecPtr, XLogReaderState::record, RedoRecPtr, 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 3061 of file xlog.c.

3062 {
3063  /*
3064  * During recovery, we don't flush WAL but update minRecoveryPoint
3065  * instead. So "needs flush" is taken to mean whether minRecoveryPoint
3066  * would need to be updated.
3067  */
3068  if (RecoveryInProgress())
3069  {
3070  /*
3071  * An invalid minRecoveryPoint means that we need to recover all the
3072  * WAL, i.e., we're doing crash recovery. We never modify the control
3073  * file's value in that case, so we can short-circuit future checks
3074  * here too. This triggers a quick exit path for the startup process,
3075  * which cannot update its local copy of minRecoveryPoint as long as
3076  * it has not replayed all WAL available when doing crash recovery.
3077  */
3079  updateMinRecoveryPoint = false;
3080 
3081  /* Quick exit if already known to be updated or cannot be updated */
3083  return false;
3084 
3085  /*
3086  * Update local copy of minRecoveryPoint. But if the lock is busy,
3087  * just return a conservative guess.
3088  */
3089  if (!LWLockConditionalAcquire(ControlFileLock, LW_SHARED))
3090  return true;
3093  LWLockRelease(ControlFileLock);
3094 
3095  /*
3096  * Check minRecoveryPoint for any other process than the startup
3097  * process doing crash recovery, which should not update the control
3098  * file value if crash recovery is still running.
3099  */
3101  updateMinRecoveryPoint = false;
3102 
3103  /* check again */
3105  return false;
3106  else
3107  return true;
3108  }
3109 
3110  /* Quick exit if already known flushed */
3111  if (record <= LogwrtResult.Flush)
3112  return false;
3113 
3114  /* read LogwrtResult and update local state */
3118 
3119  /* check again */
3120  if (record <= LogwrtResult.Flush)
3121  return false;
3122 
3123  return true;
3124 }
bool LWLockConditionalAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1343

References ControlFile, XLogwrtResult::Flush, XLogCtlData::info_lck, InRecovery, LocalMinRecoveryPoint, LocalMinRecoveryPointTLI, XLogCtlData::LogwrtResult, LogwrtResult, LW_SHARED, LWLockConditionalAcquire(), LWLockRelease(), ControlFileData::minRecoveryPoint, ControlFileData::minRecoveryPointTLI, RecoveryInProgress(), SpinLockAcquire, SpinLockRelease, updateMinRecoveryPoint, XLogCtl, and XLogRecPtrIsInvalid.

Referenced by GetVictimBuffer(), and SetHintBits().

◆ XLogPutNextOid()

void XLogPutNextOid ( Oid  nextOid)

Definition at line 7857 of file xlog.c.

7858 {
7859  XLogBeginInsert();
7860  XLogRegisterData((char *) (&nextOid), sizeof(Oid));
7861  (void) XLogInsert(RM_XLOG_ID, XLOG_NEXTOID);
7862 
7863  /*
7864  * We need not flush the NEXTOID record immediately, because any of the
7865  * just-allocated OIDs could only reach disk as part of a tuple insert or
7866  * update that would have its own XLOG record that must follow the NEXTOID
7867  * record. Therefore, the standard buffer LSN interlock applied to those
7868  * records will ensure no such OID reaches disk before the NEXTOID record
7869  * does.
7870  *
7871  * Note, however, that the above statement only covers state "within" the
7872  * database. When we use a generated OID as a file or directory name, we
7873  * are in a sense violating the basic WAL rule, because that filesystem
7874  * change may reach disk before the NEXTOID WAL record does. The impact
7875  * of this is that if a database crash occurs immediately afterward, we
7876  * might after restart re-generate the same OID and find that it conflicts
7877  * with the leftover file or directory. But since for safety's sake we
7878  * always loop until finding a nonconflicting filename, this poses no real
7879  * problem in practice. See pgsql-hackers discussion 27-Sep-2006.
7880  */
7881 }

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

Referenced by GetNewObjectId().

◆ XLogRestorePoint()

XLogRecPtr XLogRestorePoint ( const char *  rpName)

Definition at line 7912 of file xlog.c.

7913 {
7914  XLogRecPtr RecPtr;
7915  xl_restore_point xlrec;
7916 
7917  xlrec.rp_time = GetCurrentTimestamp();
7918  strlcpy(xlrec.rp_name, rpName, MAXFNAMELEN);
7919 
7920  XLogBeginInsert();
7921  XLogRegisterData((char *) &xlrec, sizeof(xl_restore_point));
7922 
7923  RecPtr = XLogInsert(RM_XLOG_ID, XLOG_RESTORE_POINT);
7924 
7925  ereport(LOG,
7926  (errmsg("restore point \"%s\" created at %X/%X",
7927  rpName, LSN_FORMAT_ARGS(RecPtr))));
7928 
7929  return RecPtr;
7930 }
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45
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 2564 of file xlog.c.

2565 {
2566  XLogRecPtr WriteRqstPtr = asyncXactLSN;
2567  bool sleeping;
2568  bool wakeup = false;
2569  XLogRecPtr prevAsyncXactLSN;
2570 
2573  sleeping = XLogCtl->WalWriterSleeping;
2574  prevAsyncXactLSN = XLogCtl->asyncXactLSN;
2575  if (XLogCtl->asyncXactLSN < asyncXactLSN)
2576  XLogCtl->asyncXactLSN = asyncXactLSN;
2578 
2579  /*
2580  * If somebody else already called this function with a more aggressive
2581  * LSN, they will have done what we needed (and perhaps more).
2582  */
2583  if (asyncXactLSN <= prevAsyncXactLSN)
2584  return;
2585 
2586  /*
2587  * If the WALWriter is sleeping, kick it to make it come out of low-power
2588  * mode, so that this async commit will reach disk within the expected
2589  * amount of time. Otherwise, determine whether it has enough WAL
2590  * available to flush, the same way that XLogBackgroundFlush() does.
2591  */
2592  if (sleeping)
2593  wakeup = true;
2594  else
2595  {
2596  int flushblocks;
2597 
2598  flushblocks =
2599  WriteRqstPtr / XLOG_BLCKSZ - LogwrtResult.Flush / XLOG_BLCKSZ;
2600 
2601  if (WalWriterFlushAfter == 0 || flushblocks >= WalWriterFlushAfter)
2602  wakeup = true;
2603  }
2604 
2607 }
void SetLatch(Latch *latch)
Definition: latch.c:632
PROC_HDR * ProcGlobal
Definition: proc.c:78
Latch * walwriterLatch
Definition: proc.h:412
static TimestampTz wakeup[NUM_WALRCV_WAKEUPS]
Definition: walreceiver.c:129

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

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

◆ XLogSetReplicationSlotMinimumLSN()

void XLogSetReplicationSlotMinimumLSN ( XLogRecPtr  lsn)

◆ XLOGShmemInit()

void XLOGShmemInit ( void  )

Definition at line 4808 of file xlog.c.

4809 {
4810  bool foundCFile,
4811  foundXLog;
4812  char *allocptr;
4813  int i;
4814  ControlFileData *localControlFile;
4815 
4816 #ifdef WAL_DEBUG
4817 
4818  /*
4819  * Create a memory context for WAL debugging that's exempt from the normal
4820  * "no pallocs in critical section" rule. Yes, that can lead to a PANIC if
4821  * an allocation fails, but wal_debug is not for production use anyway.
4822  */
4823  if (walDebugCxt == NULL)
4824  {
4826  "WAL Debug",
4828  MemoryContextAllowInCriticalSection(walDebugCxt, true);
4829  }
4830 #endif
4831 
4832 
4833  XLogCtl = (XLogCtlData *)
4834  ShmemInitStruct("XLOG Ctl", XLOGShmemSize(), &foundXLog);
4835 
4836  localControlFile = ControlFile;
4838  ShmemInitStruct("Control File", sizeof(ControlFileData), &foundCFile);
4839 
4840  if (foundCFile || foundXLog)
4841  {
4842  /* both should be present or neither */
4843  Assert(foundCFile && foundXLog);
4844 
4845  /* Initialize local copy of WALInsertLocks */
4847 
4848  if (localControlFile)
4849  pfree(localControlFile);
4850  return;
4851  }
4852  memset(XLogCtl, 0, sizeof(XLogCtlData));
4853 
4854  /*
4855  * Already have read control file locally, unless in bootstrap mode. Move
4856  * contents into shared memory.
4857  */
4858  if (localControlFile)
4859  {
4860  memcpy(ControlFile, localControlFile, sizeof(ControlFileData));
4861  pfree(localControlFile);
4862  }
4863 
4864  /*
4865  * Since XLogCtlData contains XLogRecPtr fields, its sizeof should be a
4866  * multiple of the alignment for same, so no extra alignment padding is
4867  * needed here.
4868  */
4869  allocptr = ((char *) XLogCtl) + sizeof(XLogCtlData);
4870  XLogCtl->xlblocks = (pg_atomic_uint64 *) allocptr;
4871  allocptr += sizeof(pg_atomic_uint64) * XLOGbuffers;
4872 
4873  for (i = 0; i < XLOGbuffers; i++)
4874  {
4876  }
4877 
4878  /* WAL insertion locks. Ensure they're aligned to the full padded size */
4879  allocptr += sizeof(WALInsertLockPadded) -
4880  ((uintptr_t) allocptr) % sizeof(WALInsertLockPadded);
4882  (WALInsertLockPadded *) allocptr;
4883  allocptr += sizeof(WALInsertLockPadded) * NUM_XLOGINSERT_LOCKS;
4884 
4885  for (i = 0; i < NUM_XLOGINSERT_LOCKS; i++)
4886  {
4890  }
4891 
4892  /*
4893  * Align the start of the page buffers to a full xlog block size boundary.
4894  * This simplifies some calculations in XLOG insertion. It is also
4895  * required for O_DIRECT.
4896  */
4897  allocptr = (char *) TYPEALIGN(XLOG_BLCKSZ, allocptr);
4898  XLogCtl->pages = allocptr;
4899  memset(XLogCtl->pages, 0, (Size) XLOG_BLCKSZ * XLOGbuffers);
4900 
4901  /*
4902  * Do basic initialization of XLogCtl shared data. (StartupXLOG will fill
4903  * in additional info.)
4904  */
4908  XLogCtl->WalWriterSleeping = false;
4909 
4913 }
static void pg_atomic_init_u64(volatile pg_atomic_uint64 *ptr, uint64 val)
Definition: atomics.h:448
struct pg_atomic_uint64 pg_atomic_uint64
void LWLockInitialize(LWLock *lock, int tranche_id)
Definition: lwlock.c:707
@ LWTRANCHE_WAL_INSERT
Definition: lwlock.h:188
MemoryContext TopMemoryContext
Definition: mcxt.c:137
void MemoryContextAllowInCriticalSection(MemoryContext context, bool allow)
Definition: mcxt.c:682
#define AllocSetContextCreate
Definition: memutils.h:129
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:153
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:387
#define SpinLockInit(lock)
Definition: spin.h:60
int XLogCacheBlck
Definition: xlog.c:501
WALInsertLockPadded * WALInsertLocks
Definition: xlog.c:450
slock_t insertpos_lck
Definition: xlog.c:404
union WALInsertLockPadded WALInsertLockPadded
Size XLOGShmemSize(void)
Definition: xlog.c:4758
int XLOGbuffers
Definition: xlog.c:117
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, 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 4758 of file xlog.c.

4759 {
4760  Size size;
4761 
4762  /*
4763  * If the value of wal_buffers is -1, use the preferred auto-tune value.
4764  * This isn't an amazingly clean place to do this, but we must wait till
4765  * NBuffers has received its final value, and must do it before using the
4766  * value of XLOGbuffers to do anything important.
4767  *
4768  * We prefer to report this value's source as PGC_S_DYNAMIC_DEFAULT.
4769  * However, if the DBA explicitly set wal_buffers = -1 in the config file,
4770  * then PGC_S_DYNAMIC_DEFAULT will fail to override that and we must force
4771  * the matter with PGC_S_OVERRIDE.
4772  */
4773  if (XLOGbuffers == -1)
4774  {
4775  char buf[32];
4776 
4777  snprintf(buf, sizeof(buf), "%d", XLOGChooseNumBuffers());
4778  SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4780  if (XLOGbuffers == -1) /* failed to apply it? */
4781  SetConfigOption("wal_buffers", buf, PGC_POSTMASTER,
4782  PGC_S_OVERRIDE);
4783  }
4784  Assert(XLOGbuffers > 0);
4785 
4786  /* XLogCtl */
4787  size = sizeof(XLogCtlData);
4788 
4789  /* WAL insertion locks, plus alignment */
4791  /* xlblocks array */
4793  /* extra alignment padding for XLOG I/O buffers */
4794  size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
4795  /* and the buffers themselves */
4796  size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
4797 
4798  /*
4799  * Note: we don't count ControlFileData, it comes out of the "slop factor"
4800  * added by CreateSharedMemoryAndSemaphores. This lets us use this
4801  * routine again below to compute the actual allocation size.
4802  */
4803 
4804  return size;
4805 }
#define Max(x, y)
Definition: c.h:985
void SetConfigOption(const char *name, const char *value, GucContext context, GucSource source)
Definition: guc.c:4275
@ PGC_S_DYNAMIC_DEFAULT
Definition: guc.h:110
@ 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
static pg_noinline void Size size
Definition: slab.c:607
static int XLOGChooseNumBuffers(void)
Definition: xlog.c:4511

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 9306 of file xlog.c.

9307 {
9308  ShutdownWalRcv();
9309 
9310  LWLockAcquire(ControlFileLock, LW_EXCLUSIVE);
9312  LWLockRelease(ControlFileLock);
9313 }
void ShutdownWalRcv(void)

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

Referenced by FinishWalRecovery(), and WaitForWALToBecomeAvailable().

Variable Documentation

◆ CheckPointSegments

PGDLLIMPORT int CheckPointSegments
extern

◆ CheckpointStats

◆ EnableHotStandby

◆ fullPageWrites

PGDLLIMPORT bool fullPageWrites
extern

Definition at line 122 of file xlog.c.

Referenced by BootStrapXLOG(), and UpdateFullPageWrites().

◆ log_checkpoints

PGDLLIMPORT bool log_checkpoints
extern

◆ max_slot_wal_keep_size_mb

PGDLLIMPORT int max_slot_wal_keep_size_mb
extern

Definition at line 135 of file xlog.c.

Referenced by KeepLogSeg(), and pg_get_replication_slots().

◆ max_wal_size_mb

PGDLLIMPORT int max_wal_size_mb
extern

◆ min_wal_size_mb

PGDLLIMPORT int min_wal_size_mb
extern

Definition at line 115 of file xlog.c.

Referenced by ReadControlFile(), and XLOGfileslop().

◆ ProcLastRecPtr

PGDLLIMPORT XLogRecPtr ProcLastRecPtr
extern

◆ track_wal_io_timing

PGDLLIMPORT bool track_wal_io_timing
extern

Definition at line 137 of file xlog.c.

Referenced by issue_xlog_fsync(), and XLogWrite().

◆ wal_compression

PGDLLIMPORT int wal_compression
extern

Definition at line 124 of file xlog.c.

Referenced by XLogCompressBackupBlock(), and XLogRecordAssemble().

◆ wal_consistency_checking

PGDLLIMPORT bool* wal_consistency_checking
extern

Definition at line 126 of file xlog.c.

Referenced by assign_wal_consistency_checking(), and XLogRecordAssemble().

◆ wal_consistency_checking_string

PGDLLIMPORT char* wal_consistency_checking_string
extern

Definition at line 125 of file xlog.c.

Referenced by InitializeWalConsistencyChecking().

◆ wal_decode_buffer_size

PGDLLIMPORT int wal_decode_buffer_size
extern

Definition at line 136 of file xlog.c.

Referenced by InitWalRecovery().

◆ wal_init_zero

PGDLLIMPORT bool wal_init_zero
extern

Definition at line 127 of file xlog.c.

Referenced by XLogFileInitInternal().

◆ wal_keep_size_mb

PGDLLIMPORT int wal_keep_size_mb
extern

Definition at line 116 of file xlog.c.

Referenced by KeepLogSeg(), and pg_get_replication_slots().

◆ wal_level

◆ wal_log_hints

PGDLLIMPORT bool wal_log_hints
extern

Definition at line 123 of file xlog.c.

Referenced by InitControlFile(), and XLogReportParameters().

◆ wal_recycle

PGDLLIMPORT bool wal_recycle
extern

Definition at line 128 of file xlog.c.

Referenced by RemoveXlogFile().

◆ wal_retrieve_retry_interval

PGDLLIMPORT int wal_retrieve_retry_interval
extern

◆ wal_segment_size

PGDLLIMPORT int wal_segment_size
extern

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

PGDLLIMPORT int wal_sync_method
extern

◆ XactLastCommitEnd

◆ XactLastRecEnd

◆ XLogArchiveCommand

PGDLLIMPORT char* XLogArchiveCommand
extern

◆ XLogArchiveMode

◆ XLogArchiveTimeout

PGDLLIMPORT int XLogArchiveTimeout
extern

Definition at line 118 of file xlog.c.

Referenced by CheckArchiveTimeout(), and CheckpointerMain().

◆ XLOGbuffers

PGDLLIMPORT int XLOGbuffers
extern

Definition at line 117 of file xlog.c.

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