PostgreSQL Source Code  git master
walreceiverfuncs.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * walreceiverfuncs.c
4  *
5  * This file contains functions used by the startup process to communicate
6  * with the walreceiver process. Functions implementing walreceiver itself
7  * are in walreceiver.c.
8  *
9  * Portions Copyright (c) 2010-2020, PostgreSQL Global Development Group
10  *
11  *
12  * IDENTIFICATION
13  * src/backend/replication/walreceiverfuncs.c
14  *
15  *-------------------------------------------------------------------------
16  */
17 #include "postgres.h"
18 
19 #include <sys/stat.h>
20 #include <sys/time.h>
21 #include <time.h>
22 #include <unistd.h>
23 #include <signal.h>
24 
25 #include "access/xlog_internal.h"
26 #include "postmaster/startup.h"
28 #include "storage/pmsignal.h"
29 #include "storage/shmem.h"
30 #include "utils/timestamp.h"
31 
33 
34 /*
35  * How long to wait for walreceiver to start up after requesting
36  * postmaster to launch it. In seconds.
37  */
38 #define WALRCV_STARTUP_TIMEOUT 10
39 
40 /* Report shared memory space needed by WalRcvShmemInit */
41 Size
43 {
44  Size size = 0;
45 
46  size = add_size(size, sizeof(WalRcvData));
47 
48  return size;
49 }
50 
51 /* Allocate and initialize walreceiver-related shared memory */
52 void
54 {
55  bool found;
56 
57  WalRcv = (WalRcvData *)
58  ShmemInitStruct("Wal Receiver Ctl", WalRcvShmemSize(), &found);
59 
60  if (!found)
61  {
62  /* First time through, so initialize */
63  MemSet(WalRcv, 0, WalRcvShmemSize());
64  WalRcv->walRcvState = WALRCV_STOPPED;
65  SpinLockInit(&WalRcv->mutex);
66  WalRcv->latch = NULL;
67  }
68 }
69 
70 /* Is walreceiver running (or starting up)? */
71 bool
73 {
74  WalRcvData *walrcv = WalRcv;
76  pg_time_t startTime;
77 
78  SpinLockAcquire(&walrcv->mutex);
79 
80  state = walrcv->walRcvState;
81  startTime = walrcv->startTime;
82 
83  SpinLockRelease(&walrcv->mutex);
84 
85  /*
86  * If it has taken too long for walreceiver to start up, give up. Setting
87  * the state to STOPPED ensures that if walreceiver later does start up
88  * after all, it will see that it's not supposed to be running and die
89  * without doing anything.
90  */
91  if (state == WALRCV_STARTING)
92  {
93  pg_time_t now = (pg_time_t) time(NULL);
94 
95  if ((now - startTime) > WALRCV_STARTUP_TIMEOUT)
96  {
97  SpinLockAcquire(&walrcv->mutex);
98 
99  if (walrcv->walRcvState == WALRCV_STARTING)
100  state = walrcv->walRcvState = WALRCV_STOPPED;
101 
102  SpinLockRelease(&walrcv->mutex);
103  }
104  }
105 
106  if (state != WALRCV_STOPPED)
107  return true;
108  else
109  return false;
110 }
111 
112 /*
113  * Is walreceiver running and streaming (or at least attempting to connect,
114  * or starting up)?
115  */
116 bool
118 {
119  WalRcvData *walrcv = WalRcv;
121  pg_time_t startTime;
122 
123  SpinLockAcquire(&walrcv->mutex);
124 
125  state = walrcv->walRcvState;
126  startTime = walrcv->startTime;
127 
128  SpinLockRelease(&walrcv->mutex);
129 
130  /*
131  * If it has taken too long for walreceiver to start up, give up. Setting
132  * the state to STOPPED ensures that if walreceiver later does start up
133  * after all, it will see that it's not supposed to be running and die
134  * without doing anything.
135  */
136  if (state == WALRCV_STARTING)
137  {
138  pg_time_t now = (pg_time_t) time(NULL);
139 
140  if ((now - startTime) > WALRCV_STARTUP_TIMEOUT)
141  {
142  SpinLockAcquire(&walrcv->mutex);
143 
144  if (walrcv->walRcvState == WALRCV_STARTING)
145  state = walrcv->walRcvState = WALRCV_STOPPED;
146 
147  SpinLockRelease(&walrcv->mutex);
148  }
149  }
150 
151  if (state == WALRCV_STREAMING || state == WALRCV_STARTING ||
152  state == WALRCV_RESTARTING)
153  return true;
154  else
155  return false;
156 }
157 
158 /*
159  * Stop walreceiver (if running) and wait for it to die.
160  * Executed by the Startup process.
161  */
162 void
164 {
165  WalRcvData *walrcv = WalRcv;
166  pid_t walrcvpid = 0;
167 
168  /*
169  * Request walreceiver to stop. Walreceiver will switch to WALRCV_STOPPED
170  * mode once it's finished, and will also request postmaster to not
171  * restart itself.
172  */
173  SpinLockAcquire(&walrcv->mutex);
174  switch (walrcv->walRcvState)
175  {
176  case WALRCV_STOPPED:
177  break;
178  case WALRCV_STARTING:
179  walrcv->walRcvState = WALRCV_STOPPED;
180  break;
181 
182  case WALRCV_STREAMING:
183  case WALRCV_WAITING:
184  case WALRCV_RESTARTING:
185  walrcv->walRcvState = WALRCV_STOPPING;
186  /* fall through */
187  case WALRCV_STOPPING:
188  walrcvpid = walrcv->pid;
189  break;
190  }
191  SpinLockRelease(&walrcv->mutex);
192 
193  /*
194  * Signal walreceiver process if it was still running.
195  */
196  if (walrcvpid != 0)
197  kill(walrcvpid, SIGTERM);
198 
199  /*
200  * Wait for walreceiver to acknowledge its death by setting state to
201  * WALRCV_STOPPED.
202  */
203  while (WalRcvRunning())
204  {
205  /*
206  * This possibly-long loop needs to handle interrupts of startup
207  * process.
208  */
210 
211  pg_usleep(100000); /* 100ms */
212  }
213 }
214 
215 /*
216  * Request postmaster to start walreceiver.
217  *
218  * "recptr" indicates the position where streaming should begin. "conninfo"
219  * is a libpq connection string to use. "slotname" is, optionally, the name
220  * of a replication slot to acquire. "create_temp_slot" indicates to create
221  * a temporary slot when no "slotname" is given.
222  *
223  * WAL receivers do not directly load GUC parameters used for the connection
224  * to the primary, and rely on the values passed down by the caller of this
225  * routine instead. Hence, the addition of any new parameters should happen
226  * through this code path.
227  */
228 void
229 RequestXLogStreaming(TimeLineID tli, XLogRecPtr recptr, const char *conninfo,
230  const char *slotname, bool create_temp_slot)
231 {
232  WalRcvData *walrcv = WalRcv;
233  bool launch = false;
234  pg_time_t now = (pg_time_t) time(NULL);
235  Latch *latch;
236 
237  /*
238  * We always start at the beginning of the segment. That prevents a broken
239  * segment (i.e., with no records in the first half of a segment) from
240  * being created by XLOG streaming, which might cause trouble later on if
241  * the segment is e.g archived.
242  */
243  if (XLogSegmentOffset(recptr, wal_segment_size) != 0)
244  recptr -= XLogSegmentOffset(recptr, wal_segment_size);
245 
246  SpinLockAcquire(&walrcv->mutex);
247 
248  /* It better be stopped if we try to restart it */
249  Assert(walrcv->walRcvState == WALRCV_STOPPED ||
250  walrcv->walRcvState == WALRCV_WAITING);
251 
252  if (conninfo != NULL)
253  strlcpy((char *) walrcv->conninfo, conninfo, MAXCONNINFO);
254  else
255  walrcv->conninfo[0] = '\0';
256 
257  /*
258  * Use configured replication slot if present, and ignore the value of
259  * create_temp_slot as the slot name should be persistent. Otherwise, use
260  * create_temp_slot to determine whether this WAL receiver should create a
261  * temporary slot by itself and use it, or not.
262  */
263  if (slotname != NULL && slotname[0] != '\0')
264  {
265  strlcpy((char *) walrcv->slotname, slotname, NAMEDATALEN);
266  walrcv->is_temp_slot = false;
267  }
268  else
269  {
270  walrcv->slotname[0] = '\0';
271  walrcv->is_temp_slot = create_temp_slot;
272  }
273 
274  if (walrcv->walRcvState == WALRCV_STOPPED)
275  {
276  launch = true;
277  walrcv->walRcvState = WALRCV_STARTING;
278  }
279  else
280  walrcv->walRcvState = WALRCV_RESTARTING;
281  walrcv->startTime = now;
282 
283  /*
284  * If this is the first startup of walreceiver (on this timeline),
285  * initialize flushedUpto and latestChunkStart to the starting point.
286  */
287  if (walrcv->receiveStart == 0 || walrcv->receivedTLI != tli)
288  {
289  walrcv->flushedUpto = recptr;
290  walrcv->receivedTLI = tli;
291  walrcv->latestChunkStart = recptr;
292  }
293  walrcv->receiveStart = recptr;
294  walrcv->receiveStartTLI = tli;
295 
296  latch = walrcv->latch;
297 
298  SpinLockRelease(&walrcv->mutex);
299 
300  if (launch)
302  else if (latch)
303  SetLatch(latch);
304 }
305 
306 /*
307  * Returns the last+1 byte position that walreceiver has flushed.
308  *
309  * Optionally, returns the previous chunk start, that is the first byte
310  * written in the most recent walreceiver flush cycle. Callers not
311  * interested in that value may pass NULL for latestChunkStart. Same for
312  * receiveTLI.
313  */
316 {
317  WalRcvData *walrcv = WalRcv;
318  XLogRecPtr recptr;
319 
320  SpinLockAcquire(&walrcv->mutex);
321  recptr = walrcv->flushedUpto;
322  if (latestChunkStart)
323  *latestChunkStart = walrcv->latestChunkStart;
324  if (receiveTLI)
325  *receiveTLI = walrcv->receivedTLI;
326  SpinLockRelease(&walrcv->mutex);
327 
328  return recptr;
329 }
330 
331 /*
332  * Returns the last+1 byte position that walreceiver has written.
333  * This returns a recently written value without taking a lock.
334  */
337 {
338  WalRcvData *walrcv = WalRcv;
339 
340  return pg_atomic_read_u64(&walrcv->writtenUpto);
341 }
342 
343 /*
344  * Returns the replication apply delay in ms or -1
345  * if the apply delay info is not available
346  */
347 int
349 {
350  WalRcvData *walrcv = WalRcv;
351  XLogRecPtr receivePtr;
352  XLogRecPtr replayPtr;
353  TimestampTz chunkReplayStartTime;
354 
355  SpinLockAcquire(&walrcv->mutex);
356  receivePtr = walrcv->flushedUpto;
357  SpinLockRelease(&walrcv->mutex);
358 
359  replayPtr = GetXLogReplayRecPtr(NULL);
360 
361  if (receivePtr == replayPtr)
362  return 0;
363 
364  chunkReplayStartTime = GetCurrentChunkReplayStartTime();
365 
366  if (chunkReplayStartTime == 0)
367  return -1;
368 
369  return TimestampDifferenceMilliseconds(chunkReplayStartTime,
371 }
372 
373 /*
374  * Returns the network latency in ms, note that this includes any
375  * difference in clock settings between the servers, as well as timezone.
376  */
377 int
379 {
380  WalRcvData *walrcv = WalRcv;
381  TimestampTz lastMsgSendTime;
382  TimestampTz lastMsgReceiptTime;
383 
384  SpinLockAcquire(&walrcv->mutex);
385  lastMsgSendTime = walrcv->lastMsgSendTime;
386  lastMsgReceiptTime = walrcv->lastMsgReceiptTime;
387  SpinLockRelease(&walrcv->mutex);
388 
389  return TimestampDifferenceMilliseconds(lastMsgSendTime,
390  lastMsgReceiptTime);
391 }
XLogRecPtr GetWalRcvWriteRecPtr(void)
uint32 TimeLineID
Definition: xlogdefs.h:52
int64 pg_time_t
Definition: pgtime.h:23
slock_t mutex
Definition: walreceiver.h:143
int wal_segment_size
Definition: xlog.c:117
int GetReplicationTransferLatency(void)
TimestampTz GetCurrentTimestamp(void)
Definition: timestamp.c:1578
int64 TimestampTz
Definition: timestamp.h:39
WalRcvState walRcvState
Definition: walreceiver.h:64
#define SpinLockInit(lock)
Definition: spin.h:60
TimeLineID receivedTLI
Definition: walreceiver.h:84
#define MemSet(start, val, len)
Definition: c.h:1004
#define kill(pid, sig)
Definition: win32_port.h:454
void WalRcvShmemInit(void)
void SetLatch(Latch *latch)
Definition: latch.c:505
TimestampTz lastMsgReceiptTime
Definition: walreceiver.h:98
TimestampTz lastMsgSendTime
Definition: walreceiver.h:97
pg_time_t startTime
Definition: walreceiver.h:65
WalRcvState
Definition: walreceiver.h:44
pg_atomic_uint64 writtenUpto
Definition: walreceiver.h:151
#define NAMEDATALEN
#define SpinLockAcquire(lock)
Definition: spin.h:62
void pg_usleep(long microsec)
Definition: signal.c:53
TimestampTz GetCurrentChunkReplayStartTime(void)
Definition: xlog.c:6199
void RequestXLogStreaming(TimeLineID tli, XLogRecPtr recptr, const char *conninfo, const char *slotname, bool create_temp_slot)
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:392
XLogRecPtr GetXLogReplayRecPtr(TimeLineID *replayTLI)
Definition: xlog.c:11480
pid_t pid
Definition: walreceiver.h:63
#define MAXCONNINFO
Definition: walreceiver.h:36
XLogRecPtr latestChunkStart
Definition: walreceiver.h:92
Definition: latch.h:110
TimeLineID receiveStartTLI
Definition: walreceiver.h:74
Latch * latch
Definition: walreceiver.h:141
int GetReplicationApplyDelay(void)
void ShutdownWalRcv(void)
#define SpinLockRelease(lock)
Definition: spin.h:64
Size add_size(Size s1, Size s2)
Definition: shmem.c:498
static TimeLineID receiveTLI
Definition: xlog.c:214
#define XLogSegmentOffset(xlogptr, wal_segsz_bytes)
#define WALRCV_STARTUP_TIMEOUT
Size WalRcvShmemSize(void)
size_t strlcpy(char *dst, const char *src, size_t siz)
Definition: strlcpy.c:45
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:800
Definition: regguts.h:298
bool WalRcvStreaming(void)
size_t Size
Definition: c.h:528
XLogRecPtr GetWalRcvFlushRecPtr(XLogRecPtr *latestChunkStart, TimeLineID *receiveTLI)
bool is_temp_slot
Definition: walreceiver.h:129
WalRcvData * WalRcv
static uint64 pg_atomic_read_u64(volatile pg_atomic_uint64 *ptr)
Definition: atomics.h:429
void HandleStartupProcInterrupts(void)
Definition: startup.c:121
bool WalRcvRunning(void)
void SendPostmasterSignal(PMSignalReason reason)
Definition: pmsignal.c:146
XLogRecPtr receiveStart
Definition: walreceiver.h:73
XLogRecPtr flushedUpto
Definition: walreceiver.h:83
Datum now(PG_FUNCTION_ARGS)
Definition: timestamp.c:1542
char slotname[NAMEDATALEN]
Definition: walreceiver.h:123
long TimestampDifferenceMilliseconds(TimestampTz start_time, TimestampTz stop_time)
Definition: timestamp.c:1691
char conninfo[MAXCONNINFO]
Definition: walreceiver.h:110