PostgreSQL Source Code  git master
checkpointer.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * checkpointer.c
4  *
5  * The checkpointer is new as of Postgres 9.2. It handles all checkpoints.
6  * Checkpoints are automatically dispatched after a certain amount of time has
7  * elapsed since the last one, and it can be signaled to perform requested
8  * checkpoints as well. (The GUC parameter that mandates a checkpoint every
9  * so many WAL segments is implemented by having backends signal when they
10  * fill WAL segments; the checkpointer itself doesn't watch for the
11  * condition.)
12  *
13  * The checkpointer is started by the postmaster as soon as the startup
14  * subprocess finishes, or as soon as recovery begins if we are doing archive
15  * recovery. It remains alive until the postmaster commands it to terminate.
16  * Normal termination is by SIGUSR2, which instructs the checkpointer to
17  * execute a shutdown checkpoint and then exit(0). (All backends must be
18  * stopped before SIGUSR2 is issued!) Emergency termination is by SIGQUIT;
19  * like any backend, the checkpointer will simply abort and exit on SIGQUIT.
20  *
21  * If the checkpointer exits unexpectedly, the postmaster treats that the same
22  * as a backend crash: shared memory may be corrupted, so remaining backends
23  * should be killed by SIGQUIT and then a recovery cycle started. (Even if
24  * shared memory isn't corrupted, we have lost information about which
25  * files need to be fsync'd for the next checkpoint, and so a system
26  * restart needs to be forced.)
27  *
28  *
29  * Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
30  *
31  *
32  * IDENTIFICATION
33  * src/backend/postmaster/checkpointer.c
34  *
35  *-------------------------------------------------------------------------
36  */
37 #include "postgres.h"
38 
39 #include <sys/time.h>
40 #include <time.h>
41 
42 #include "access/xlog.h"
43 #include "access/xlog_internal.h"
44 #include "libpq/pqsignal.h"
45 #include "miscadmin.h"
46 #include "pgstat.h"
47 #include "postmaster/bgwriter.h"
48 #include "postmaster/interrupt.h"
49 #include "replication/syncrep.h"
50 #include "storage/bufmgr.h"
52 #include "storage/fd.h"
53 #include "storage/ipc.h"
54 #include "storage/lwlock.h"
55 #include "storage/proc.h"
56 #include "storage/procsignal.h"
57 #include "storage/shmem.h"
58 #include "storage/smgr.h"
59 #include "storage/spin.h"
60 #include "utils/guc.h"
61 #include "utils/memutils.h"
62 #include "utils/resowner.h"
63 
64 
65 /*----------
66  * Shared memory area for communication between checkpointer and backends
67  *
68  * The ckpt counters allow backends to watch for completion of a checkpoint
69  * request they send. Here's how it works:
70  * * At start of a checkpoint, checkpointer reads (and clears) the request
71  * flags and increments ckpt_started, while holding ckpt_lck.
72  * * On completion of a checkpoint, checkpointer sets ckpt_done to
73  * equal ckpt_started.
74  * * On failure of a checkpoint, checkpointer increments ckpt_failed
75  * and sets ckpt_done to equal ckpt_started.
76  *
77  * The algorithm for backends is:
78  * 1. Record current values of ckpt_failed and ckpt_started, and
79  * set request flags, while holding ckpt_lck.
80  * 2. Send signal to request checkpoint.
81  * 3. Sleep until ckpt_started changes. Now you know a checkpoint has
82  * begun since you started this algorithm (although *not* that it was
83  * specifically initiated by your signal), and that it is using your flags.
84  * 4. Record new value of ckpt_started.
85  * 5. Sleep until ckpt_done >= saved value of ckpt_started. (Use modulo
86  * arithmetic here in case counters wrap around.) Now you know a
87  * checkpoint has started and completed, but not whether it was
88  * successful.
89  * 6. If ckpt_failed is different from the originally saved value,
90  * assume request failed; otherwise it was definitely successful.
91  *
92  * ckpt_flags holds the OR of the checkpoint request flags sent by all
93  * requesting backends since the last checkpoint start. The flags are
94  * chosen so that OR'ing is the correct way to combine multiple requests.
95  *
96  * num_backend_writes is used to count the number of buffer writes performed
97  * by user backend processes. This counter should be wide enough that it
98  * can't overflow during a single processing cycle. num_backend_fsync
99  * counts the subset of those writes that also had to do their own fsync,
100  * because the checkpointer failed to absorb their request.
101  *
102  * The requests array holds fsync requests sent by backends and not yet
103  * absorbed by the checkpointer.
104  *
105  * Unlike the checkpoint fields, num_backend_writes, num_backend_fsync, and
106  * the requests fields are protected by CheckpointerCommLock.
107  *----------
108  */
109 typedef struct
110 {
111  SyncRequestType type; /* request type */
112  FileTag ftag; /* file identifier */
114 
115 typedef struct
116 {
117  pid_t checkpointer_pid; /* PID (0 if not started) */
118 
119  slock_t ckpt_lck; /* protects all the ckpt_* fields */
120 
121  int ckpt_started; /* advances when checkpoint starts */
122  int ckpt_done; /* advances when checkpoint done */
123  int ckpt_failed; /* advances when checkpoint fails */
124 
125  int ckpt_flags; /* checkpoint flags, as defined in xlog.h */
126 
127  ConditionVariable start_cv; /* signaled when ckpt_started advances */
128  ConditionVariable done_cv; /* signaled when ckpt_done advances */
129 
130  uint32 num_backend_writes; /* counts user backend buffer writes */
131  uint32 num_backend_fsync; /* counts user backend fsync calls */
132 
133  int num_requests; /* current # of requests */
134  int max_requests; /* allocated array size */
137 
139 
140 /* interval for calling AbsorbSyncRequests in CheckpointWriteDelay */
141 #define WRITES_PER_ABSORB 1000
142 
143 /*
144  * GUC parameters
145  */
149 
150 /*
151  * Private state
152  */
153 static bool ckpt_active = false;
154 
155 /* these values are valid when ckpt_active is true: */
158 static double ckpt_cached_elapsed;
159 
162 
163 /* Prototypes for private functions */
164 
165 static void HandleCheckpointerInterrupts(void);
166 static void CheckArchiveTimeout(void);
167 static bool IsCheckpointOnSchedule(double progress);
168 static bool ImmediateCheckpointRequested(void);
169 static bool CompactCheckpointerRequestQueue(void);
170 static void UpdateSharedMemoryConfig(void);
171 
172 /* Signal handlers */
173 static void ReqCheckpointHandler(SIGNAL_ARGS);
174 
175 
176 /*
177  * Main entry point for checkpointer process
178  *
179  * This is invoked from AuxiliaryProcessMain, which has already created the
180  * basic execution environment, but not enabled signals yet.
181  */
182 void
184 {
185  sigjmp_buf local_sigjmp_buf;
186  MemoryContext checkpointer_context;
187 
188  CheckpointerShmem->checkpointer_pid = MyProcPid;
189 
190  /*
191  * Properly accept or ignore signals the postmaster might send us
192  *
193  * Note: we deliberately ignore SIGTERM, because during a standard Unix
194  * system shutdown cycle, init will SIGTERM all processes at once. We
195  * want to wait for the backends to exit, whereupon the postmaster will
196  * tell us it's okay to shut down (via SIGUSR2).
197  */
199  pqsignal(SIGINT, ReqCheckpointHandler); /* request checkpoint */
200  pqsignal(SIGTERM, SIG_IGN); /* ignore SIGTERM */
201  /* SIGQUIT handler was already set up by InitPostmasterChild */
206 
207  /*
208  * Reset some signals that are accepted by postmaster but not here
209  */
211 
212  /*
213  * Initialize so that first time-driven event happens at the correct time.
214  */
216 
217  /*
218  * Create a memory context that we will do all our work in. We do this so
219  * that we can reset the context during error recovery and thereby avoid
220  * possible memory leaks. Formerly this code just ran in
221  * TopMemoryContext, but resetting that would be a really bad idea.
222  */
223  checkpointer_context = AllocSetContextCreate(TopMemoryContext,
224  "Checkpointer",
226  MemoryContextSwitchTo(checkpointer_context);
227 
228  /*
229  * If an exception is encountered, processing resumes here.
230  *
231  * You might wonder why this isn't coded as an infinite loop around a
232  * PG_TRY construct. The reason is that this is the bottom of the
233  * exception stack, and so with PG_TRY there would be no exception handler
234  * in force at all during the CATCH part. By leaving the outermost setjmp
235  * always active, we have at least some chance of recovering from an error
236  * during error recovery. (If we get into an infinite loop thereby, it
237  * will soon be stopped by overflow of elog.c's internal state stack.)
238  *
239  * Note that we use sigsetjmp(..., 1), so that the prevailing signal mask
240  * (to wit, BlockSig) will be restored when longjmp'ing to here. Thus,
241  * signals other than SIGQUIT will be blocked until we complete error
242  * recovery. It might seem that this policy makes the HOLD_INTERRUPTS()
243  * call redundant, but it is not since InterruptPending might be set
244  * already.
245  */
246  if (sigsetjmp(local_sigjmp_buf, 1) != 0)
247  {
248  /* Since not using PG_TRY, must reset error stack by hand */
249  error_context_stack = NULL;
250 
251  /* Prevent interrupts while cleaning up */
252  HOLD_INTERRUPTS();
253 
254  /* Report the error to the server log */
255  EmitErrorReport();
256 
257  /*
258  * These operations are really just a minimal subset of
259  * AbortTransaction(). We don't have very many resources to worry
260  * about in checkpointer, but we do have LWLocks, buffers, and temp
261  * files.
262  */
266  AbortBufferIO();
267  UnlockBuffers();
269  AtEOXact_Buffers(false);
270  AtEOXact_SMgr();
271  AtEOXact_Files(false);
272  AtEOXact_HashTables(false);
273 
274  /* Warn any waiting backends that the checkpoint failed. */
275  if (ckpt_active)
276  {
277  SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
278  CheckpointerShmem->ckpt_failed++;
279  CheckpointerShmem->ckpt_done = CheckpointerShmem->ckpt_started;
280  SpinLockRelease(&CheckpointerShmem->ckpt_lck);
281 
282  ConditionVariableBroadcast(&CheckpointerShmem->done_cv);
283 
284  ckpt_active = false;
285  }
286 
287  /*
288  * Now return to normal top-level context and clear ErrorContext for
289  * next time.
290  */
291  MemoryContextSwitchTo(checkpointer_context);
292  FlushErrorState();
293 
294  /* Flush any leaked data in the top-level context */
295  MemoryContextResetAndDeleteChildren(checkpointer_context);
296 
297  /* Now we can allow interrupts again */
299 
300  /*
301  * Sleep at least 1 second after any error. A write error is likely
302  * to be repeated, and we don't want to be filling the error logs as
303  * fast as we can.
304  */
305  pg_usleep(1000000L);
306 
307  /*
308  * Close all open files after any error. This is helpful on Windows,
309  * where holding deleted files open causes various strange errors.
310  * It's not clear we need it elsewhere, but shouldn't hurt.
311  */
312  smgrcloseall();
313  }
314 
315  /* We can now handle ereport(ERROR) */
316  PG_exception_stack = &local_sigjmp_buf;
317 
318  /*
319  * Unblock signals (they were blocked when the postmaster forked us)
320  */
322 
323  /*
324  * Ensure all shared memory values are set correctly for the config. Doing
325  * this here ensures no race conditions from other concurrent updaters.
326  */
328 
329  /*
330  * Advertise our latch that backends can use to wake us up while we're
331  * sleeping.
332  */
334 
335  /*
336  * Loop forever
337  */
338  for (;;)
339  {
340  bool do_checkpoint = false;
341  int flags = 0;
342  pg_time_t now;
343  int elapsed_secs;
344  int cur_timeout;
345 
346  /* Clear any already-pending wakeups */
348 
349  /*
350  * Process any requests or signals received recently.
351  */
354 
355  /*
356  * Detect a pending checkpoint request by checking whether the flags
357  * word in shared memory is nonzero. We shouldn't need to acquire the
358  * ckpt_lck for this.
359  */
360  if (((volatile CheckpointerShmemStruct *) CheckpointerShmem)->ckpt_flags)
361  {
362  do_checkpoint = true;
364  }
365 
366  /*
367  * Force a checkpoint if too much time has elapsed since the last one.
368  * Note that we count a timed checkpoint in stats only when this
369  * occurs without an external request, but we set the CAUSE_TIME flag
370  * bit even if there is also an external request.
371  */
372  now = (pg_time_t) time(NULL);
373  elapsed_secs = now - last_checkpoint_time;
374  if (elapsed_secs >= CheckPointTimeout)
375  {
376  if (!do_checkpoint)
378  do_checkpoint = true;
379  flags |= CHECKPOINT_CAUSE_TIME;
380  }
381 
382  /*
383  * Do a checkpoint if requested.
384  */
385  if (do_checkpoint)
386  {
387  bool ckpt_performed = false;
388  bool do_restartpoint;
389 
390  /*
391  * Check if we should perform a checkpoint or a restartpoint. As a
392  * side-effect, RecoveryInProgress() initializes TimeLineID if
393  * it's not set yet.
394  */
395  do_restartpoint = RecoveryInProgress();
396 
397  /*
398  * Atomically fetch the request flags to figure out what kind of a
399  * checkpoint we should perform, and increase the started-counter
400  * to acknowledge that we've started a new checkpoint.
401  */
402  SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
403  flags |= CheckpointerShmem->ckpt_flags;
404  CheckpointerShmem->ckpt_flags = 0;
405  CheckpointerShmem->ckpt_started++;
406  SpinLockRelease(&CheckpointerShmem->ckpt_lck);
407 
408  ConditionVariableBroadcast(&CheckpointerShmem->start_cv);
409 
410  /*
411  * The end-of-recovery checkpoint is a real checkpoint that's
412  * performed while we're still in recovery.
413  */
414  if (flags & CHECKPOINT_END_OF_RECOVERY)
415  do_restartpoint = false;
416 
417  /*
418  * We will warn if (a) too soon since last checkpoint (whatever
419  * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
420  * since the last checkpoint start. Note in particular that this
421  * implementation will not generate warnings caused by
422  * CheckPointTimeout < CheckPointWarning.
423  */
424  if (!do_restartpoint &&
425  (flags & CHECKPOINT_CAUSE_XLOG) &&
426  elapsed_secs < CheckPointWarning)
427  ereport(LOG,
428  (errmsg_plural("checkpoints are occurring too frequently (%d second apart)",
429  "checkpoints are occurring too frequently (%d seconds apart)",
430  elapsed_secs,
431  elapsed_secs),
432  errhint("Consider increasing the configuration parameter \"max_wal_size\".")));
433 
434  /*
435  * Initialize checkpointer-private variables used during
436  * checkpoint.
437  */
438  ckpt_active = true;
439  if (do_restartpoint)
441  else
445 
446  /*
447  * Do the checkpoint.
448  */
449  if (!do_restartpoint)
450  {
451  CreateCheckPoint(flags);
452  ckpt_performed = true;
453  }
454  else
455  ckpt_performed = CreateRestartPoint(flags);
456 
457  /*
458  * After any checkpoint, close all smgr files. This is so we
459  * won't hang onto smgr references to deleted files indefinitely.
460  */
461  smgrcloseall();
462 
463  /*
464  * Indicate checkpoint completion to any waiting backends.
465  */
466  SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
467  CheckpointerShmem->ckpt_done = CheckpointerShmem->ckpt_started;
468  SpinLockRelease(&CheckpointerShmem->ckpt_lck);
469 
470  ConditionVariableBroadcast(&CheckpointerShmem->done_cv);
471 
472  if (ckpt_performed)
473  {
474  /*
475  * Note we record the checkpoint start time not end time as
476  * last_checkpoint_time. This is so that time-driven
477  * checkpoints happen at a predictable spacing.
478  */
479  last_checkpoint_time = now;
480  }
481  else
482  {
483  /*
484  * We were not able to perform the restartpoint (checkpoints
485  * throw an ERROR in case of error). Most likely because we
486  * have not received any new checkpoint WAL records since the
487  * last restartpoint. Try again in 15 s.
488  */
489  last_checkpoint_time = now - CheckPointTimeout + 15;
490  }
491 
492  ckpt_active = false;
493  }
494 
495  /* Check for archive_timeout and switch xlog files if necessary. */
497 
498  /*
499  * Send off activity statistics to the stats collector. (The reason
500  * why we re-use bgwriter-related code for this is that the bgwriter
501  * and checkpointer used to be just one process. It's probably not
502  * worth the trouble to split the stats support into two independent
503  * stats message types.)
504  */
506 
507  /*
508  * If any checkpoint flags have been set, redo the loop to handle the
509  * checkpoint without sleeping.
510  */
511  if (((volatile CheckpointerShmemStruct *) CheckpointerShmem)->ckpt_flags)
512  continue;
513 
514  /*
515  * Sleep until we are signaled or it's time for another checkpoint or
516  * xlog file switch.
517  */
518  now = (pg_time_t) time(NULL);
519  elapsed_secs = now - last_checkpoint_time;
520  if (elapsed_secs >= CheckPointTimeout)
521  continue; /* no sleep for us ... */
522  cur_timeout = CheckPointTimeout - elapsed_secs;
524  {
525  elapsed_secs = now - last_xlog_switch_time;
526  if (elapsed_secs >= XLogArchiveTimeout)
527  continue; /* no sleep for us ... */
528  cur_timeout = Min(cur_timeout, XLogArchiveTimeout - elapsed_secs);
529  }
530 
531  (void) WaitLatch(MyLatch,
533  cur_timeout * 1000L /* convert to ms */ ,
535  }
536 }
537 
538 /*
539  * Process any new interrupts.
540  */
541 static void
543 {
546 
548  {
549  ConfigReloadPending = false;
551 
552  /*
553  * Checkpointer is the last process to shut down, so we ask it to hold
554  * the keys for a range of other tasks required most of which have
555  * nothing to do with checkpointing at all.
556  *
557  * For various reasons, some config values can change dynamically so
558  * the primary copy of them is held in shared memory to make sure all
559  * backends see the same value. We make Checkpointer responsible for
560  * updating the shared memory copy if the parameter setting changes
561  * because of SIGHUP.
562  */
564  }
566  {
567  /*
568  * From here on, elog(ERROR) should end with exit(1), not send control
569  * back to the sigsetjmp block above
570  */
571  ExitOnAnyError = true;
572  /* Close down the database */
573  ShutdownXLOG(0, 0);
574  /* Normal exit from the checkpointer is here */
575  proc_exit(0); /* done */
576  }
577 }
578 
579 /*
580  * CheckArchiveTimeout -- check for archive_timeout and switch xlog files
581  *
582  * This will switch to a new WAL file and force an archive file write if
583  * meaningful activity is recorded in the current WAL file. This includes most
584  * writes, including just a single checkpoint record, but excludes WAL records
585  * that were inserted with the XLOG_MARK_UNIMPORTANT flag being set (like
586  * snapshots of running transactions). Such records, depending on
587  * configuration, occur on regular intervals and don't contain important
588  * information. This avoids generating archives with a few unimportant
589  * records.
590  */
591 static void
593 {
594  pg_time_t now;
595  pg_time_t last_time;
596  XLogRecPtr last_switch_lsn;
597 
599  return;
600 
601  now = (pg_time_t) time(NULL);
602 
603  /* First we do a quick check using possibly-stale local state. */
604  if ((int) (now - last_xlog_switch_time) < XLogArchiveTimeout)
605  return;
606 
607  /*
608  * Update local state ... note that last_xlog_switch_time is the last time
609  * a switch was performed *or requested*.
610  */
611  last_time = GetLastSegSwitchData(&last_switch_lsn);
612 
614 
615  /* Now we can do the real checks */
616  if ((int) (now - last_xlog_switch_time) >= XLogArchiveTimeout)
617  {
618  /*
619  * Switch segment only when "important" WAL has been logged since the
620  * last segment switch (last_switch_lsn points to end of segment
621  * switch occurred in).
622  */
623  if (GetLastImportantRecPtr() > last_switch_lsn)
624  {
625  XLogRecPtr switchpoint;
626 
627  /* mark switch as unimportant, avoids triggering checkpoints */
628  switchpoint = RequestXLogSwitch(true);
629 
630  /*
631  * If the returned pointer points exactly to a segment boundary,
632  * assume nothing happened.
633  */
634  if (XLogSegmentOffset(switchpoint, wal_segment_size) != 0)
635  elog(DEBUG1, "write-ahead log switch forced (archive_timeout=%d)",
637  }
638 
639  /*
640  * Update state in any case, so we don't retry constantly when the
641  * system is idle.
642  */
644  }
645 }
646 
647 /*
648  * Returns true if an immediate checkpoint request is pending. (Note that
649  * this does not check the *current* checkpoint's IMMEDIATE flag, but whether
650  * there is one pending behind it.)
651  */
652 static bool
654 {
656 
657  /*
658  * We don't need to acquire the ckpt_lck in this case because we're only
659  * looking at a single flag bit.
660  */
661  if (cps->ckpt_flags & CHECKPOINT_IMMEDIATE)
662  return true;
663  return false;
664 }
665 
666 /*
667  * CheckpointWriteDelay -- control rate of checkpoint
668  *
669  * This function is called after each page write performed by BufferSync().
670  * It is responsible for throttling BufferSync()'s write rate to hit
671  * checkpoint_completion_target.
672  *
673  * The checkpoint request flags should be passed in; currently the only one
674  * examined is CHECKPOINT_IMMEDIATE, which disables delays between writes.
675  *
676  * 'progress' is an estimate of how much of the work has been done, as a
677  * fraction between 0.0 meaning none, and 1.0 meaning all done.
678  */
679 void
680 CheckpointWriteDelay(int flags, double progress)
681 {
682  static int absorb_counter = WRITES_PER_ABSORB;
683 
684  /* Do nothing if checkpoint is being executed by non-checkpointer process */
685  if (!AmCheckpointerProcess())
686  return;
687 
688  /*
689  * Perform the usual duties and take a nap, unless we're behind schedule,
690  * in which case we just try to catch up as quickly as possible.
691  */
692  if (!(flags & CHECKPOINT_IMMEDIATE) &&
695  IsCheckpointOnSchedule(progress))
696  {
698  {
699  ConfigReloadPending = false;
701  /* update shmem copies of config variables */
703  }
704 
706  absorb_counter = WRITES_PER_ABSORB;
707 
709 
710  /*
711  * Report interim activity statistics to the stats collector.
712  */
714 
715  /*
716  * This sleep used to be connected to bgwriter_delay, typically 200ms.
717  * That resulted in more frequent wakeups if not much work to do.
718  * Checkpointer and bgwriter are no longer related so take the Big
719  * Sleep.
720  */
721  pg_usleep(100000L);
722  }
723  else if (--absorb_counter <= 0)
724  {
725  /*
726  * Absorb pending fsync requests after each WRITES_PER_ABSORB write
727  * operations even when we don't sleep, to prevent overflow of the
728  * fsync request queue.
729  */
731  absorb_counter = WRITES_PER_ABSORB;
732  }
733 
734  /* Check for barrier events. */
737 }
738 
739 /*
740  * IsCheckpointOnSchedule -- are we on schedule to finish this checkpoint
741  * (or restartpoint) in time?
742  *
743  * Compares the current progress against the time/segments elapsed since last
744  * checkpoint, and returns true if the progress we've made this far is greater
745  * than the elapsed time/segments.
746  */
747 static bool
749 {
750  XLogRecPtr recptr;
751  struct timeval now;
752  double elapsed_xlogs,
753  elapsed_time;
754 
756 
757  /* Scale progress according to checkpoint_completion_target. */
758  progress *= CheckPointCompletionTarget;
759 
760  /*
761  * Check against the cached value first. Only do the more expensive
762  * calculations once we reach the target previously calculated. Since
763  * neither time or WAL insert pointer moves backwards, a freshly
764  * calculated value can only be greater than or equal to the cached value.
765  */
766  if (progress < ckpt_cached_elapsed)
767  return false;
768 
769  /*
770  * Check progress against WAL segments written and CheckPointSegments.
771  *
772  * We compare the current WAL insert location against the location
773  * computed before calling CreateCheckPoint. The code in XLogInsert that
774  * actually triggers a checkpoint when CheckPointSegments is exceeded
775  * compares against RedoRecPtr, so this is not completely accurate.
776  * However, it's good enough for our purposes, we're only calculating an
777  * estimate anyway.
778  *
779  * During recovery, we compare last replayed WAL record's location with
780  * the location computed before calling CreateRestartPoint. That maintains
781  * the same pacing as we have during checkpoints in normal operation, but
782  * we might exceed max_wal_size by a fair amount. That's because there can
783  * be a large gap between a checkpoint's redo-pointer and the checkpoint
784  * record itself, and we only start the restartpoint after we've seen the
785  * checkpoint record. (The gap is typically up to CheckPointSegments *
786  * checkpoint_completion_target where checkpoint_completion_target is the
787  * value that was in effect when the WAL was generated).
788  */
789  if (RecoveryInProgress())
790  recptr = GetXLogReplayRecPtr(NULL);
791  else
792  recptr = GetInsertRecPtr();
793  elapsed_xlogs = (((double) (recptr - ckpt_start_recptr)) /
795 
796  if (progress < elapsed_xlogs)
797  {
798  ckpt_cached_elapsed = elapsed_xlogs;
799  return false;
800  }
801 
802  /*
803  * Check progress against time elapsed and checkpoint_timeout.
804  */
805  gettimeofday(&now, NULL);
806  elapsed_time = ((double) ((pg_time_t) now.tv_sec - ckpt_start_time) +
807  now.tv_usec / 1000000.0) / CheckPointTimeout;
808 
809  if (progress < elapsed_time)
810  {
812  return false;
813  }
814 
815  /* It looks like we're on schedule. */
816  return true;
817 }
818 
819 
820 /* --------------------------------
821  * signal handler routines
822  * --------------------------------
823  */
824 
825 /* SIGINT: set flag to run a normal checkpoint right away */
826 static void
828 {
829  int save_errno = errno;
830 
831  /*
832  * The signaling process should have set ckpt_flags nonzero, so all we
833  * need do is ensure that our main loop gets kicked out of any wait.
834  */
835  SetLatch(MyLatch);
836 
837  errno = save_errno;
838 }
839 
840 
841 /* --------------------------------
842  * communication with backends
843  * --------------------------------
844  */
845 
846 /*
847  * CheckpointerShmemSize
848  * Compute space needed for checkpointer-related shared memory
849  */
850 Size
852 {
853  Size size;
854 
855  /*
856  * Currently, the size of the requests[] array is arbitrarily set equal to
857  * NBuffers. This may prove too large or small ...
858  */
859  size = offsetof(CheckpointerShmemStruct, requests);
860  size = add_size(size, mul_size(NBuffers, sizeof(CheckpointerRequest)));
861 
862  return size;
863 }
864 
865 /*
866  * CheckpointerShmemInit
867  * Allocate and initialize checkpointer-related shared memory
868  */
869 void
871 {
872  Size size = CheckpointerShmemSize();
873  bool found;
874 
875  CheckpointerShmem = (CheckpointerShmemStruct *)
876  ShmemInitStruct("Checkpointer Data",
877  size,
878  &found);
879 
880  if (!found)
881  {
882  /*
883  * First time through, so initialize. Note that we zero the whole
884  * requests array; this is so that CompactCheckpointerRequestQueue can
885  * assume that any pad bytes in the request structs are zeroes.
886  */
887  MemSet(CheckpointerShmem, 0, size);
888  SpinLockInit(&CheckpointerShmem->ckpt_lck);
889  CheckpointerShmem->max_requests = NBuffers;
890  ConditionVariableInit(&CheckpointerShmem->start_cv);
891  ConditionVariableInit(&CheckpointerShmem->done_cv);
892  }
893 }
894 
895 /*
896  * RequestCheckpoint
897  * Called in backend processes to request a checkpoint
898  *
899  * flags is a bitwise OR of the following:
900  * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
901  * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
902  * CHECKPOINT_IMMEDIATE: finish the checkpoint ASAP,
903  * ignoring checkpoint_completion_target parameter.
904  * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
905  * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
906  * CHECKPOINT_END_OF_RECOVERY).
907  * CHECKPOINT_WAIT: wait for completion before returning (otherwise,
908  * just signal checkpointer to do it, and return).
909  * CHECKPOINT_CAUSE_XLOG: checkpoint is requested due to xlog filling.
910  * (This affects logging, and in particular enables CheckPointWarning.)
911  */
912 void
914 {
915  int ntries;
916  int old_failed,
917  old_started;
918 
919  /*
920  * If in a standalone backend, just do it ourselves.
921  */
923  {
924  /*
925  * There's no point in doing slow checkpoints in a standalone backend,
926  * because there's no other backends the checkpoint could disrupt.
927  */
929 
930  /*
931  * After any checkpoint, close all smgr files. This is so we won't
932  * hang onto smgr references to deleted files indefinitely.
933  */
934  smgrcloseall();
935 
936  return;
937  }
938 
939  /*
940  * Atomically set the request flags, and take a snapshot of the counters.
941  * When we see ckpt_started > old_started, we know the flags we set here
942  * have been seen by checkpointer.
943  *
944  * Note that we OR the flags with any existing flags, to avoid overriding
945  * a "stronger" request by another backend. The flag senses must be
946  * chosen to make this work!
947  */
948  SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
949 
950  old_failed = CheckpointerShmem->ckpt_failed;
951  old_started = CheckpointerShmem->ckpt_started;
952  CheckpointerShmem->ckpt_flags |= (flags | CHECKPOINT_REQUESTED);
953 
954  SpinLockRelease(&CheckpointerShmem->ckpt_lck);
955 
956  /*
957  * Send signal to request checkpoint. It's possible that the checkpointer
958  * hasn't started yet, or is in process of restarting, so we will retry a
959  * few times if needed. (Actually, more than a few times, since on slow
960  * or overloaded buildfarm machines, it's been observed that the
961  * checkpointer can take several seconds to start.) However, if not told
962  * to wait for the checkpoint to occur, we consider failure to send the
963  * signal to be nonfatal and merely LOG it. The checkpointer should see
964  * the request when it does start, with or without getting a signal.
965  */
966 #define MAX_SIGNAL_TRIES 600 /* max wait 60.0 sec */
967  for (ntries = 0;; ntries++)
968  {
969  if (CheckpointerShmem->checkpointer_pid == 0)
970  {
971  if (ntries >= MAX_SIGNAL_TRIES || !(flags & CHECKPOINT_WAIT))
972  {
973  elog((flags & CHECKPOINT_WAIT) ? ERROR : LOG,
974  "could not signal for checkpoint: checkpointer is not running");
975  break;
976  }
977  }
978  else if (kill(CheckpointerShmem->checkpointer_pid, SIGINT) != 0)
979  {
980  if (ntries >= MAX_SIGNAL_TRIES || !(flags & CHECKPOINT_WAIT))
981  {
982  elog((flags & CHECKPOINT_WAIT) ? ERROR : LOG,
983  "could not signal for checkpoint: %m");
984  break;
985  }
986  }
987  else
988  break; /* signal sent successfully */
989 
991  pg_usleep(100000L); /* wait 0.1 sec, then retry */
992  }
993 
994  /*
995  * If requested, wait for completion. We detect completion according to
996  * the algorithm given above.
997  */
998  if (flags & CHECKPOINT_WAIT)
999  {
1000  int new_started,
1001  new_failed;
1002 
1003  /* Wait for a new checkpoint to start. */
1004  ConditionVariablePrepareToSleep(&CheckpointerShmem->start_cv);
1005  for (;;)
1006  {
1007  SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
1008  new_started = CheckpointerShmem->ckpt_started;
1009  SpinLockRelease(&CheckpointerShmem->ckpt_lck);
1010 
1011  if (new_started != old_started)
1012  break;
1013 
1014  ConditionVariableSleep(&CheckpointerShmem->start_cv,
1016  }
1018 
1019  /*
1020  * We are waiting for ckpt_done >= new_started, in a modulo sense.
1021  */
1022  ConditionVariablePrepareToSleep(&CheckpointerShmem->done_cv);
1023  for (;;)
1024  {
1025  int new_done;
1026 
1027  SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
1028  new_done = CheckpointerShmem->ckpt_done;
1029  new_failed = CheckpointerShmem->ckpt_failed;
1030  SpinLockRelease(&CheckpointerShmem->ckpt_lck);
1031 
1032  if (new_done - new_started >= 0)
1033  break;
1034 
1035  ConditionVariableSleep(&CheckpointerShmem->done_cv,
1037  }
1039 
1040  if (new_failed != old_failed)
1041  ereport(ERROR,
1042  (errmsg("checkpoint request failed"),
1043  errhint("Consult recent messages in the server log for details.")));
1044  }
1045 }
1046 
1047 /*
1048  * ForwardSyncRequest
1049  * Forward a file-fsync request from a backend to the checkpointer
1050  *
1051  * Whenever a backend is compelled to write directly to a relation
1052  * (which should be seldom, if the background writer is getting its job done),
1053  * the backend calls this routine to pass over knowledge that the relation
1054  * is dirty and must be fsync'd before next checkpoint. We also use this
1055  * opportunity to count such writes for statistical purposes.
1056  *
1057  * To avoid holding the lock for longer than necessary, we normally write
1058  * to the requests[] queue without checking for duplicates. The checkpointer
1059  * will have to eliminate dups internally anyway. However, if we discover
1060  * that the queue is full, we make a pass over the entire queue to compact
1061  * it. This is somewhat expensive, but the alternative is for the backend
1062  * to perform its own fsync, which is far more expensive in practice. It
1063  * is theoretically possible a backend fsync might still be necessary, if
1064  * the queue is full and contains no duplicate entries. In that case, we
1065  * let the backend know by returning false.
1066  */
1067 bool
1069 {
1070  CheckpointerRequest *request;
1071  bool too_full;
1072 
1073  if (!IsUnderPostmaster)
1074  return false; /* probably shouldn't even get here */
1075 
1076  if (AmCheckpointerProcess())
1077  elog(ERROR, "ForwardSyncRequest must not be called in checkpointer");
1078 
1079  LWLockAcquire(CheckpointerCommLock, LW_EXCLUSIVE);
1080 
1081  /* Count all backend writes regardless of if they fit in the queue */
1083  CheckpointerShmem->num_backend_writes++;
1084 
1085  /*
1086  * If the checkpointer isn't running or the request queue is full, the
1087  * backend will have to perform its own fsync request. But before forcing
1088  * that to happen, we can try to compact the request queue.
1089  */
1090  if (CheckpointerShmem->checkpointer_pid == 0 ||
1091  (CheckpointerShmem->num_requests >= CheckpointerShmem->max_requests &&
1093  {
1094  /*
1095  * Count the subset of writes where backends have to do their own
1096  * fsync
1097  */
1099  CheckpointerShmem->num_backend_fsync++;
1100  LWLockRelease(CheckpointerCommLock);
1101  return false;
1102  }
1103 
1104  /* OK, insert request */
1105  request = &CheckpointerShmem->requests[CheckpointerShmem->num_requests++];
1106  request->ftag = *ftag;
1107  request->type = type;
1108 
1109  /* If queue is more than half full, nudge the checkpointer to empty it */
1110  too_full = (CheckpointerShmem->num_requests >=
1111  CheckpointerShmem->max_requests / 2);
1112 
1113  LWLockRelease(CheckpointerCommLock);
1114 
1115  /* ... but not till after we release the lock */
1116  if (too_full && ProcGlobal->checkpointerLatch)
1118 
1119  return true;
1120 }
1121 
1122 /*
1123  * CompactCheckpointerRequestQueue
1124  * Remove duplicates from the request queue to avoid backend fsyncs.
1125  * Returns "true" if any entries were removed.
1126  *
1127  * Although a full fsync request queue is not common, it can lead to severe
1128  * performance problems when it does happen. So far, this situation has
1129  * only been observed to occur when the system is under heavy write load,
1130  * and especially during the "sync" phase of a checkpoint. Without this
1131  * logic, each backend begins doing an fsync for every block written, which
1132  * gets very expensive and can slow down the whole system.
1133  *
1134  * Trying to do this every time the queue is full could lose if there
1135  * aren't any removable entries. But that should be vanishingly rare in
1136  * practice: there's one queue entry per shared buffer.
1137  */
1138 static bool
1140 {
1141  struct CheckpointerSlotMapping
1142  {
1143  CheckpointerRequest request;
1144  int slot;
1145  };
1146 
1147  int n,
1148  preserve_count;
1149  int num_skipped = 0;
1150  HASHCTL ctl;
1151  HTAB *htab;
1152  bool *skip_slot;
1153 
1154  /* must hold CheckpointerCommLock in exclusive mode */
1155  Assert(LWLockHeldByMe(CheckpointerCommLock));
1156 
1157  /* Initialize skip_slot array */
1158  skip_slot = palloc0(sizeof(bool) * CheckpointerShmem->num_requests);
1159 
1160  /* Initialize temporary hash table */
1161  MemSet(&ctl, 0, sizeof(ctl));
1162  ctl.keysize = sizeof(CheckpointerRequest);
1163  ctl.entrysize = sizeof(struct CheckpointerSlotMapping);
1164  ctl.hcxt = CurrentMemoryContext;
1165 
1166  htab = hash_create("CompactCheckpointerRequestQueue",
1167  CheckpointerShmem->num_requests,
1168  &ctl,
1170 
1171  /*
1172  * The basic idea here is that a request can be skipped if it's followed
1173  * by a later, identical request. It might seem more sensible to work
1174  * backwards from the end of the queue and check whether a request is
1175  * *preceded* by an earlier, identical request, in the hopes of doing less
1176  * copying. But that might change the semantics, if there's an
1177  * intervening SYNC_FORGET_REQUEST or SYNC_FILTER_REQUEST, so we do it
1178  * this way. It would be possible to be even smarter if we made the code
1179  * below understand the specific semantics of such requests (it could blow
1180  * away preceding entries that would end up being canceled anyhow), but
1181  * it's not clear that the extra complexity would buy us anything.
1182  */
1183  for (n = 0; n < CheckpointerShmem->num_requests; n++)
1184  {
1185  CheckpointerRequest *request;
1186  struct CheckpointerSlotMapping *slotmap;
1187  bool found;
1188 
1189  /*
1190  * We use the request struct directly as a hashtable key. This
1191  * assumes that any padding bytes in the structs are consistently the
1192  * same, which should be okay because we zeroed them in
1193  * CheckpointerShmemInit. Note also that RelFileNode had better
1194  * contain no pad bytes.
1195  */
1196  request = &CheckpointerShmem->requests[n];
1197  slotmap = hash_search(htab, request, HASH_ENTER, &found);
1198  if (found)
1199  {
1200  /* Duplicate, so mark the previous occurrence as skippable */
1201  skip_slot[slotmap->slot] = true;
1202  num_skipped++;
1203  }
1204  /* Remember slot containing latest occurrence of this request value */
1205  slotmap->slot = n;
1206  }
1207 
1208  /* Done with the hash table. */
1209  hash_destroy(htab);
1210 
1211  /* If no duplicates, we're out of luck. */
1212  if (!num_skipped)
1213  {
1214  pfree(skip_slot);
1215  return false;
1216  }
1217 
1218  /* We found some duplicates; remove them. */
1219  preserve_count = 0;
1220  for (n = 0; n < CheckpointerShmem->num_requests; n++)
1221  {
1222  if (skip_slot[n])
1223  continue;
1224  CheckpointerShmem->requests[preserve_count++] = CheckpointerShmem->requests[n];
1225  }
1226  ereport(DEBUG1,
1227  (errmsg("compacted fsync request queue from %d entries to %d entries",
1228  CheckpointerShmem->num_requests, preserve_count)));
1229  CheckpointerShmem->num_requests = preserve_count;
1230 
1231  /* Cleanup. */
1232  pfree(skip_slot);
1233  return true;
1234 }
1235 
1236 /*
1237  * AbsorbSyncRequests
1238  * Retrieve queued sync requests and pass them to sync mechanism.
1239  *
1240  * This is exported because it must be called during CreateCheckPoint;
1241  * we have to be sure we have accepted all pending requests just before
1242  * we start fsync'ing. Since CreateCheckPoint sometimes runs in
1243  * non-checkpointer processes, do nothing if not checkpointer.
1244  */
1245 void
1247 {
1248  CheckpointerRequest *requests = NULL;
1249  CheckpointerRequest *request;
1250  int n;
1251 
1252  if (!AmCheckpointerProcess())
1253  return;
1254 
1255  LWLockAcquire(CheckpointerCommLock, LW_EXCLUSIVE);
1256 
1257  /* Transfer stats counts into pending pgstats message */
1259  BgWriterStats.m_buf_fsync_backend += CheckpointerShmem->num_backend_fsync;
1260 
1261  CheckpointerShmem->num_backend_writes = 0;
1262  CheckpointerShmem->num_backend_fsync = 0;
1263 
1264  /*
1265  * We try to avoid holding the lock for a long time by copying the request
1266  * array, and processing the requests after releasing the lock.
1267  *
1268  * Once we have cleared the requests from shared memory, we have to PANIC
1269  * if we then fail to absorb them (eg, because our hashtable runs out of
1270  * memory). This is because the system cannot run safely if we are unable
1271  * to fsync what we have been told to fsync. Fortunately, the hashtable
1272  * is so small that the problem is quite unlikely to arise in practice.
1273  */
1274  n = CheckpointerShmem->num_requests;
1275  if (n > 0)
1276  {
1277  requests = (CheckpointerRequest *) palloc(n * sizeof(CheckpointerRequest));
1278  memcpy(requests, CheckpointerShmem->requests, n * sizeof(CheckpointerRequest));
1279  }
1280 
1282 
1283  CheckpointerShmem->num_requests = 0;
1284 
1285  LWLockRelease(CheckpointerCommLock);
1286 
1287  for (request = requests; n > 0; request++, n--)
1288  RememberSyncRequest(&request->ftag, request->type);
1289 
1290  END_CRIT_SECTION();
1291 
1292  if (requests)
1293  pfree(requests);
1294 }
1295 
1296 /*
1297  * Update any shared memory configurations based on config parameters
1298  */
1299 static void
1301 {
1302  /* update global shmem state for sync rep */
1304 
1305  /*
1306  * If full_page_writes has been changed by SIGHUP, we update it in shared
1307  * memory and write an XLOG_FPW_CHANGE record.
1308  */
1310 
1311  elog(DEBUG2, "checkpointer updated shared memory configuration values");
1312 }
1313 
1314 /*
1315  * FirstCallSinceLastCheckpoint allows a process to take an action once
1316  * per checkpoint cycle by asynchronously checking for checkpoint completion.
1317  */
1318 bool
1320 {
1321  static int ckpt_done = 0;
1322  int new_done;
1323  bool FirstCall = false;
1324 
1325  SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
1326  new_done = CheckpointerShmem->ckpt_done;
1327  SpinLockRelease(&CheckpointerShmem->ckpt_lck);
1328 
1329  if (new_done != ckpt_done)
1330  FirstCall = true;
1331 
1332  ckpt_done = new_done;
1333 
1334  return FirstCall;
1335 }
XLogRecPtr GetLastImportantRecPtr(void)
Definition: xlog.c:8444
int slock_t
Definition: s_lock.h:934
PgStat_Counter m_buf_fsync_backend
Definition: pgstat.h:433
void SyncRepUpdateSyncStandbysDefined(void)
Definition: syncrep.c:948
void RememberSyncRequest(const FileTag *ftag, SyncRequestType type)
Definition: sync.c:437
bool IsPostmasterEnvironment
Definition: globals.c:108
void CheckpointWriteDelay(int flags, double progress)
Definition: checkpointer.c:680
SyncRequestType type
Definition: checkpointer.c:111
void hash_destroy(HTAB *hashp)
Definition: dynahash.c:827
static bool IsCheckpointOnSchedule(double progress)
Definition: checkpointer.c:748
int gettimeofday(struct timeval *tp, struct timezone *tzp)
Definition: gettimeofday.c:104
#define AllocSetContextCreate
Definition: memutils.h:170
XLogRecPtr RequestXLogSwitch(bool mark_unimportant)
Definition: xlog.c:9683
#define DEBUG1
Definition: elog.h:25
XLogRecPtr GetInsertRecPtr(void)
Definition: xlog.c:8410
int MyProcPid
Definition: globals.c:40
int errhint(const char *fmt,...)
Definition: elog.c:1071
int64 pg_time_t
Definition: pgtime.h:23
pg_time_t GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
Definition: xlog.c:8473
#define HASH_CONTEXT
Definition: hsearch.h:91
#define HASH_ELEM
Definition: hsearch.h:85
#define WL_TIMEOUT
Definition: latch.h:127
void ProcessConfigFile(GucContext context)
int wal_segment_size
Definition: xlog.c:117
static void ReqCheckpointHandler(SIGNAL_ARGS)
Definition: checkpointer.c:827
int XLogArchiveTimeout
Definition: xlog.c:94
MemoryContext hcxt
Definition: hsearch.h:77
#define SIGUSR1
Definition: win32_port.h:165
bool LWLockHeldByMe(LWLock *l)
Definition: lwlock.c:1928
int errmsg_plural(const char *fmt_singular, const char *fmt_plural, unsigned long n,...)
Definition: elog.c:934
void CreateCheckPoint(int flags)
Definition: xlog.c:8706
#define SIGCHLD
Definition: win32_port.h:163
PGPROC * MyProc
Definition: proc.c:67
PgStat_Counter m_timed_checkpoints
Definition: pgstat.h:427
void AtEOXact_Buffers(bool isCommit)
Definition: bufmgr.c:2478
void SignalHandlerForConfigReload(SIGNAL_ARGS)
Definition: interrupt.c:56
void ReleaseAuxProcessResources(bool isCommit)
Definition: resowner.c:879
#define SpinLockInit(lock)
Definition: spin.h:60
#define Min(x, y)
Definition: c.h:927
void ProcessProcSignalBarrier(void)
Definition: procsignal.c:452
static void HandleCheckpointerInterrupts(void)
Definition: checkpointer.c:542
#define END_CRIT_SECTION()
Definition: miscadmin.h:134
bool CreateRestartPoint(int flags)
Definition: xlog.c:9247
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
void CheckpointerShmemInit(void)
Definition: checkpointer.c:870
PgStat_MsgBgWriter BgWriterStats
Definition: pgstat.c:142
Size entrysize
Definition: hsearch.h:72
int CheckPointWarning
Definition: checkpointer.c:147
CheckpointerRequest requests[FLEXIBLE_ARRAY_MEMBER]
Definition: checkpointer.c:135
#define FLEXIBLE_ARRAY_MEMBER
Definition: c.h:283
#define START_CRIT_SECTION()
Definition: miscadmin.h:132
void ConditionVariableBroadcast(ConditionVariable *cv)
void proc_exit(int code)
Definition: ipc.c:104
PROC_HDR * ProcGlobal
Definition: proc.c:79
#define MemSet(start, val, len)
Definition: c.h:949
#define kill(pid, sig)
Definition: win32_port.h:426
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:919
#define SIGPIPE
Definition: win32_port.h:158
#define SIGUSR2
Definition: win32_port.h:166
#define LOG
Definition: elog.h:26
bool RecoveryInProgress(void)
Definition: xlog.c:8074
void SetLatch(Latch *latch)
Definition: latch.c:505
void FlushErrorState(void)
Definition: elog.c:1568
void ResetLatch(Latch *latch)
Definition: latch.c:588
#define PG_SETMASK(mask)
Definition: pqsignal.h:19
void AtEOXact_Files(bool isCommit)
Definition: fd.c:2904
Latch procLatch
Definition: proc.h:121
int WaitLatch(Latch *latch, int wakeEvents, long timeout, uint32 wait_event_info)
Definition: latch.c:390
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1812
void smgrcloseall(void)
Definition: smgr.c:287
#define RESUME_INTERRUPTS()
Definition: miscadmin.h:118
ErrorContextCallback * error_context_stack
Definition: elog.c:92
void ConditionVariablePrepareToSleep(ConditionVariable *cv)
#define CHECKPOINT_CAUSE_XLOG
Definition: xlog.h:232
PgStat_Counter m_requested_checkpoints
Definition: pgstat.h:428
#define SpinLockAcquire(lock)
Definition: spin.h:62
static bool ImmediateCheckpointRequested(void)
Definition: checkpointer.c:653
void ConditionVariableInit(ConditionVariable *cv)
void pg_usleep(long microsec)
Definition: signal.c:53
Definition: dynahash.c:218
void AtEOXact_SMgr(void)
Definition: smgr.c:661
void AbsorbSyncRequests(void)
void pfree(void *pointer)
Definition: mcxt.c:1057
void ConditionVariableCancelSleep(void)
#define AmBackgroundWriterProcess()
Definition: miscadmin.h:432
#define ERROR
Definition: elog.h:43
#define AmCheckpointerProcess()
Definition: miscadmin.h:433
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:392
static bool CompactCheckpointerRequestQueue(void)
bool ForwardSyncRequest(const FileTag *ftag, SyncRequestType type)
XLogRecPtr GetXLogReplayRecPtr(TimeLineID *replayTLI)
Definition: xlog.c:11480
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:192
#define DEBUG2
Definition: elog.h:24
bool IsUnderPostmaster
Definition: globals.c:109
#define CHECKPOINT_REQUESTED
Definition: xlog.h:230
int CheckPointTimeout
Definition: checkpointer.c:146
#define CHECKPOINT_END_OF_RECOVERY
Definition: xlog.h:222
#define SIGHUP
Definition: win32_port.h:153
unsigned int uint32
Definition: c.h:374
sigset_t UnBlockSig
Definition: pqsignal.c:22
static void pgstat_report_wait_end(void)
Definition: pgstat.h:1386
MemoryContext CurrentMemoryContext
Definition: mcxt.c:38
MemoryContext TopMemoryContext
Definition: mcxt.c:44
volatile sig_atomic_t ShutdownRequestPending
Definition: interrupt.c:27
Definition: guc.h:72
ConditionVariable done_cv
Definition: checkpointer.c:128
int CheckPointSegments
Definition: xlog.c:130
void SignalHandlerForShutdownRequest(SIGNAL_ARGS)
Definition: interrupt.c:104
static CheckpointerShmemStruct * CheckpointerShmem
Definition: checkpointer.c:138
#define SIG_IGN
Definition: win32_port.h:150
void ShutdownXLOG(int code, Datum arg)
Definition: xlog.c:8490
void UnlockBuffers(void)
Definition: bufmgr.c:3729
int progress
Definition: pgbench.c:235
#define MemoryContextResetAndDeleteChildren(ctx)
Definition: memutils.h:67
#define SpinLockRelease(lock)
Definition: spin.h:64
#define HASH_BLOBS
Definition: hsearch.h:86
bool ExitOnAnyError
Definition: globals.c:113
static void UpdateSharedMemoryConfig(void)
Size mul_size(Size s1, Size s2)
Definition: shmem.c:515
void UpdateFullPageWrites(void)
Definition: xlog.c:9789
void * palloc0(Size size)
Definition: mcxt.c:981
bool FirstCallSinceLastCheckpoint(void)
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:326
Size add_size(Size s1, Size s2)
Definition: shmem.c:498
#define WRITES_PER_ABSORB
Definition: checkpointer.c:141
#define XLogSegmentOffset(xlogptr, wal_segsz_bytes)
Size keysize
Definition: hsearch.h:71
Size CheckpointerShmemSize(void)
Definition: checkpointer.c:851
void EmitErrorReport(void)
Definition: elog.c:1423
static pg_time_t last_xlog_switch_time
Definition: checkpointer.c:161
void pgstat_send_bgwriter(void)
Definition: pgstat.c:4397
void CheckpointerMain(void)
Definition: checkpointer.c:183
static bool ckpt_active
Definition: checkpointer.c:153
#define ereport(elevel,...)
Definition: elog.h:144
static pg_time_t ckpt_start_time
Definition: checkpointer.c:156
pqsigfunc pqsignal(int signum, pqsigfunc handler)
Definition: signal.c:170
static XLogRecPtr ckpt_start_recptr
Definition: checkpointer.c:157
#define CHECKPOINT_WAIT
Definition: xlog.h:229
Latch * checkpointerLatch
Definition: proc.h:340
#define Max(x, y)
Definition: c.h:921
#define SIG_DFL
Definition: win32_port.h:148
void ConditionVariableSleep(ConditionVariable *cv, uint32 wait_event_info)
#define SIGNAL_ARGS
Definition: c.h:1273
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:745
volatile sig_atomic_t ProcSignalBarrierPending
Definition: globals.c:35
PgStat_Counter m_buf_written_backend
Definition: pgstat.h:432
size_t Size
Definition: c.h:473
#define SIGALRM
Definition: win32_port.h:159
#define MAX_SIGNAL_TRIES
SyncRequestType
Definition: sync.h:23
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1208
void AbortBufferIO(void)
Definition: bufmgr.c:4207
static double elapsed_time(instr_time *starttime)
Definition: explain.c:1016
sigjmp_buf * PG_exception_stack
Definition: elog.c:94
void * palloc(Size size)
Definition: mcxt.c:950
int errmsg(const char *fmt,...)
Definition: elog.c:824
static double ckpt_cached_elapsed
Definition: checkpointer.c:158
static void CheckArchiveTimeout(void)
Definition: checkpointer.c:592
#define HOLD_INTERRUPTS()
Definition: miscadmin.h:116
#define elog(elevel,...)
Definition: elog.h:214
volatile sig_atomic_t ConfigReloadPending
Definition: interrupt.c:26
#define CHECKPOINT_CAUSE_TIME
Definition: xlog.h:233
#define CHECKPOINT_IMMEDIATE
Definition: xlog.h:224
int NBuffers
Definition: globals.c:132
struct Latch * MyLatch
Definition: globals.c:54
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:99
void LWLockReleaseAll(void)
Definition: lwlock.c:1911
void procsignal_sigusr1_handler(SIGNAL_ARGS)
Definition: procsignal.c:551
void AtEOXact_HashTables(bool isCommit)
Definition: dynahash.c:1848
static pg_time_t last_checkpoint_time
Definition: checkpointer.c:160
#define WL_LATCH_SET
Definition: latch.h:124
Datum now(PG_FUNCTION_ARGS)
Definition: timestamp.c:1538
Definition: sync.h:45
#define offsetof(type, field)
Definition: c.h:668
#define WL_EXIT_ON_PM_DEATH
Definition: latch.h:129
void RequestCheckpoint(int flags)
Definition: checkpointer.c:913
double CheckPointCompletionTarget
Definition: checkpointer.c:148
ConditionVariable start_cv
Definition: checkpointer.c:127