PostgreSQL Source Code: src/backend/postmaster/checkpointer.c Source File

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 normal termination sequence is that checkpointer is instructed to
14 * execute the shutdown checkpoint by SIGINT. After that checkpointer waits
15 * to be terminated via SIGUSR2, which instructs the checkpointer to exit(0).
16 * All backends must be stopped before SIGINT or SIGUSR2 is issued!
17 *
18 * Emergency termination is by SIGQUIT; like any backend, the checkpointer
19 * 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) 1996-2025, 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 "access/xlogrecovery.h"
45#include "catalog/pg_authid.h"
46#include "commands/defrem.h"
47#include "libpq/pqsignal.h"
48#include "miscadmin.h"
49#include "pgstat.h"
50#include "postmaster/auxprocess.h"
51#include "postmaster/bgwriter.h"
52#include "postmaster/interrupt.h"
53#include "replication/syncrep.h"
54#include "storage/aio_subsys.h"
55#include "storage/bufmgr.h"
56#include "storage/condition_variable.h"
57#include "storage/fd.h"
58#include "storage/ipc.h"
59#include "storage/lwlock.h"
60#include "storage/pmsignal.h"
61#include "storage/proc.h"
62#include "storage/procsignal.h"
63#include "storage/shmem.h"
64#include "storage/smgr.h"
65#include "storage/spin.h"
66#include "utils/acl.h"
67#include "utils/guc.h"
68#include "utils/memutils.h"
69#include "utils/resowner.h"
70
71
72/*----------
73 * Shared memory area for communication between checkpointer and backends
74 *
75 * The ckpt counters allow backends to watch for completion of a checkpoint
76 * request they send. Here's how it works:
77 * * At start of a checkpoint, checkpointer reads (and clears) the request
78 * flags and increments ckpt_started, while holding ckpt_lck.
79 * * On completion of a checkpoint, checkpointer sets ckpt_done to
80 * equal ckpt_started.
81 * * On failure of a checkpoint, checkpointer increments ckpt_failed
82 * and sets ckpt_done to equal ckpt_started.
83 *
84 * The algorithm for backends is:
85 * 1. Record current values of ckpt_failed and ckpt_started, and
86 * set request flags, while holding ckpt_lck.
87 * 2. Send signal to request checkpoint.
88 * 3. Sleep until ckpt_started changes. Now you know a checkpoint has
89 * begun since you started this algorithm (although *not* that it was
90 * specifically initiated by your signal), and that it is using your flags.
91 * 4. Record new value of ckpt_started.
92 * 5. Sleep until ckpt_done >= saved value of ckpt_started. (Use modulo
93 * arithmetic here in case counters wrap around.) Now you know a
94 * checkpoint has started and completed, but not whether it was
95 * successful.
96 * 6. If ckpt_failed is different from the originally saved value,
97 * assume request failed; otherwise it was definitely successful.
98 *
99 * ckpt_flags holds the OR of the checkpoint request flags sent by all
100 * requesting backends since the last checkpoint start. The flags are
101 * chosen so that OR'ing is the correct way to combine multiple requests.
102 *
103 * The requests array holds fsync requests sent by backends and not yet
104 * absorbed by the checkpointer.
105 *
106 * Unlike the checkpoint fields, requests related fields are protected by
107 * CheckpointerCommLock.
108 *----------
109 */
110 typedef struct
111{
112 SyncRequestType type; /* request type */
113 FileTag ftag; /* file identifier */
114} CheckpointerRequest;
115
116 typedef struct
117{
118 pid_t checkpointer_pid; /* PID (0 if not started) */
119
120 slock_t ckpt_lck; /* protects all the ckpt_* fields */
121
122 int ckpt_started; /* advances when checkpoint starts */
123 int ckpt_done; /* advances when checkpoint done */
124 int ckpt_failed; /* advances when checkpoint fails */
125
126 int ckpt_flags; /* checkpoint flags, as defined in xlog.h */
127
128 ConditionVariable start_cv; /* signaled when ckpt_started advances */
129 ConditionVariable done_cv; /* signaled when ckpt_done advances */
130
131 int num_requests; /* current # of requests */
132 int max_requests; /* allocated array size */
133
134 int head; /* Index of the first request in the ring
135 * buffer */
136 int tail; /* Index of the last request in the ring
137 * buffer */
138
139 /* The ring buffer of pending checkpointer requests */
140 CheckpointerRequest requests[FLEXIBLE_ARRAY_MEMBER];
141} CheckpointerShmemStruct;
142
143 static CheckpointerShmemStruct *CheckpointerShmem;
144
145/* interval for calling AbsorbSyncRequests in CheckpointWriteDelay */
146 #define WRITES_PER_ABSORB 1000
147
148/* Maximum number of checkpointer requests to process in one batch */
149 #define CKPT_REQ_BATCH_SIZE 10000
150
151/* Max number of requests the checkpointer request queue can hold */
152 #define MAX_CHECKPOINT_REQUESTS 10000000
153
154/*
155 * GUC parameters
156 */
157 int CheckPointTimeout = 300;
158 int CheckPointWarning = 30;
159 double CheckPointCompletionTarget = 0.9;
160
161/*
162 * Private state
163 */
164 static bool ckpt_active = false;
165 static volatile sig_atomic_t ShutdownXLOGPending = false;
166
167/* these values are valid when ckpt_active is true: */
168 static pg_time_t ckpt_start_time;
169 static XLogRecPtr ckpt_start_recptr;
170 static double ckpt_cached_elapsed;
171
172 static pg_time_t last_checkpoint_time;
173 static pg_time_t last_xlog_switch_time;
174
175/* Prototypes for private functions */
176
177static void ProcessCheckpointerInterrupts(void);
178static void CheckArchiveTimeout(void);
179static bool IsCheckpointOnSchedule(double progress);
180static bool FastCheckpointRequested(void);
181static bool CompactCheckpointerRequestQueue(void);
182static void UpdateSharedMemoryConfig(void);
183
184/* Signal handlers */
185static void ReqShutdownXLOG(SIGNAL_ARGS);
186
187
188/*
189 * Main entry point for checkpointer process
190 *
191 * This is invoked from AuxiliaryProcessMain, which has already created the
192 * basic execution environment, but not enabled signals yet.
193 */
194void
195 CheckpointerMain(const void *startup_data, size_t startup_data_len)
196{
197 sigjmp_buf local_sigjmp_buf;
198 MemoryContext checkpointer_context;
199
200 Assert(startup_data_len == 0);
201
202 MyBackendType = B_CHECKPOINTER;
203 AuxiliaryProcessMainCommon();
204
205 CheckpointerShmem->checkpointer_pid = MyProcPid;
206
207 /*
208 * Properly accept or ignore signals the postmaster might send us
209 *
210 * Note: we deliberately ignore SIGTERM, because during a standard Unix
211 * system shutdown cycle, init will SIGTERM all processes at once. We
212 * want to wait for the backends to exit, whereupon the postmaster will
213 * tell us it's okay to shut down (via SIGUSR2).
214 */
215 pqsignal(SIGHUP, SignalHandlerForConfigReload);
216 pqsignal(SIGINT, ReqShutdownXLOG);
217 pqsignal(SIGTERM, SIG_IGN); /* ignore SIGTERM */
218 /* SIGQUIT handler was already set up by InitPostmasterChild */
219 pqsignal(SIGALRM, SIG_IGN);
220 pqsignal(SIGPIPE, SIG_IGN);
221 pqsignal(SIGUSR1, procsignal_sigusr1_handler);
222 pqsignal(SIGUSR2, SignalHandlerForShutdownRequest);
223
224 /*
225 * Reset some signals that are accepted by postmaster but not here
226 */
227 pqsignal(SIGCHLD, SIG_DFL);
228
229 /*
230 * Initialize so that first time-driven event happens at the correct time.
231 */
232 last_checkpoint_time = last_xlog_switch_time = (pg_time_t) time(NULL);
233
234 /*
235 * Write out stats after shutdown. This needs to be called by exactly one
236 * process during a normal shutdown, and since checkpointer is shut down
237 * very late...
238 *
239 * While e.g. walsenders are active after the shutdown checkpoint has been
240 * written (and thus could produce more stats), checkpointer stays around
241 * after the shutdown checkpoint has been written. postmaster will only
242 * signal checkpointer to exit after all processes that could emit stats
243 * have been shut down.
244 */
245 before_shmem_exit(pgstat_before_server_shutdown, 0);
246
247 /*
248 * Create a memory context that we will do all our work in. We do this so
249 * that we can reset the context during error recovery and thereby avoid
250 * possible memory leaks. Formerly this code just ran in
251 * TopMemoryContext, but resetting that would be a really bad idea.
252 */
253 checkpointer_context = AllocSetContextCreate(TopMemoryContext,
254 "Checkpointer",
255 ALLOCSET_DEFAULT_SIZES);
256 MemoryContextSwitchTo(checkpointer_context);
257
258 /*
259 * If an exception is encountered, processing resumes here.
260 *
261 * You might wonder why this isn't coded as an infinite loop around a
262 * PG_TRY construct. The reason is that this is the bottom of the
263 * exception stack, and so with PG_TRY there would be no exception handler
264 * in force at all during the CATCH part. By leaving the outermost setjmp
265 * always active, we have at least some chance of recovering from an error
266 * during error recovery. (If we get into an infinite loop thereby, it
267 * will soon be stopped by overflow of elog.c's internal state stack.)
268 *
269 * Note that we use sigsetjmp(..., 1), so that the prevailing signal mask
270 * (to wit, BlockSig) will be restored when longjmp'ing to here. Thus,
271 * signals other than SIGQUIT will be blocked until we complete error
272 * recovery. It might seem that this policy makes the HOLD_INTERRUPTS()
273 * call redundant, but it is not since InterruptPending might be set
274 * already.
275 */
276 if (sigsetjmp(local_sigjmp_buf, 1) != 0)
277 {
278 /* Since not using PG_TRY, must reset error stack by hand */
279 error_context_stack = NULL;
280
281 /* Prevent interrupts while cleaning up */
282 HOLD_INTERRUPTS();
283
284 /* Report the error to the server log */
285 EmitErrorReport();
286
287 /*
288 * These operations are really just a minimal subset of
289 * AbortTransaction(). We don't have very many resources to worry
290 * about in checkpointer, but we do have LWLocks, buffers, and temp
291 * files.
292 */
293 LWLockReleaseAll();
294 ConditionVariableCancelSleep();
295 pgstat_report_wait_end();
296 pgaio_error_cleanup();
297 UnlockBuffers();
298 ReleaseAuxProcessResources(false);
299 AtEOXact_Buffers(false);
300 AtEOXact_SMgr();
301 AtEOXact_Files(false);
302 AtEOXact_HashTables(false);
303
304 /* Warn any waiting backends that the checkpoint failed. */
305 if (ckpt_active)
306 {
307 SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
308 CheckpointerShmem->ckpt_failed++;
309 CheckpointerShmem->ckpt_done = CheckpointerShmem->ckpt_started;
310 SpinLockRelease(&CheckpointerShmem->ckpt_lck);
311
312 ConditionVariableBroadcast(&CheckpointerShmem->done_cv);
313
314 ckpt_active = false;
315 }
316
317 /*
318 * Now return to normal top-level context and clear ErrorContext for
319 * next time.
320 */
321 MemoryContextSwitchTo(checkpointer_context);
322 FlushErrorState();
323
324 /* Flush any leaked data in the top-level context */
325 MemoryContextReset(checkpointer_context);
326
327 /* Now we can allow interrupts again */
328 RESUME_INTERRUPTS();
329
330 /*
331 * Sleep at least 1 second after any error. A write error is likely
332 * to be repeated, and we don't want to be filling the error logs as
333 * fast as we can.
334 */
335 pg_usleep(1000000L);
336 }
337
338 /* We can now handle ereport(ERROR) */
339 PG_exception_stack = &local_sigjmp_buf;
340
341 /*
342 * Unblock signals (they were blocked when the postmaster forked us)
343 */
344 sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
345
346 /*
347 * Ensure all shared memory values are set correctly for the config. Doing
348 * this here ensures no race conditions from other concurrent updaters.
349 */
350 UpdateSharedMemoryConfig();
351
352 /*
353 * Advertise our proc number that backends can use to wake us up while
354 * we're sleeping.
355 */
356 ProcGlobal->checkpointerProc = MyProcNumber;
357
358 /*
359 * Loop until we've been asked to write the shutdown checkpoint or
360 * terminate.
361 */
362 for (;;)
363 {
364 bool do_checkpoint = false;
365 int flags = 0;
366 pg_time_t now;
367 int elapsed_secs;
368 int cur_timeout;
369 bool chkpt_or_rstpt_requested = false;
370 bool chkpt_or_rstpt_timed = false;
371
372 /* Clear any already-pending wakeups */
373 ResetLatch(MyLatch);
374
375 /*
376 * Process any requests or signals received recently.
377 */
378 AbsorbSyncRequests();
379
380 ProcessCheckpointerInterrupts();
381 if (ShutdownXLOGPending || ShutdownRequestPending)
382 break;
383
384 /*
385 * Detect a pending checkpoint request by checking whether the flags
386 * word in shared memory is nonzero. We shouldn't need to acquire the
387 * ckpt_lck for this.
388 */
389 if (((volatile CheckpointerShmemStruct *) CheckpointerShmem)->ckpt_flags)
390 {
391 do_checkpoint = true;
392 chkpt_or_rstpt_requested = true;
393 }
394
395 /*
396 * Force a checkpoint if too much time has elapsed since the last one.
397 * Note that we count a timed checkpoint in stats only when this
398 * occurs without an external request, but we set the CAUSE_TIME flag
399 * bit even if there is also an external request.
400 */
401 now = (pg_time_t) time(NULL);
402 elapsed_secs = now - last_checkpoint_time;
403 if (elapsed_secs >= CheckPointTimeout)
404 {
405 if (!do_checkpoint)
406 chkpt_or_rstpt_timed = true;
407 do_checkpoint = true;
408 flags |= CHECKPOINT_CAUSE_TIME;
409 }
410
411 /*
412 * Do a checkpoint if requested.
413 */
414 if (do_checkpoint)
415 {
416 bool ckpt_performed = false;
417 bool do_restartpoint;
418
419 /* Check if we should perform a checkpoint or a restartpoint. */
420 do_restartpoint = RecoveryInProgress();
421
422 /*
423 * Atomically fetch the request flags to figure out what kind of a
424 * checkpoint we should perform, and increase the started-counter
425 * to acknowledge that we've started a new checkpoint.
426 */
427 SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
428 flags |= CheckpointerShmem->ckpt_flags;
429 CheckpointerShmem->ckpt_flags = 0;
430 CheckpointerShmem->ckpt_started++;
431 SpinLockRelease(&CheckpointerShmem->ckpt_lck);
432
433 ConditionVariableBroadcast(&CheckpointerShmem->start_cv);
434
435 /*
436 * The end-of-recovery checkpoint is a real checkpoint that's
437 * performed while we're still in recovery.
438 */
439 if (flags & CHECKPOINT_END_OF_RECOVERY)
440 do_restartpoint = false;
441
442 if (chkpt_or_rstpt_timed)
443 {
444 chkpt_or_rstpt_timed = false;
445 if (do_restartpoint)
446 PendingCheckpointerStats.restartpoints_timed++;
447 else
448 PendingCheckpointerStats.num_timed++;
449 }
450
451 if (chkpt_or_rstpt_requested)
452 {
453 chkpt_or_rstpt_requested = false;
454 if (do_restartpoint)
455 PendingCheckpointerStats.restartpoints_requested++;
456 else
457 PendingCheckpointerStats.num_requested++;
458 }
459
460 /*
461 * We will warn if (a) too soon since last checkpoint (whatever
462 * caused it) and (b) somebody set the CHECKPOINT_CAUSE_XLOG flag
463 * since the last checkpoint start. Note in particular that this
464 * implementation will not generate warnings caused by
465 * CheckPointTimeout < CheckPointWarning.
466 */
467 if (!do_restartpoint &&
468 (flags & CHECKPOINT_CAUSE_XLOG) &&
469 elapsed_secs < CheckPointWarning)
470 ereport(LOG,
471 (errmsg_plural("checkpoints are occurring too frequently (%d second apart)",
472 "checkpoints are occurring too frequently (%d seconds apart)",
473 elapsed_secs,
474 elapsed_secs),
475 errhint("Consider increasing the configuration parameter \"%s\".", "max_wal_size")));
476
477 /*
478 * Initialize checkpointer-private variables used during
479 * checkpoint.
480 */
481 ckpt_active = true;
482 if (do_restartpoint)
483 ckpt_start_recptr = GetXLogReplayRecPtr(NULL);
484 else
485 ckpt_start_recptr = GetInsertRecPtr();
486 ckpt_start_time = now;
487 ckpt_cached_elapsed = 0;
488
489 /*
490 * Do the checkpoint.
491 */
492 if (!do_restartpoint)
493 ckpt_performed = CreateCheckPoint(flags);
494 else
495 ckpt_performed = CreateRestartPoint(flags);
496
497 /*
498 * After any checkpoint, free all smgr objects. Otherwise we
499 * would never do so for dropped relations, as the checkpointer
500 * does not process shared invalidation messages or call
501 * AtEOXact_SMgr().
502 */
503 smgrdestroyall();
504
505 /*
506 * Indicate checkpoint completion to any waiting backends.
507 */
508 SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
509 CheckpointerShmem->ckpt_done = CheckpointerShmem->ckpt_started;
510 SpinLockRelease(&CheckpointerShmem->ckpt_lck);
511
512 ConditionVariableBroadcast(&CheckpointerShmem->done_cv);
513
514 if (!do_restartpoint)
515 {
516 /*
517 * Note we record the checkpoint start time not end time as
518 * last_checkpoint_time. This is so that time-driven
519 * checkpoints happen at a predictable spacing.
520 */
521 last_checkpoint_time = now;
522
523 if (ckpt_performed)
524 PendingCheckpointerStats.num_performed++;
525 }
526 else
527 {
528 if (ckpt_performed)
529 {
530 /*
531 * The same as for checkpoint. Please see the
532 * corresponding comment.
533 */
534 last_checkpoint_time = now;
535
536 PendingCheckpointerStats.restartpoints_performed++;
537 }
538 else
539 {
540 /*
541 * We were not able to perform the restartpoint
542 * (checkpoints throw an ERROR in case of error). Most
543 * likely because we have not received any new checkpoint
544 * WAL records since the last restartpoint. Try again in
545 * 15 s.
546 */
547 last_checkpoint_time = now - CheckPointTimeout + 15;
548 }
549 }
550
551 ckpt_active = false;
552
553 /*
554 * We may have received an interrupt during the checkpoint and the
555 * latch might have been reset (e.g. in CheckpointWriteDelay).
556 */
557 ProcessCheckpointerInterrupts();
558 if (ShutdownXLOGPending || ShutdownRequestPending)
559 break;
560 }
561
562 /* Check for archive_timeout and switch xlog files if necessary. */
563 CheckArchiveTimeout();
564
565 /* Report pending statistics to the cumulative stats system */
566 pgstat_report_checkpointer();
567 pgstat_report_wal(true);
568
569 /*
570 * If any checkpoint flags have been set, redo the loop to handle the
571 * checkpoint without sleeping.
572 */
573 if (((volatile CheckpointerShmemStruct *) CheckpointerShmem)->ckpt_flags)
574 continue;
575
576 /*
577 * Sleep until we are signaled or it's time for another checkpoint or
578 * xlog file switch.
579 */
580 now = (pg_time_t) time(NULL);
581 elapsed_secs = now - last_checkpoint_time;
582 if (elapsed_secs >= CheckPointTimeout)
583 continue; /* no sleep for us ... */
584 cur_timeout = CheckPointTimeout - elapsed_secs;
585 if (XLogArchiveTimeout > 0 && !RecoveryInProgress())
586 {
587 elapsed_secs = now - last_xlog_switch_time;
588 if (elapsed_secs >= XLogArchiveTimeout)
589 continue; /* no sleep for us ... */
590 cur_timeout = Min(cur_timeout, XLogArchiveTimeout - elapsed_secs);
591 }
592
593 (void) WaitLatch(MyLatch,
594 WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
595 cur_timeout * 1000L /* convert to ms */ ,
596 WAIT_EVENT_CHECKPOINTER_MAIN);
597 }
598
599 /*
600 * From here on, elog(ERROR) should end with exit(1), not send control
601 * back to the sigsetjmp block above.
602 */
603 ExitOnAnyError = true;
604
605 if (ShutdownXLOGPending)
606 {
607 /*
608 * Close down the database.
609 *
610 * Since ShutdownXLOG() creates restartpoint or checkpoint, and
611 * updates the statistics, increment the checkpoint request and flush
612 * out pending statistic.
613 */
614 PendingCheckpointerStats.num_requested++;
615 ShutdownXLOG(0, 0);
616 pgstat_report_checkpointer();
617 pgstat_report_wal(true);
618
619 /*
620 * Tell postmaster that we're done.
621 */
622 SendPostmasterSignal(PMSIGNAL_XLOG_IS_SHUTDOWN);
623 ShutdownXLOGPending = false;
624 }
625
626 /*
627 * Wait until we're asked to shut down. By separating the writing of the
628 * shutdown checkpoint from checkpointer exiting, checkpointer can perform
629 * some should-be-as-late-as-possible work like writing out stats.
630 */
631 for (;;)
632 {
633 /* Clear any already-pending wakeups */
634 ResetLatch(MyLatch);
635
636 ProcessCheckpointerInterrupts();
637
638 if (ShutdownRequestPending)
639 break;
640
641 (void) WaitLatch(MyLatch,
642 WL_LATCH_SET | WL_EXIT_ON_PM_DEATH,
643 0,
644 WAIT_EVENT_CHECKPOINTER_SHUTDOWN);
645 }
646
647 /* Normal exit from the checkpointer is here */
648 proc_exit(0); /* done */
649}
650
651/*
652 * Process any new interrupts.
653 */
654static void
655 ProcessCheckpointerInterrupts(void)
656{
657 if (ProcSignalBarrierPending)
658 ProcessProcSignalBarrier();
659
660 if (ConfigReloadPending)
661 {
662 ConfigReloadPending = false;
663 ProcessConfigFile(PGC_SIGHUP);
664
665 /*
666 * Checkpointer is the last process to shut down, so we ask it to hold
667 * the keys for a range of other tasks required most of which have
668 * nothing to do with checkpointing at all.
669 *
670 * For various reasons, some config values can change dynamically so
671 * the primary copy of them is held in shared memory to make sure all
672 * backends see the same value. We make Checkpointer responsible for
673 * updating the shared memory copy if the parameter setting changes
674 * because of SIGHUP.
675 */
676 UpdateSharedMemoryConfig();
677 }
678
679 /* Perform logging of memory contexts of this process */
680 if (LogMemoryContextPending)
681 ProcessLogMemoryContextInterrupt();
682}
683
684/*
685 * CheckArchiveTimeout -- check for archive_timeout and switch xlog files
686 *
687 * This will switch to a new WAL file and force an archive file write if
688 * meaningful activity is recorded in the current WAL file. This includes most
689 * writes, including just a single checkpoint record, but excludes WAL records
690 * that were inserted with the XLOG_MARK_UNIMPORTANT flag being set (like
691 * snapshots of running transactions). Such records, depending on
692 * configuration, occur on regular intervals and don't contain important
693 * information. This avoids generating archives with a few unimportant
694 * records.
695 */
696static void
697 CheckArchiveTimeout(void)
698{
699 pg_time_t now;
700 pg_time_t last_time;
701 XLogRecPtr last_switch_lsn;
702
703 if (XLogArchiveTimeout <= 0 || RecoveryInProgress())
704 return;
705
706 now = (pg_time_t) time(NULL);
707
708 /* First we do a quick check using possibly-stale local state. */
709 if ((int) (now - last_xlog_switch_time) < XLogArchiveTimeout)
710 return;
711
712 /*
713 * Update local state ... note that last_xlog_switch_time is the last time
714 * a switch was performed *or requested*.
715 */
716 last_time = GetLastSegSwitchData(&last_switch_lsn);
717
718 last_xlog_switch_time = Max(last_xlog_switch_time, last_time);
719
720 /* Now we can do the real checks */
721 if ((int) (now - last_xlog_switch_time) >= XLogArchiveTimeout)
722 {
723 /*
724 * Switch segment only when "important" WAL has been logged since the
725 * last segment switch (last_switch_lsn points to end of segment
726 * switch occurred in).
727 */
728 if (GetLastImportantRecPtr() > last_switch_lsn)
729 {
730 XLogRecPtr switchpoint;
731
732 /* mark switch as unimportant, avoids triggering checkpoints */
733 switchpoint = RequestXLogSwitch(true);
734
735 /*
736 * If the returned pointer points exactly to a segment boundary,
737 * assume nothing happened.
738 */
739 if (XLogSegmentOffset(switchpoint, wal_segment_size) != 0)
740 elog(DEBUG1, "write-ahead log switch forced (\"archive_timeout\"=%d)",
741 XLogArchiveTimeout);
742 }
743
744 /*
745 * Update state in any case, so we don't retry constantly when the
746 * system is idle.
747 */
748 last_xlog_switch_time = now;
749 }
750}
751
752/*
753 * Returns true if a fast checkpoint request is pending. (Note that this does
754 * not check the *current* checkpoint's FAST flag, but whether there is one
755 * pending behind it.)
756 */
757static bool
758 FastCheckpointRequested(void)
759{
760 volatile CheckpointerShmemStruct *cps = CheckpointerShmem;
761
762 /*
763 * We don't need to acquire the ckpt_lck in this case because we're only
764 * looking at a single flag bit.
765 */
766 if (cps->ckpt_flags & CHECKPOINT_FAST)
767 return true;
768 return false;
769}
770
771/*
772 * CheckpointWriteDelay -- control rate of checkpoint
773 *
774 * This function is called after each page write performed by BufferSync().
775 * It is responsible for throttling BufferSync()'s write rate to hit
776 * checkpoint_completion_target.
777 *
778 * The checkpoint request flags should be passed in; currently the only one
779 * examined is CHECKPOINT_FAST, which disables delays between writes.
780 *
781 * 'progress' is an estimate of how much of the work has been done, as a
782 * fraction between 0.0 meaning none, and 1.0 meaning all done.
783 */
784void
785 CheckpointWriteDelay(int flags, double progress)
786{
787 static int absorb_counter = WRITES_PER_ABSORB;
788
789 /* Do nothing if checkpoint is being executed by non-checkpointer process */
790 if (!AmCheckpointerProcess())
791 return;
792
793 /*
794 * Perform the usual duties and take a nap, unless we're behind schedule,
795 * in which case we just try to catch up as quickly as possible.
796 */
797 if (!(flags & CHECKPOINT_FAST) &&
798 !ShutdownXLOGPending &&
799 !ShutdownRequestPending &&
800 !FastCheckpointRequested() &&
801 IsCheckpointOnSchedule(progress))
802 {
803 if (ConfigReloadPending)
804 {
805 ConfigReloadPending = false;
806 ProcessConfigFile(PGC_SIGHUP);
807 /* update shmem copies of config variables */
808 UpdateSharedMemoryConfig();
809 }
810
811 AbsorbSyncRequests();
812 absorb_counter = WRITES_PER_ABSORB;
813
814 CheckArchiveTimeout();
815
816 /* Report interim statistics to the cumulative stats system */
817 pgstat_report_checkpointer();
818
819 /*
820 * This sleep used to be connected to bgwriter_delay, typically 200ms.
821 * That resulted in more frequent wakeups if not much work to do.
822 * Checkpointer and bgwriter are no longer related so take the Big
823 * Sleep.
824 */
825 WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH | WL_TIMEOUT,
826 100,
827 WAIT_EVENT_CHECKPOINT_WRITE_DELAY);
828 ResetLatch(MyLatch);
829 }
830 else if (--absorb_counter <= 0)
831 {
832 /*
833 * Absorb pending fsync requests after each WRITES_PER_ABSORB write
834 * operations even when we don't sleep, to prevent overflow of the
835 * fsync request queue.
836 */
837 AbsorbSyncRequests();
838 absorb_counter = WRITES_PER_ABSORB;
839 }
840
841 /* Check for barrier events. */
842 if (ProcSignalBarrierPending)
843 ProcessProcSignalBarrier();
844}
845
846/*
847 * IsCheckpointOnSchedule -- are we on schedule to finish this checkpoint
848 * (or restartpoint) in time?
849 *
850 * Compares the current progress against the time/segments elapsed since last
851 * checkpoint, and returns true if the progress we've made this far is greater
852 * than the elapsed time/segments.
853 */
854static bool
855 IsCheckpointOnSchedule(double progress)
856{
857 XLogRecPtr recptr;
858 struct timeval now;
859 double elapsed_xlogs,
860 elapsed_time;
861
862 Assert(ckpt_active);
863
864 /* Scale progress according to checkpoint_completion_target. */
865 progress *= CheckPointCompletionTarget;
866
867 /*
868 * Check against the cached value first. Only do the more expensive
869 * calculations once we reach the target previously calculated. Since
870 * neither time or WAL insert pointer moves backwards, a freshly
871 * calculated value can only be greater than or equal to the cached value.
872 */
873 if (progress < ckpt_cached_elapsed)
874 return false;
875
876 /*
877 * Check progress against WAL segments written and CheckPointSegments.
878 *
879 * We compare the current WAL insert location against the location
880 * computed before calling CreateCheckPoint. The code in XLogInsert that
881 * actually triggers a checkpoint when CheckPointSegments is exceeded
882 * compares against RedoRecPtr, so this is not completely accurate.
883 * However, it's good enough for our purposes, we're only calculating an
884 * estimate anyway.
885 *
886 * During recovery, we compare last replayed WAL record's location with
887 * the location computed before calling CreateRestartPoint. That maintains
888 * the same pacing as we have during checkpoints in normal operation, but
889 * we might exceed max_wal_size by a fair amount. That's because there can
890 * be a large gap between a checkpoint's redo-pointer and the checkpoint
891 * record itself, and we only start the restartpoint after we've seen the
892 * checkpoint record. (The gap is typically up to CheckPointSegments *
893 * checkpoint_completion_target where checkpoint_completion_target is the
894 * value that was in effect when the WAL was generated).
895 */
896 if (RecoveryInProgress())
897 recptr = GetXLogReplayRecPtr(NULL);
898 else
899 recptr = GetInsertRecPtr();
900 elapsed_xlogs = (((double) (recptr - ckpt_start_recptr)) /
901 wal_segment_size) / CheckPointSegments;
902
903 if (progress < elapsed_xlogs)
904 {
905 ckpt_cached_elapsed = elapsed_xlogs;
906 return false;
907 }
908
909 /*
910 * Check progress against time elapsed and checkpoint_timeout.
911 */
912 gettimeofday(&now, NULL);
913 elapsed_time = ((double) ((pg_time_t) now.tv_sec - ckpt_start_time) +
914 now.tv_usec / 1000000.0) / CheckPointTimeout;
915
916 if (progress < elapsed_time)
917 {
918 ckpt_cached_elapsed = elapsed_time;
919 return false;
920 }
921
922 /* It looks like we're on schedule. */
923 return true;
924}
925
926
927/* --------------------------------
928 * signal handler routines
929 * --------------------------------
930 */
931
932/* SIGINT: set flag to trigger writing of shutdown checkpoint */
933static void
934 ReqShutdownXLOG(SIGNAL_ARGS)
935{
936 ShutdownXLOGPending = true;
937 SetLatch(MyLatch);
938}
939
940
941/* --------------------------------
942 * communication with backends
943 * --------------------------------
944 */
945
946/*
947 * CheckpointerShmemSize
948 * Compute space needed for checkpointer-related shared memory
949 */
950Size
951 CheckpointerShmemSize(void)
952{
953 Size size;
954
955 /*
956 * The size of the requests[] array is arbitrarily set equal to NBuffers.
957 * But there is a cap of MAX_CHECKPOINT_REQUESTS to prevent accumulating
958 * too many checkpoint requests in the ring buffer.
959 */
960 size = offsetof(CheckpointerShmemStruct, requests);
961 size = add_size(size, mul_size(Min(NBuffers,
962 MAX_CHECKPOINT_REQUESTS),
963 sizeof(CheckpointerRequest)));
964
965 return size;
966}
967
968/*
969 * CheckpointerShmemInit
970 * Allocate and initialize checkpointer-related shared memory
971 */
972void
973 CheckpointerShmemInit(void)
974{
975 Size size = CheckpointerShmemSize();
976 bool found;
977
978 CheckpointerShmem = (CheckpointerShmemStruct *)
979 ShmemInitStruct("Checkpointer Data",
980 size,
981 &found);
982
983 if (!found)
984 {
985 /*
986 * First time through, so initialize. Note that we zero the whole
987 * requests array; this is so that CompactCheckpointerRequestQueue can
988 * assume that any pad bytes in the request structs are zeroes.
989 */
990 MemSet(CheckpointerShmem, 0, size);
991 SpinLockInit(&CheckpointerShmem->ckpt_lck);
992 CheckpointerShmem->max_requests = Min(NBuffers, MAX_CHECKPOINT_REQUESTS);
993 CheckpointerShmem->head = CheckpointerShmem->tail = 0;
994 ConditionVariableInit(&CheckpointerShmem->start_cv);
995 ConditionVariableInit(&CheckpointerShmem->done_cv);
996 }
997}
998
999/*
1000 * ExecCheckpoint
1001 * Primary entry point for manual CHECKPOINT commands
1002 *
1003 * This is mainly a wrapper for RequestCheckpoint().
1004 */
1005void
1006 ExecCheckpoint(ParseState *pstate, CheckPointStmt *stmt)
1007{
1008 bool fast = true;
1009 bool unlogged = false;
1010
1011 foreach_ptr(DefElem, opt, stmt->options)
1012 {
1013 if (strcmp(opt->defname, "mode") == 0)
1014 {
1015 char *mode = defGetString(opt);
1016
1017 if (strcmp(mode, "spread") == 0)
1018 fast = false;
1019 else if (strcmp(mode, "fast") != 0)
1020 ereport(ERROR,
1021 (errcode(ERRCODE_SYNTAX_ERROR),
1022 errmsg("unrecognized MODE option \"%s\"", mode),
1023 parser_errposition(pstate, opt->location)));
1024 }
1025 else if (strcmp(opt->defname, "flush_unlogged") == 0)
1026 unlogged = defGetBoolean(opt);
1027 else
1028 ereport(ERROR,
1029 (errcode(ERRCODE_SYNTAX_ERROR),
1030 errmsg("unrecognized CHECKPOINT option \"%s\"", opt->defname),
1031 parser_errposition(pstate, opt->location)));
1032 }
1033
1034 if (!has_privs_of_role(GetUserId(), ROLE_PG_CHECKPOINT))
1035 ereport(ERROR,
1036 (errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
1037 /* translator: %s is name of an SQL command (e.g., CHECKPOINT) */
1038 errmsg("permission denied to execute %s command",
1039 "CHECKPOINT"),
1040 errdetail("Only roles with privileges of the \"%s\" role may execute this command.",
1041 "pg_checkpoint")));
1042
1043 RequestCheckpoint(CHECKPOINT_WAIT |
1044 (fast ? CHECKPOINT_FAST : 0) |
1045 (unlogged ? CHECKPOINT_FLUSH_UNLOGGED : 0) |
1046 (RecoveryInProgress() ? 0 : CHECKPOINT_FORCE));
1047}
1048
1049/*
1050 * RequestCheckpoint
1051 * Called in backend processes to request a checkpoint
1052 *
1053 * flags is a bitwise OR of the following:
1054 * CHECKPOINT_IS_SHUTDOWN: checkpoint is for database shutdown.
1055 * CHECKPOINT_END_OF_RECOVERY: checkpoint is for end of WAL recovery.
1056 * CHECKPOINT_FAST: finish the checkpoint ASAP,
1057 * ignoring checkpoint_completion_target parameter.
1058 * CHECKPOINT_FORCE: force a checkpoint even if no XLOG activity has occurred
1059 * since the last one (implied by CHECKPOINT_IS_SHUTDOWN or
1060 * CHECKPOINT_END_OF_RECOVERY, and the CHECKPOINT command).
1061 * CHECKPOINT_WAIT: wait for completion before returning (otherwise,
1062 * just signal checkpointer to do it, and return).
1063 * CHECKPOINT_CAUSE_XLOG: checkpoint is requested due to xlog filling.
1064 * (This affects logging, and in particular enables CheckPointWarning.)
1065 */
1066void
1067 RequestCheckpoint(int flags)
1068{
1069 int ntries;
1070 int old_failed,
1071 old_started;
1072
1073 /*
1074 * If in a standalone backend, just do it ourselves.
1075 */
1076 if (!IsPostmasterEnvironment)
1077 {
1078 /*
1079 * There's no point in doing slow checkpoints in a standalone backend,
1080 * because there's no other backends the checkpoint could disrupt.
1081 */
1082 CreateCheckPoint(flags | CHECKPOINT_FAST);
1083
1084 /* Free all smgr objects, as CheckpointerMain() normally would. */
1085 smgrdestroyall();
1086
1087 return;
1088 }
1089
1090 /*
1091 * Atomically set the request flags, and take a snapshot of the counters.
1092 * When we see ckpt_started > old_started, we know the flags we set here
1093 * have been seen by checkpointer.
1094 *
1095 * Note that we OR the flags with any existing flags, to avoid overriding
1096 * a "stronger" request by another backend. The flag senses must be
1097 * chosen to make this work!
1098 */
1099 SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
1100
1101 old_failed = CheckpointerShmem->ckpt_failed;
1102 old_started = CheckpointerShmem->ckpt_started;
1103 CheckpointerShmem->ckpt_flags |= (flags | CHECKPOINT_REQUESTED);
1104
1105 SpinLockRelease(&CheckpointerShmem->ckpt_lck);
1106
1107 /*
1108 * Set checkpointer's latch to request checkpoint. It's possible that the
1109 * checkpointer hasn't started yet, so we will retry a few times if
1110 * needed. (Actually, more than a few times, since on slow or overloaded
1111 * buildfarm machines, it's been observed that the checkpointer can take
1112 * several seconds to start.) However, if not told to wait for the
1113 * checkpoint to occur, we consider failure to set the latch to be
1114 * nonfatal and merely LOG it. The checkpointer should see the request
1115 * when it does start, with or without the SetLatch().
1116 */
1117#define MAX_SIGNAL_TRIES 600 /* max wait 60.0 sec */
1118 for (ntries = 0;; ntries++)
1119 {
1120 volatile PROC_HDR *procglobal = ProcGlobal;
1121 ProcNumber checkpointerProc = procglobal->checkpointerProc;
1122
1123 if (checkpointerProc == INVALID_PROC_NUMBER)
1124 {
1125 if (ntries >= MAX_SIGNAL_TRIES || !(flags & CHECKPOINT_WAIT))
1126 {
1127 elog((flags & CHECKPOINT_WAIT) ? ERROR : LOG,
1128 "could not notify checkpoint: checkpointer is not running");
1129 break;
1130 }
1131 }
1132 else
1133 {
1134 SetLatch(&GetPGProcByNumber(checkpointerProc)->procLatch);
1135 /* notified successfully */
1136 break;
1137 }
1138
1139 CHECK_FOR_INTERRUPTS();
1140 pg_usleep(100000L); /* wait 0.1 sec, then retry */
1141 }
1142
1143 /*
1144 * If requested, wait for completion. We detect completion according to
1145 * the algorithm given above.
1146 */
1147 if (flags & CHECKPOINT_WAIT)
1148 {
1149 int new_started,
1150 new_failed;
1151
1152 /* Wait for a new checkpoint to start. */
1153 ConditionVariablePrepareToSleep(&CheckpointerShmem->start_cv);
1154 for (;;)
1155 {
1156 SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
1157 new_started = CheckpointerShmem->ckpt_started;
1158 SpinLockRelease(&CheckpointerShmem->ckpt_lck);
1159
1160 if (new_started != old_started)
1161 break;
1162
1163 ConditionVariableSleep(&CheckpointerShmem->start_cv,
1164 WAIT_EVENT_CHECKPOINT_START);
1165 }
1166 ConditionVariableCancelSleep();
1167
1168 /*
1169 * We are waiting for ckpt_done >= new_started, in a modulo sense.
1170 */
1171 ConditionVariablePrepareToSleep(&CheckpointerShmem->done_cv);
1172 for (;;)
1173 {
1174 int new_done;
1175
1176 SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
1177 new_done = CheckpointerShmem->ckpt_done;
1178 new_failed = CheckpointerShmem->ckpt_failed;
1179 SpinLockRelease(&CheckpointerShmem->ckpt_lck);
1180
1181 if (new_done - new_started >= 0)
1182 break;
1183
1184 ConditionVariableSleep(&CheckpointerShmem->done_cv,
1185 WAIT_EVENT_CHECKPOINT_DONE);
1186 }
1187 ConditionVariableCancelSleep();
1188
1189 if (new_failed != old_failed)
1190 ereport(ERROR,
1191 (errmsg("checkpoint request failed"),
1192 errhint("Consult recent messages in the server log for details.")));
1193 }
1194}
1195
1196/*
1197 * ForwardSyncRequest
1198 * Forward a file-fsync request from a backend to the checkpointer
1199 *
1200 * Whenever a backend is compelled to write directly to a relation
1201 * (which should be seldom, if the background writer is getting its job done),
1202 * the backend calls this routine to pass over knowledge that the relation
1203 * is dirty and must be fsync'd before next checkpoint. We also use this
1204 * opportunity to count such writes for statistical purposes.
1205 *
1206 * To avoid holding the lock for longer than necessary, we normally write
1207 * to the requests[] queue without checking for duplicates. The checkpointer
1208 * will have to eliminate dups internally anyway. However, if we discover
1209 * that the queue is full, we make a pass over the entire queue to compact
1210 * it. This is somewhat expensive, but the alternative is for the backend
1211 * to perform its own fsync, which is far more expensive in practice. It
1212 * is theoretically possible a backend fsync might still be necessary, if
1213 * the queue is full and contains no duplicate entries. In that case, we
1214 * let the backend know by returning false.
1215 */
1216bool
1217 ForwardSyncRequest(const FileTag *ftag, SyncRequestType type)
1218{
1219 CheckpointerRequest *request;
1220 bool too_full;
1221 int insert_pos;
1222
1223 if (!IsUnderPostmaster)
1224 return false; /* probably shouldn't even get here */
1225
1226 if (AmCheckpointerProcess())
1227 elog(ERROR, "ForwardSyncRequest must not be called in checkpointer");
1228
1229 LWLockAcquire(CheckpointerCommLock, LW_EXCLUSIVE);
1230
1231 /*
1232 * If the checkpointer isn't running or the request queue is full, the
1233 * backend will have to perform its own fsync request. But before forcing
1234 * that to happen, we can try to compact the request queue.
1235 */
1236 if (CheckpointerShmem->checkpointer_pid == 0 ||
1237 (CheckpointerShmem->num_requests >= CheckpointerShmem->max_requests &&
1238 !CompactCheckpointerRequestQueue()))
1239 {
1240 LWLockRelease(CheckpointerCommLock);
1241 return false;
1242 }
1243
1244 /* OK, insert request */
1245 insert_pos = CheckpointerShmem->tail;
1246 request = &CheckpointerShmem->requests[insert_pos];
1247 request->ftag = *ftag;
1248 request->type = type;
1249
1250 CheckpointerShmem->tail = (CheckpointerShmem->tail + 1) % CheckpointerShmem->max_requests;
1251 CheckpointerShmem->num_requests++;
1252
1253 /* If queue is more than half full, nudge the checkpointer to empty it */
1254 too_full = (CheckpointerShmem->num_requests >=
1255 CheckpointerShmem->max_requests / 2);
1256
1257 LWLockRelease(CheckpointerCommLock);
1258
1259 /* ... but not till after we release the lock */
1260 if (too_full)
1261 {
1262 volatile PROC_HDR *procglobal = ProcGlobal;
1263 ProcNumber checkpointerProc = procglobal->checkpointerProc;
1264
1265 if (checkpointerProc != INVALID_PROC_NUMBER)
1266 SetLatch(&GetPGProcByNumber(checkpointerProc)->procLatch);
1267 }
1268
1269 return true;
1270}
1271
1272/*
1273 * CompactCheckpointerRequestQueue
1274 * Remove duplicates from the request queue to avoid backend fsyncs.
1275 * Returns "true" if any entries were removed.
1276 *
1277 * Although a full fsync request queue is not common, it can lead to severe
1278 * performance problems when it does happen. So far, this situation has
1279 * only been observed to occur when the system is under heavy write load,
1280 * and especially during the "sync" phase of a checkpoint. Without this
1281 * logic, each backend begins doing an fsync for every block written, which
1282 * gets very expensive and can slow down the whole system.
1283 *
1284 * Trying to do this every time the queue is full could lose if there
1285 * aren't any removable entries. But that should be vanishingly rare in
1286 * practice: there's one queue entry per shared buffer.
1287 */
1288static bool
1289 CompactCheckpointerRequestQueue(void)
1290{
1291 struct CheckpointerSlotMapping
1292 {
1293 CheckpointerRequest request;
1294 int ring_idx;
1295 };
1296
1297 int n;
1298 int num_skipped = 0;
1299 int head;
1300 int max_requests;
1301 int num_requests;
1302 int read_idx,
1303 write_idx;
1304 HASHCTL ctl;
1305 HTAB *htab;
1306 bool *skip_slot;
1307
1308 /* must hold CheckpointerCommLock in exclusive mode */
1309 Assert(LWLockHeldByMe(CheckpointerCommLock));
1310
1311 /* Avoid memory allocations in a critical section. */
1312 if (CritSectionCount > 0)
1313 return false;
1314
1315 max_requests = CheckpointerShmem->max_requests;
1316 num_requests = CheckpointerShmem->num_requests;
1317
1318 /* Initialize skip_slot array */
1319 skip_slot = palloc0(sizeof(bool) * max_requests);
1320
1321 head = CheckpointerShmem->head;
1322
1323 /* Initialize temporary hash table */
1324 ctl.keysize = sizeof(CheckpointerRequest);
1325 ctl.entrysize = sizeof(struct CheckpointerSlotMapping);
1326 ctl.hcxt = CurrentMemoryContext;
1327
1328 htab = hash_create("CompactCheckpointerRequestQueue",
1329 CheckpointerShmem->num_requests,
1330 &ctl,
1331 HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
1332
1333 /*
1334 * The basic idea here is that a request can be skipped if it's followed
1335 * by a later, identical request. It might seem more sensible to work
1336 * backwards from the end of the queue and check whether a request is
1337 * *preceded* by an earlier, identical request, in the hopes of doing less
1338 * copying. But that might change the semantics, if there's an
1339 * intervening SYNC_FORGET_REQUEST or SYNC_FILTER_REQUEST, so we do it
1340 * this way. It would be possible to be even smarter if we made the code
1341 * below understand the specific semantics of such requests (it could blow
1342 * away preceding entries that would end up being canceled anyhow), but
1343 * it's not clear that the extra complexity would buy us anything.
1344 */
1345 read_idx = head;
1346 for (n = 0; n < num_requests; n++)
1347 {
1348 CheckpointerRequest *request;
1349 struct CheckpointerSlotMapping *slotmap;
1350 bool found;
1351
1352 /*
1353 * We use the request struct directly as a hashtable key. This
1354 * assumes that any padding bytes in the structs are consistently the
1355 * same, which should be okay because we zeroed them in
1356 * CheckpointerShmemInit. Note also that RelFileLocator had better
1357 * contain no pad bytes.
1358 */
1359 request = &CheckpointerShmem->requests[read_idx];
1360 slotmap = hash_search(htab, request, HASH_ENTER, &found);
1361 if (found)
1362 {
1363 /* Duplicate, so mark the previous occurrence as skippable */
1364 skip_slot[slotmap->ring_idx] = true;
1365 num_skipped++;
1366 }
1367 /* Remember slot containing latest occurrence of this request value */
1368 slotmap->ring_idx = read_idx;
1369
1370 /* Move to the next request in the ring buffer */
1371 read_idx = (read_idx + 1) % max_requests;
1372 }
1373
1374 /* Done with the hash table. */
1375 hash_destroy(htab);
1376
1377 /* If no duplicates, we're out of luck. */
1378 if (!num_skipped)
1379 {
1380 pfree(skip_slot);
1381 return false;
1382 }
1383
1384 /* We found some duplicates; remove them. */
1385 read_idx = write_idx = head;
1386 for (n = 0; n < num_requests; n++)
1387 {
1388 /* If this slot is NOT skipped, keep it */
1389 if (!skip_slot[read_idx])
1390 {
1391 /* If the read and write positions are different, copy the request */
1392 if (write_idx != read_idx)
1393 CheckpointerShmem->requests[write_idx] =
1394 CheckpointerShmem->requests[read_idx];
1395
1396 /* Advance the write position */
1397 write_idx = (write_idx + 1) % max_requests;
1398 }
1399
1400 read_idx = (read_idx + 1) % max_requests;
1401 }
1402
1403 /*
1404 * Update ring buffer state: head remains the same, tail moves, count
1405 * decreases
1406 */
1407 CheckpointerShmem->tail = write_idx;
1408 CheckpointerShmem->num_requests -= num_skipped;
1409
1410 ereport(DEBUG1,
1411 (errmsg_internal("compacted fsync request queue from %d entries to %d entries",
1412 num_requests, CheckpointerShmem->num_requests)));
1413
1414 /* Cleanup. */
1415 pfree(skip_slot);
1416 return true;
1417}
1418
1419/*
1420 * AbsorbSyncRequests
1421 * Retrieve queued sync requests and pass them to sync mechanism.
1422 *
1423 * This is exported because it must be called during CreateCheckPoint;
1424 * we have to be sure we have accepted all pending requests just before
1425 * we start fsync'ing. Since CreateCheckPoint sometimes runs in
1426 * non-checkpointer processes, do nothing if not checkpointer.
1427 */
1428void
1429 AbsorbSyncRequests(void)
1430{
1431 CheckpointerRequest *requests = NULL;
1432 CheckpointerRequest *request;
1433 int n,
1434 i;
1435 bool loop;
1436
1437 if (!AmCheckpointerProcess())
1438 return;
1439
1440 do
1441 {
1442 LWLockAcquire(CheckpointerCommLock, LW_EXCLUSIVE);
1443
1444 /*---
1445 * We try to avoid holding the lock for a long time by:
1446 * 1. Copying the request array and processing the requests after
1447 * releasing the lock;
1448 * 2. Processing not the whole queue, but only batches of
1449 * CKPT_REQ_BATCH_SIZE at once.
1450 *
1451 * Once we have cleared the requests from shared memory, we must
1452 * PANIC if we then fail to absorb them (e.g., because our hashtable
1453 * runs out of memory). This is because the system cannot run safely
1454 * if we are unable to fsync what we have been told to fsync.
1455 * Fortunately, the hashtable is so small that the problem is quite
1456 * unlikely to arise in practice.
1457 *
1458 * Note: The maximum possible size of a ring buffer is
1459 * MAX_CHECKPOINT_REQUESTS entries, which fit into a maximum palloc
1460 * allocation size of 1Gb. Our maximum batch size,
1461 * CKPT_REQ_BATCH_SIZE, is even smaller.
1462 */
1463 n = Min(CheckpointerShmem->num_requests, CKPT_REQ_BATCH_SIZE);
1464 if (n > 0)
1465 {
1466 if (!requests)
1467 requests = (CheckpointerRequest *) palloc(n * sizeof(CheckpointerRequest));
1468
1469 for (i = 0; i < n; i++)
1470 {
1471 requests[i] = CheckpointerShmem->requests[CheckpointerShmem->head];
1472 CheckpointerShmem->head = (CheckpointerShmem->head + 1) % CheckpointerShmem->max_requests;
1473 }
1474
1475 CheckpointerShmem->num_requests -= n;
1476
1477 }
1478
1479 START_CRIT_SECTION();
1480
1481 /* Are there any requests in the queue? If so, keep going. */
1482 loop = CheckpointerShmem->num_requests != 0;
1483
1484 LWLockRelease(CheckpointerCommLock);
1485
1486 for (request = requests; n > 0; request++, n--)
1487 RememberSyncRequest(&request->ftag, request->type);
1488
1489 END_CRIT_SECTION();
1490 } while (loop);
1491
1492 if (requests)
1493 pfree(requests);
1494}
1495
1496/*
1497 * Update any shared memory configurations based on config parameters
1498 */
1499static void
1500 UpdateSharedMemoryConfig(void)
1501{
1502 /* update global shmem state for sync rep */
1503 SyncRepUpdateSyncStandbysDefined();
1504
1505 /*
1506 * If full_page_writes has been changed by SIGHUP, we update it in shared
1507 * memory and write an XLOG_FPW_CHANGE record.
1508 */
1509 UpdateFullPageWrites();
1510
1511 elog(DEBUG2, "checkpointer updated shared memory configuration values");
1512}
1513
1514/*
1515 * FirstCallSinceLastCheckpoint allows a process to take an action once
1516 * per checkpoint cycle by asynchronously checking for checkpoint completion.
1517 */
1518bool
1519 FirstCallSinceLastCheckpoint(void)
1520{
1521 static int ckpt_done = 0;
1522 int new_done;
1523 bool FirstCall = false;
1524
1525 SpinLockAcquire(&CheckpointerShmem->ckpt_lck);
1526 new_done = CheckpointerShmem->ckpt_done;
1527 SpinLockRelease(&CheckpointerShmem->ckpt_lck);
1528
1529 if (new_done != ckpt_done)
1530 FirstCall = true;
1531
1532 ckpt_done = new_done;
1533
1534 return FirstCall;
1535}
bool has_privs_of_role(Oid member, Oid role)
Definition: acl.c:5284
void pgaio_error_cleanup(void)
Definition: aio.c:1162
void AuxiliaryProcessMainCommon(void)
Definition: auxprocess.c:39
sigset_t UnBlockSig
Definition: pqsignal.c:22
Datum now(PG_FUNCTION_ARGS)
Definition: timestamp.c:1609
void AtEOXact_Buffers(bool isCommit)
Definition: bufmgr.c:3965
void UnlockBuffers(void)
Definition: bufmgr.c:5544
#define Min(x, y)
Definition: c.h:1003
#define Max(x, y)
Definition: c.h:997
#define SIGNAL_ARGS
Definition: c.h:1348
#define FLEXIBLE_ARRAY_MEMBER
Definition: c.h:470
#define MemSet(start, val, len)
Definition: c.h:1019
size_t Size
Definition: c.h:610
static void UpdateSharedMemoryConfig(void)
Definition: checkpointer.c:1500
static bool FastCheckpointRequested(void)
Definition: checkpointer.c:758
static XLogRecPtr ckpt_start_recptr
Definition: checkpointer.c:169
static bool IsCheckpointOnSchedule(double progress)
Definition: checkpointer.c:855
bool ForwardSyncRequest(const FileTag *ftag, SyncRequestType type)
Definition: checkpointer.c:1217
static void ReqShutdownXLOG(SIGNAL_ARGS)
Definition: checkpointer.c:934
static void CheckArchiveTimeout(void)
Definition: checkpointer.c:697
static double ckpt_cached_elapsed
Definition: checkpointer.c:170
static bool ckpt_active
Definition: checkpointer.c:164
void CheckpointerMain(const void *startup_data, size_t startup_data_len)
Definition: checkpointer.c:195
static bool CompactCheckpointerRequestQueue(void)
Definition: checkpointer.c:1289
static void ProcessCheckpointerInterrupts(void)
Definition: checkpointer.c:655
static volatile sig_atomic_t ShutdownXLOGPending
Definition: checkpointer.c:165
#define CKPT_REQ_BATCH_SIZE
Definition: checkpointer.c:149
#define MAX_SIGNAL_TRIES
void AbsorbSyncRequests(void)
Definition: checkpointer.c:1429
#define WRITES_PER_ABSORB
Definition: checkpointer.c:146
double CheckPointCompletionTarget
Definition: checkpointer.c:159
static pg_time_t last_xlog_switch_time
Definition: checkpointer.c:173
#define MAX_CHECKPOINT_REQUESTS
Definition: checkpointer.c:152
int CheckPointWarning
Definition: checkpointer.c:158
void CheckpointerShmemInit(void)
Definition: checkpointer.c:973
bool FirstCallSinceLastCheckpoint(void)
Definition: checkpointer.c:1519
static CheckpointerShmemStruct * CheckpointerShmem
Definition: checkpointer.c:143
int CheckPointTimeout
Definition: checkpointer.c:157
void RequestCheckpoint(int flags)
Definition: checkpointer.c:1067
static pg_time_t last_checkpoint_time
Definition: checkpointer.c:172
void ExecCheckpoint(ParseState *pstate, CheckPointStmt *stmt)
Definition: checkpointer.c:1006
void CheckpointWriteDelay(int flags, double progress)
Definition: checkpointer.c:785
static pg_time_t ckpt_start_time
Definition: checkpointer.c:168
Size CheckpointerShmemSize(void)
Definition: checkpointer.c:951
bool ConditionVariableCancelSleep(void)
void ConditionVariableBroadcast(ConditionVariable *cv)
void ConditionVariablePrepareToSleep(ConditionVariable *cv)
void ConditionVariableInit(ConditionVariable *cv)
void ConditionVariableSleep(ConditionVariable *cv, uint32 wait_event_info)
char * defGetString(DefElem *def)
Definition: define.c:35
bool defGetBoolean(DefElem *def)
Definition: define.c:94
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:952
void AtEOXact_HashTables(bool isCommit)
Definition: dynahash.c:1931
HTAB * hash_create(const char *tabname, int64 nelem, const HASHCTL *info, int flags)
Definition: dynahash.c:358
void hash_destroy(HTAB *hashp)
Definition: dynahash.c:865
int errmsg_plural(const char *fmt_singular, const char *fmt_plural, unsigned long n,...)
Definition: elog.c:1184
int errmsg_internal(const char *fmt,...)
Definition: elog.c:1161
void EmitErrorReport(void)
Definition: elog.c:1695
int errdetail(const char *fmt,...)
Definition: elog.c:1207
ErrorContextCallback * error_context_stack
Definition: elog.c:95
void FlushErrorState(void)
Definition: elog.c:1875
int errhint(const char *fmt,...)
Definition: elog.c:1321
int errcode(int sqlerrcode)
Definition: elog.c:854
int errmsg(const char *fmt,...)
Definition: elog.c:1071
sigjmp_buf * PG_exception_stack
Definition: elog.c:97
#define LOG
Definition: elog.h:31
#define DEBUG2
Definition: elog.h:29
#define DEBUG1
Definition: elog.h:30
#define ERROR
Definition: elog.h:39
#define elog(elevel,...)
Definition: elog.h:226
#define ereport(elevel,...)
Definition: elog.h:150
static double elapsed_time(instr_time *starttime)
Definition: explain.c:1164
void AtEOXact_Files(bool isCommit)
Definition: fd.c:3226
volatile sig_atomic_t LogMemoryContextPending
Definition: globals.c:41
volatile sig_atomic_t ProcSignalBarrierPending
Definition: globals.c:40
int NBuffers
Definition: globals.c:142
int MyProcPid
Definition: globals.c:47
ProcNumber MyProcNumber
Definition: globals.c:90
bool IsUnderPostmaster
Definition: globals.c:120
volatile uint32 CritSectionCount
Definition: globals.c:45
bool ExitOnAnyError
Definition: globals.c:123
bool IsPostmasterEnvironment
Definition: globals.c:119
struct Latch * MyLatch
Definition: globals.c:63
void ProcessConfigFile(GucContext context)
Definition: guc-file.l:120
@ PGC_SIGHUP
Definition: guc.h:75
Assert(PointerIsAligned(start, uint64))
@ HASH_ENTER
Definition: hsearch.h:114
#define HASH_CONTEXT
Definition: hsearch.h:102
#define HASH_ELEM
Definition: hsearch.h:95
#define HASH_BLOBS
Definition: hsearch.h:97
#define stmt
Definition: indent_codes.h:59
void SignalHandlerForShutdownRequest(SIGNAL_ARGS)
Definition: interrupt.c:104
volatile sig_atomic_t ShutdownRequestPending
Definition: interrupt.c:28
volatile sig_atomic_t ConfigReloadPending
Definition: interrupt.c:27
void SignalHandlerForConfigReload(SIGNAL_ARGS)
Definition: interrupt.c:61
void before_shmem_exit(pg_on_exit_callback function, Datum arg)
Definition: ipc.c:337
void proc_exit(int code)
Definition: ipc.c:104
i
int i
Definition: isn.c:77
void SetLatch(Latch *latch)
Definition: latch.c:290
void ResetLatch(Latch *latch)
Definition: latch.c:374
int WaitLatch(Latch *latch, int wakeEvents, long timeout, uint32 wait_event_info)
Definition: latch.c:172
bool LWLockHeldByMe(LWLock *lock)
Definition: lwlock.c:1977
bool LWLockAcquire(LWLock *lock, LWLockMode mode)
Definition: lwlock.c:1174
void LWLockRelease(LWLock *lock)
Definition: lwlock.c:1894
void LWLockReleaseAll(void)
Definition: lwlock.c:1945
@ LW_EXCLUSIVE
Definition: lwlock.h:112
void MemoryContextReset(MemoryContext context)
Definition: mcxt.c:400
void pfree(void *pointer)
Definition: mcxt.c:1594
void * palloc0(Size size)
Definition: mcxt.c:1395
MemoryContext TopMemoryContext
Definition: mcxt.c:166
void * palloc(Size size)
Definition: mcxt.c:1365
MemoryContext CurrentMemoryContext
Definition: mcxt.c:160
void ProcessLogMemoryContextInterrupt(void)
Definition: mcxt.c:1337
#define AllocSetContextCreate
Definition: memutils.h:129
#define ALLOCSET_DEFAULT_SIZES
Definition: memutils.h:160
#define AmCheckpointerProcess()
Definition: miscadmin.h:388
#define RESUME_INTERRUPTS()
Definition: miscadmin.h:135
#define START_CRIT_SECTION()
Definition: miscadmin.h:149
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:122
#define HOLD_INTERRUPTS()
Definition: miscadmin.h:133
@ B_CHECKPOINTER
Definition: miscadmin.h:362
#define END_CRIT_SECTION()
Definition: miscadmin.h:151
Oid GetUserId(void)
Definition: miscinit.c:469
BackendType MyBackendType
Definition: miscinit.c:64
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:124
int parser_errposition(ParseState *pstate, int location)
Definition: parse_node.c:106
static PgChecksumMode mode
Definition: pg_checksums.c:55
#define foreach_ptr(type, var, lst)
Definition: pg_list.h:469
static int progress
Definition: pgbench.c:262
void pgstat_before_server_shutdown(int code, Datum arg)
Definition: pgstat.c:561
void pgstat_report_checkpointer(void)
PgStat_CheckpointerStats PendingCheckpointerStats
void pgstat_report_wal(bool force)
Definition: pgstat_wal.c:46
int64 pg_time_t
Definition: pgtime.h:23
void SendPostmasterSignal(PMSignalReason reason)
Definition: pmsignal.c:165
@ PMSIGNAL_XLOG_IS_SHUTDOWN
Definition: pmsignal.h:44
#define pqsignal
Definition: port.h:531
#define GetPGProcByNumber(n)
Definition: proc.h:440
#define INVALID_PROC_NUMBER
Definition: procnumber.h:26
int ProcNumber
Definition: procnumber.h:24
void ProcessProcSignalBarrier(void)
Definition: procsignal.c:499
void procsignal_sigusr1_handler(SIGNAL_ARGS)
Definition: procsignal.c:674
tree ctl
Definition: radixtree.h:1838
void ReleaseAuxProcessResources(bool isCommit)
Definition: resowner.c:1016
Size add_size(Size s1, Size s2)
Definition: shmem.c:493
Size mul_size(Size s1, Size s2)
Definition: shmem.c:510
void * ShmemInitStruct(const char *name, Size size, bool *foundPtr)
Definition: shmem.c:387
void pg_usleep(long microsec)
Definition: signal.c:53
void smgrdestroyall(void)
Definition: smgr.c:386
void AtEOXact_SMgr(void)
Definition: smgr.c:1008
#define SpinLockInit(lock)
Definition: spin.h:57
#define SpinLockRelease(lock)
Definition: spin.h:61
#define SpinLockAcquire(lock)
Definition: spin.h:59
PROC_HDR * ProcGlobal
Definition: proc.c:78
SyncRequestType type
Definition: checkpointer.c:112
ConditionVariable done_cv
Definition: checkpointer.c:129
ConditionVariable start_cv
Definition: checkpointer.c:128
CheckpointerRequest requests[FLEXIBLE_ARRAY_MEMBER]
Definition: checkpointer.c:140
Definition: sync.h:51
Definition: hsearch.h:66
Definition: dynahash.c:222
Definition: proc.h:386
ProcNumber checkpointerProc
Definition: proc.h:425
PgStat_Counter restartpoints_requested
Definition: pgstat.h:260
PgStat_Counter num_requested
Definition: pgstat.h:257
PgStat_Counter num_performed
Definition: pgstat.h:258
PgStat_Counter restartpoints_timed
Definition: pgstat.h:259
PgStat_Counter num_timed
Definition: pgstat.h:256
PgStat_Counter restartpoints_performed
Definition: pgstat.h:261
void RememberSyncRequest(const FileTag *ftag, SyncRequestType type)
Definition: sync.c:487
SyncRequestType
Definition: sync.h:24
void SyncRepUpdateSyncStandbysDefined(void)
Definition: syncrep.c:964
static void pgstat_report_wait_end(void)
Definition: wait_event.h:85
const char * type
#define WL_TIMEOUT
Definition: waiteventset.h:37
#define WL_EXIT_ON_PM_DEATH
Definition: waiteventset.h:39
#define WL_LATCH_SET
Definition: waiteventset.h:34
#define SIGCHLD
Definition: win32_port.h:168
#define SIGHUP
Definition: win32_port.h:158
#define SIGPIPE
Definition: win32_port.h:163
#define SIGUSR1
Definition: win32_port.h:170
#define SIGALRM
Definition: win32_port.h:164
#define SIGUSR2
Definition: win32_port.h:171
int gettimeofday(struct timeval *tp, void *tzp)
void UpdateFullPageWrites(void)
Definition: xlog.c:8216
bool RecoveryInProgress(void)
Definition: xlog.c:6386
XLogRecPtr RequestXLogSwitch(bool mark_unimportant)
Definition: xlog.c:8110
bool CreateRestartPoint(int flags)
Definition: xlog.c:7639
XLogRecPtr GetInsertRecPtr(void)
Definition: xlog.c:6534
int wal_segment_size
Definition: xlog.c:144
void ShutdownXLOG(int code, Datum arg)
Definition: xlog.c:6654
int XLogArchiveTimeout
Definition: xlog.c:119
pg_time_t GetLastSegSwitchData(XLogRecPtr *lastSwitchLSN)
Definition: xlog.c:6637
XLogRecPtr GetLastImportantRecPtr(void)
Definition: xlog.c:6608
bool CreateCheckPoint(int flags)
Definition: xlog.c:6941
int CheckPointSegments
Definition: xlog.c:157
#define CHECKPOINT_FLUSH_UNLOGGED
Definition: xlog.h:143
#define CHECKPOINT_CAUSE_XLOG
Definition: xlog.h:148
#define CHECKPOINT_END_OF_RECOVERY
Definition: xlog.h:140
#define CHECKPOINT_CAUSE_TIME
Definition: xlog.h:149
#define CHECKPOINT_REQUESTED
Definition: xlog.h:146
#define CHECKPOINT_FORCE
Definition: xlog.h:142
#define CHECKPOINT_WAIT
Definition: xlog.h:145
#define CHECKPOINT_FAST
Definition: xlog.h:141
#define XLogSegmentOffset(xlogptr, wal_segsz_bytes)
Definition: xlog_internal.h:106
uint64 XLogRecPtr
Definition: xlogdefs.h:21
XLogRecPtr GetXLogReplayRecPtr(TimeLineID *replayTLI)
Definition: xlogrecovery.c:4561

AltStyle によって変換されたページ (->オリジナル) /