Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pgarch.c
4 : *
5 : * PostgreSQL WAL archiver
6 : *
7 : * All functions relating to archiver are included here
8 : *
9 : * - All functions executed by archiver process
10 : *
11 : * - archiver is forked from postmaster, and the two
12 : * processes then communicate using signals. All functions
13 : * executed by postmaster are included in this file.
14 : *
15 : * Initial author: Simon Riggs simon@2ndquadrant.com
16 : *
17 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
18 : * Portions Copyright (c) 1994, Regents of the University of California
19 : *
20 : *
21 : * IDENTIFICATION
22 : * src/backend/postmaster/pgarch.c
23 : *
24 : *-------------------------------------------------------------------------
25 : */
26 : #include "postgres.h"
27 :
28 : #include <time.h>
29 : #include <sys/stat.h>
30 : #include <unistd.h>
31 :
32 : #include "access/xlog.h"
33 : #include "access/xlog_internal.h"
34 : #include "archive/archive_module.h"
35 : #include "archive/shell_archive.h"
36 : #include "lib/binaryheap.h"
37 : #include "libpq/pqsignal.h"
38 : #include "pgstat.h"
39 : #include "postmaster/auxprocess.h"
40 : #include "postmaster/interrupt.h"
41 : #include "postmaster/pgarch.h"
42 : #include "storage/condition_variable.h"
43 : #include "storage/aio_subsys.h"
44 : #include "storage/fd.h"
45 : #include "storage/ipc.h"
46 : #include "storage/latch.h"
47 : #include "storage/pmsignal.h"
48 : #include "storage/proc.h"
49 : #include "storage/procsignal.h"
50 : #include "storage/shmem.h"
51 : #include "utils/guc.h"
52 : #include "utils/memutils.h"
53 : #include "utils/ps_status.h"
54 : #include "utils/resowner.h"
55 : #include "utils/timeout.h"
56 :
57 :
58 : /* ----------
59 : * Timer definitions.
60 : * ----------
61 : */
62 : #define PGARCH_AUTOWAKE_INTERVAL 60 /* How often to force a poll of the
63 : * archive status directory; in seconds. */
64 : #define PGARCH_RESTART_INTERVAL 10 /* How often to attempt to restart a
65 : * failed archiver; in seconds. */
66 :
67 : /*
68 : * Maximum number of retries allowed when attempting to archive a WAL
69 : * file.
70 : */
71 : #define NUM_ARCHIVE_RETRIES 3
72 :
73 : /*
74 : * Maximum number of retries allowed when attempting to remove an
75 : * orphan archive status file.
76 : */
77 : #define NUM_ORPHAN_CLEANUP_RETRIES 3
78 :
79 : /*
80 : * Maximum number of .ready files to gather per directory scan.
81 : */
82 : #define NUM_FILES_PER_DIRECTORY_SCAN 64
83 :
84 : /* Shared memory area for archiver process */
85 : typedef struct PgArchData
86 : {
87 : int pgprocno; /* proc number of archiver process */
88 :
89 : /*
90 : * Forces a directory scan in pgarch_readyXlog().
91 : */
92 : pg_atomic_uint32 force_dir_scan;
93 : } PgArchData;
94 :
95 : char *XLogArchiveLibrary = "";
96 : char *arch_module_check_errdetail_string;
97 :
98 :
99 : /* ----------
100 : * Local data
101 : * ----------
102 : */
103 : static time_t last_sigterm_time = 0;
104 : static PgArchData *PgArch = NULL;
105 : static const ArchiveModuleCallbacks *ArchiveCallbacks;
106 : static ArchiveModuleState *archive_module_state;
107 : static MemoryContext archive_context;
108 :
109 :
110 : /*
111 : * Stuff for tracking multiple files to archive from each scan of
112 : * archive_status. Minimizing the number of directory scans when there are
113 : * many files to archive can significantly improve archival rate.
114 : *
115 : * arch_heap is a max-heap that is used during the directory scan to track
116 : * the highest-priority files to archive. After the directory scan
117 : * completes, the file names are stored in ascending order of priority in
118 : * arch_files. pgarch_readyXlog() returns files from arch_files until it
119 : * is empty, at which point another directory scan must be performed.
120 : *
121 : * We only need this data in the archiver process, so make it a palloc'd
122 : * struct rather than a bunch of static arrays.
123 : */
124 : struct arch_files_state
125 : {
126 : binaryheap *arch_heap;
127 : int arch_files_size; /* number of live entries in arch_files[] */
128 : char *arch_files[NUM_FILES_PER_DIRECTORY_SCAN];
129 : /* buffers underlying heap, and later arch_files[], entries: */
130 : char arch_filenames[NUM_FILES_PER_DIRECTORY_SCAN][MAX_XFN_CHARS + 1];
131 : };
132 :
133 : static struct arch_files_state *arch_files = NULL;
134 :
135 : /*
136 : * Flags set by interrupt handlers for later service in the main loop.
137 : */
138 : static volatile sig_atomic_t ready_to_stop = false;
139 :
140 : /* ----------
141 : * Local function forward declarations
142 : * ----------
143 : */
144 : static void pgarch_waken_stop(SIGNAL_ARGS);
145 : static void pgarch_MainLoop(void);
146 : static void pgarch_ArchiverCopyLoop(void);
147 : static bool pgarch_archiveXlog(char *xlog);
148 : static bool pgarch_readyXlog(char *xlog);
149 : static void pgarch_archiveDone(char *xlog);
150 : static void pgarch_die(int code, Datum arg);
151 : static void ProcessPgArchInterrupts(void);
152 : static int ready_file_comparator(Datum a, Datum b, void *arg);
153 : static void LoadArchiveLibrary(void);
154 : static void pgarch_call_module_shutdown_cb(int code, Datum arg);
155 :
156 : /* Report shared memory space needed by PgArchShmemInit */
157 : Size
158 8106 : PgArchShmemSize(void)
159 : {
160 8106 : Size size = 0;
161 :
162 8106 : size = add_size(size, sizeof(PgArchData));
163 :
164 8106 : return size;
165 : }
166 :
167 : /* Allocate and initialize archiver-related shared memory */
168 : void
169 2100 : PgArchShmemInit(void)
170 : {
171 : bool found;
172 :
173 2100 : PgArch = (PgArchData *)
174 2100 : ShmemInitStruct("Archiver Data", PgArchShmemSize(), &found);
175 :
176 2100 : if (!found)
177 : {
178 : /* First time through, so initialize */
179 4200 : MemSet(PgArch, 0, PgArchShmemSize());
180 2100 : PgArch->pgprocno = INVALID_PROC_NUMBER;
181 2100 : pg_atomic_init_u32(&PgArch->force_dir_scan, 0);
182 : }
183 2100 : }
184 :
185 : /*
186 : * PgArchCanRestart
187 : *
188 : * Return true and archiver is allowed to restart if enough time has
189 : * passed since it was launched last to reach PGARCH_RESTART_INTERVAL.
190 : * Otherwise return false.
191 : *
192 : * This is a safety valve to protect against continuous respawn attempts if the
193 : * archiver is dying immediately at launch. Note that since we will retry to
194 : * launch the archiver from the postmaster main loop, we will get another
195 : * chance later.
196 : */
197 : bool
198 104 : PgArchCanRestart(void)
199 : {
200 : static time_t last_pgarch_start_time = 0;
201 104 : time_t curtime = time(NULL);
202 :
203 : /*
204 : * Return false and don't restart archiver if too soon since last archiver
205 : * start.
206 : */
207 104 : if ((unsigned int) (curtime - last_pgarch_start_time) <
208 : (unsigned int) PGARCH_RESTART_INTERVAL)
209 0 : return false;
210 :
211 104 : last_pgarch_start_time = curtime;
212 104 : return true;
213 : }
214 :
215 :
216 : /* Main entry point for archiver process */
217 : void
218 30 : PgArchiverMain(const void *startup_data, size_t startup_data_len)
219 : {
220 : Assert(startup_data_len == 0);
221 :
222 30 : MyBackendType = B_ARCHIVER;
223 30 : AuxiliaryProcessMainCommon();
224 :
225 : /*
226 : * Ignore all signals usually bound to some action in the postmaster,
227 : * except for SIGHUP, SIGTERM, SIGUSR1, SIGUSR2, and SIGQUIT.
228 : */
229 30 : pqsignal(SIGHUP, SignalHandlerForConfigReload);
230 30 : pqsignal(SIGINT, SIG_IGN);
231 30 : pqsignal(SIGTERM, SignalHandlerForShutdownRequest);
232 : /* SIGQUIT handler was already set up by InitPostmasterChild */
233 30 : pqsignal(SIGALRM, SIG_IGN);
234 30 : pqsignal(SIGPIPE, SIG_IGN);
235 30 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
236 30 : pqsignal(SIGUSR2, pgarch_waken_stop);
237 :
238 : /* Reset some signals that are accepted by postmaster but not here */
239 30 : pqsignal(SIGCHLD, SIG_DFL);
240 :
241 : /* Unblock signals (they were blocked when the postmaster forked us) */
242 30 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
243 :
244 : /* We shouldn't be launched unnecessarily. */
245 : Assert(XLogArchivingActive());
246 :
247 : /* Arrange to clean up at archiver exit */
248 30 : on_shmem_exit(pgarch_die, 0);
249 :
250 : /*
251 : * Advertise our proc number so that backends can use our latch to wake us
252 : * up while we're sleeping.
253 : */
254 30 : PgArch->pgprocno = MyProcNumber;
255 :
256 : /* Create workspace for pgarch_readyXlog() */
257 30 : arch_files = palloc(sizeof(struct arch_files_state));
258 30 : arch_files->arch_files_size = 0;
259 :
260 : /* Initialize our max-heap for prioritizing files to archive. */
261 30 : arch_files->arch_heap = binaryheap_allocate(NUM_FILES_PER_DIRECTORY_SCAN,
262 : ready_file_comparator, NULL);
263 :
264 : /* Initialize our memory context. */
265 30 : archive_context = AllocSetContextCreate(TopMemoryContext,
266 : "archiver",
267 : ALLOCSET_DEFAULT_SIZES);
268 :
269 : /* Load the archive_library. */
270 30 : LoadArchiveLibrary();
271 :
272 30 : pgarch_MainLoop();
273 :
274 30 : proc_exit(0);
275 : }
276 :
277 : /*
278 : * Wake up the archiver
279 : */
280 : void
281 924 : PgArchWakeup(void)
282 : {
283 924 : int arch_pgprocno = PgArch->pgprocno;
284 :
285 : /*
286 : * We don't acquire ProcArrayLock here. It's actually fine because
287 : * procLatch isn't ever freed, so we just can potentially set the wrong
288 : * process' (or no process') latch. Even in that case the archiver will
289 : * be relaunched shortly and will start archiving.
290 : */
291 924 : if (arch_pgprocno != INVALID_PROC_NUMBER)
292 888 : SetLatch(&ProcGlobal->allProcs[arch_pgprocno].procLatch);
293 924 : }
294 :
295 :
296 : /* SIGUSR2 signal handler for archiver process */
297 : static void
298 30 : pgarch_waken_stop(SIGNAL_ARGS)
299 : {
300 : /* set flag to do a final cycle and shut down afterwards */
301 30 : ready_to_stop = true;
302 30 : SetLatch(MyLatch);
303 30 : }
304 :
305 : /*
306 : * pgarch_MainLoop
307 : *
308 : * Main loop for archiver
309 : */
310 : static void
311 614 : pgarch_MainLoop(void)
312 : {
313 : bool time_to_stop;
314 :
315 : /*
316 : * There shouldn't be anything for the archiver to do except to wait for a
317 : * signal ... however, the archiver exists to protect our data, so it
318 : * wakes up occasionally to allow itself to be proactive.
319 : */
320 : do
321 : {
322 614 : ResetLatch(MyLatch);
323 :
324 : /* When we get SIGUSR2, we do one more archive cycle, then exit */
325 614 : time_to_stop = ready_to_stop;
326 :
327 : /* Check for barrier events and config update */
328 614 : ProcessPgArchInterrupts();
329 :
330 : /*
331 : * If we've gotten SIGTERM, we normally just sit and do nothing until
332 : * SIGUSR2 arrives. However, that means a random SIGTERM would
333 : * disable archiving indefinitely, which doesn't seem like a good
334 : * idea. If more than 60 seconds pass since SIGTERM, exit anyway, so
335 : * that the postmaster can start a new archiver if needed.
336 : */
337 614 : if (ShutdownRequestPending)
338 : {
339 0 : time_t curtime = time(NULL);
340 :
341 0 : if (last_sigterm_time == 0)
342 0 : last_sigterm_time = curtime;
343 0 : else if ((unsigned int) (curtime - last_sigterm_time) >=
344 : (unsigned int) 60)
345 0 : break;
346 : }
347 :
348 : /* Do what we're here for */
349 614 : pgarch_ArchiverCopyLoop();
350 :
351 : /*
352 : * Sleep until a signal is received, or until a poll is forced by
353 : * PGARCH_AUTOWAKE_INTERVAL, or until postmaster dies.
354 : */
355 614 : if (!time_to_stop) /* Don't wait during last iteration */
356 : {
357 : int rc;
358 :
359 584 : rc = WaitLatch(MyLatch,
360 : WL_LATCH_SET | WL_TIMEOUT | WL_POSTMASTER_DEATH,
361 : PGARCH_AUTOWAKE_INTERVAL * 1000L,
362 : WAIT_EVENT_ARCHIVER_MAIN);
363 584 : if (rc & WL_POSTMASTER_DEATH)
364 0 : time_to_stop = true;
365 : }
366 :
367 : /*
368 : * The archiver quits either when the postmaster dies (not expected)
369 : * or after completing one more archiving cycle after receiving
370 : * SIGUSR2.
371 : */
372 614 : } while (!time_to_stop);
373 30 : }
374 :
375 : /*
376 : * pgarch_ArchiverCopyLoop
377 : *
378 : * Archives all outstanding xlogs then returns
379 : */
380 : static void
381 614 : pgarch_ArchiverCopyLoop(void)
382 : {
383 : char xlog[MAX_XFN_CHARS + 1];
384 :
385 : /* force directory scan in the first call to pgarch_readyXlog() */
386 614 : arch_files->arch_files_size = 0;
387 :
388 : /*
389 : * loop through all xlogs with archive_status of .ready and archive
390 : * them...mostly we expect this to be a single file, though it is possible
391 : * some backend will add files onto the list of those that need archiving
392 : * while we are still copying earlier archives
393 : */
394 1330 : while (pgarch_readyXlog(xlog))
395 : {
396 724 : int failures = 0;
397 724 : int failures_orphan = 0;
398 :
399 : for (;;)
400 10 : {
401 : struct stat stat_buf;
402 : char pathname[MAXPGPATH];
403 :
404 : /*
405 : * Do not initiate any more archive commands after receiving
406 : * SIGTERM, nor after the postmaster has died unexpectedly. The
407 : * first condition is to try to keep from having init SIGKILL the
408 : * command, and the second is to avoid conflicts with another
409 : * archiver spawned by a newer postmaster.
410 : */
411 734 : if (ShutdownRequestPending || !PostmasterIsAlive())
412 8 : return;
413 :
414 : /*
415 : * Check for barrier events and config update. This is so that
416 : * we'll adopt a new setting for archive_command as soon as
417 : * possible, even if there is a backlog of files to be archived.
418 : */
419 734 : ProcessPgArchInterrupts();
420 :
421 : /* Reset variables that might be set by the callback */
422 734 : arch_module_check_errdetail_string = NULL;
423 :
424 : /* can't do anything if not configured ... */
425 734 : if (ArchiveCallbacks->check_configured_cb != NULL &&
426 734 : !ArchiveCallbacks->check_configured_cb(archive_module_state))
427 : {
428 4 : ereport(WARNING,
429 : (errmsg("\"archive_mode\" enabled, yet archiving is not configured"),
430 : arch_module_check_errdetail_string ?
431 : errdetail_internal("%s", arch_module_check_errdetail_string) : 0));
432 4 : return;
433 : }
434 :
435 : /*
436 : * Since archive status files are not removed in a durable manner,
437 : * a system crash could leave behind .ready files for WAL segments
438 : * that have already been recycled or removed. In this case,
439 : * simply remove the orphan status file and move on. unlink() is
440 : * used here as even on subsequent crashes the same orphan files
441 : * would get removed, so there is no need to worry about
442 : * durability.
443 : */
444 730 : snprintf(pathname, MAXPGPATH, XLOGDIR "/%s", xlog);
445 730 : if (stat(pathname, &stat_buf) != 0 && errno == ENOENT)
446 : {
447 : char xlogready[MAXPGPATH];
448 :
449 0 : StatusFilePath(xlogready, xlog, ".ready");
450 0 : if (unlink(xlogready) == 0)
451 : {
452 0 : ereport(WARNING,
453 : (errmsg("removed orphan archive status file \"%s\"",
454 : xlogready)));
455 :
456 : /* leave loop and move to the next status file */
457 0 : break;
458 : }
459 :
460 0 : if (++failures_orphan >= NUM_ORPHAN_CLEANUP_RETRIES)
461 : {
462 0 : ereport(WARNING,
463 : (errmsg("removal of orphan archive status file \"%s\" failed too many times, will try again later",
464 : xlogready)));
465 :
466 : /* give up cleanup of orphan status files */
467 0 : return;
468 : }
469 :
470 : /* wait a bit before retrying */
471 0 : pg_usleep(1000000L);
472 0 : continue;
473 : }
474 :
475 730 : if (pgarch_archiveXlog(xlog))
476 : {
477 : /* successful */
478 716 : pgarch_archiveDone(xlog);
479 :
480 : /*
481 : * Tell the cumulative stats system about the WAL file that we
482 : * successfully archived
483 : */
484 716 : pgstat_report_archiver(xlog, false);
485 :
486 716 : break; /* out of inner retry loop */
487 : }
488 : else
489 : {
490 : /*
491 : * Tell the cumulative stats system about the WAL file that we
492 : * failed to archive
493 : */
494 14 : pgstat_report_archiver(xlog, true);
495 :
496 14 : if (++failures >= NUM_ARCHIVE_RETRIES)
497 : {
498 4 : ereport(WARNING,
499 : (errmsg("archiving write-ahead log file \"%s\" failed too many times, will try again later",
500 : xlog)));
501 4 : return; /* give up archiving for now */
502 : }
503 10 : pg_usleep(1000000L); /* wait a bit before retrying */
504 : }
505 : }
506 : }
507 : }
508 :
509 : /*
510 : * pgarch_archiveXlog
511 : *
512 : * Invokes archive_file_cb to copy one archive file to wherever it should go
513 : *
514 : * Returns true if successful
515 : */
516 : static bool
517 730 : pgarch_archiveXlog(char *xlog)
518 : {
519 : sigjmp_buf local_sigjmp_buf;
520 : MemoryContext oldcontext;
521 : char pathname[MAXPGPATH];
522 : char activitymsg[MAXFNAMELEN + 16];
523 : bool ret;
524 :
525 730 : snprintf(pathname, MAXPGPATH, XLOGDIR "/%s", xlog);
526 :
527 : /* Report archive activity in PS display */
528 730 : snprintf(activitymsg, sizeof(activitymsg), "archiving %s", xlog);
529 730 : set_ps_display(activitymsg);
530 :
531 730 : oldcontext = MemoryContextSwitchTo(archive_context);
532 :
533 : /*
534 : * Since the archiver operates at the bottom of the exception stack,
535 : * ERRORs turn into FATALs and cause the archiver process to restart.
536 : * However, using ereport(ERROR, ...) when there are problems is easy to
537 : * code and maintain. Therefore, we create our own exception handler to
538 : * catch ERRORs and return false instead of restarting the archiver
539 : * whenever there is a failure.
540 : *
541 : * We assume ERRORs from the archiving callback are the most common
542 : * exceptions experienced by the archiver, so we opt to handle exceptions
543 : * here instead of PgArchiverMain() to avoid reinitializing the archiver
544 : * too frequently. We could instead add a sigsetjmp() block to
545 : * PgArchiverMain() and use PG_TRY/PG_CATCH here, but the extra code to
546 : * avoid the odd archiver restart doesn't seem worth it.
547 : */
548 730 : if (sigsetjmp(local_sigjmp_buf, 1) != 0)
549 : {
550 : /* Since not using PG_TRY, must reset error stack by hand */
551 0 : error_context_stack = NULL;
552 :
553 : /* Prevent interrupts while cleaning up */
554 0 : HOLD_INTERRUPTS();
555 :
556 : /* Report the error to the server log. */
557 0 : EmitErrorReport();
558 :
559 : /*
560 : * Try to clean up anything the archive module left behind. We try to
561 : * cover anything that an archive module could conceivably have left
562 : * behind, but it is of course possible that modules could be doing
563 : * unexpected things that require additional cleanup. Module authors
564 : * should be sure to do any extra required cleanup in a PG_CATCH block
565 : * within the archiving callback, and they are encouraged to notify
566 : * the pgsql-hackers mailing list so that we can add it here.
567 : */
568 0 : disable_all_timeouts(false);
569 0 : LWLockReleaseAll();
570 0 : ConditionVariableCancelSleep();
571 0 : pgstat_report_wait_end();
572 0 : pgaio_error_cleanup();
573 0 : ReleaseAuxProcessResources(false);
574 0 : AtEOXact_Files(false);
575 0 : AtEOXact_HashTables(false);
576 :
577 : /*
578 : * Return to the original memory context and clear ErrorContext for
579 : * next time.
580 : */
581 0 : MemoryContextSwitchTo(oldcontext);
582 0 : FlushErrorState();
583 :
584 : /* Flush any leaked data */
585 0 : MemoryContextReset(archive_context);
586 :
587 : /* Remove our exception handler */
588 0 : PG_exception_stack = NULL;
589 :
590 : /* Now we can allow interrupts again */
591 0 : RESUME_INTERRUPTS();
592 :
593 : /* Report failure so that the archiver retries this file */
594 0 : ret = false;
595 : }
596 : else
597 : {
598 : /* Enable our exception handler */
599 730 : PG_exception_stack = &local_sigjmp_buf;
600 :
601 : /* Archive the file! */
602 730 : ret = ArchiveCallbacks->archive_file_cb(archive_module_state,
603 : xlog, pathname);
604 :
605 : /* Remove our exception handler */
606 730 : PG_exception_stack = NULL;
607 :
608 : /* Reset our memory context and switch back to the original one */
609 730 : MemoryContextSwitchTo(oldcontext);
610 730 : MemoryContextReset(archive_context);
611 : }
612 :
613 730 : if (ret)
614 716 : snprintf(activitymsg, sizeof(activitymsg), "last was %s", xlog);
615 : else
616 14 : snprintf(activitymsg, sizeof(activitymsg), "failed on %s", xlog);
617 730 : set_ps_display(activitymsg);
618 :
619 730 : return ret;
620 : }
621 :
622 : /*
623 : * pgarch_readyXlog
624 : *
625 : * Return name of the oldest xlog file that has not yet been archived.
626 : * No notification is set that file archiving is now in progress, so
627 : * this would need to be extended if multiple concurrent archival
628 : * tasks were created. If a failure occurs, we will completely
629 : * re-copy the file at the next available opportunity.
630 : *
631 : * It is important that we return the oldest, so that we archive xlogs
632 : * in order that they were written, for two reasons:
633 : * 1) to maintain the sequential chain of xlogs required for recovery
634 : * 2) because the oldest ones will sooner become candidates for
635 : * recycling at time of checkpoint
636 : *
637 : * NOTE: the "oldest" comparison will consider any .history file to be older
638 : * than any other file except another .history file. Segments on a timeline
639 : * with a smaller ID will be older than all segments on a timeline with a
640 : * larger ID; the net result being that past timelines are given higher
641 : * priority for archiving. This seems okay, or at least not obviously worth
642 : * changing.
643 : */
644 : static bool
645 1330 : pgarch_readyXlog(char *xlog)
646 : {
647 : char XLogArchiveStatusDir[MAXPGPATH];
648 : DIR *rldir;
649 : struct dirent *rlde;
650 :
651 : /*
652 : * If a directory scan was requested, clear the stored file names and
653 : * proceed.
654 : */
655 1330 : if (pg_atomic_exchange_u32(&PgArch->force_dir_scan, 0) == 1)
656 6 : arch_files->arch_files_size = 0;
657 :
658 : /*
659 : * If we still have stored file names from the previous directory scan,
660 : * try to return one of those. We check to make sure the status file is
661 : * still present, as the archive_command for a previous file may have
662 : * already marked it done.
663 : */
664 1330 : while (arch_files->arch_files_size > 0)
665 : {
666 : struct stat st;
667 : char status_file[MAXPGPATH];
668 : char *arch_file;
669 :
670 152 : arch_files->arch_files_size--;
671 152 : arch_file = arch_files->arch_files[arch_files->arch_files_size];
672 152 : StatusFilePath(status_file, arch_file, ".ready");
673 :
674 152 : if (stat(status_file, &st) == 0)
675 : {
676 152 : strcpy(xlog, arch_file);
677 152 : return true;
678 : }
679 0 : else if (errno != ENOENT)
680 0 : ereport(ERROR,
681 : (errcode_for_file_access(),
682 : errmsg("could not stat file \"%s\": %m", status_file)));
683 : }
684 :
685 : /* arch_heap is probably empty, but let's make sure */
686 1178 : binaryheap_reset(arch_files->arch_heap);
687 :
688 : /*
689 : * Open the archive status directory and read through the list of files
690 : * with the .ready suffix, looking for the earliest files.
691 : */
692 1178 : snprintf(XLogArchiveStatusDir, MAXPGPATH, XLOGDIR "/archive_status");
693 1178 : rldir = AllocateDir(XLogArchiveStatusDir);
694 :
695 6730 : while ((rlde = ReadDir(rldir, XLogArchiveStatusDir)) != NULL)
696 : {
697 5552 : int basenamelen = (int) strlen(rlde->d_name) - 6;
698 : char basename[MAX_XFN_CHARS + 1];
699 : char *arch_file;
700 :
701 : /* Ignore entries with unexpected number of characters */
702 5552 : if (basenamelen < MIN_XFN_CHARS ||
703 : basenamelen > MAX_XFN_CHARS)
704 4826 : continue;
705 :
706 : /* Ignore entries with unexpected characters */
707 3148 : if (strspn(rlde->d_name, VALID_XFN_CHARS) < basenamelen)
708 0 : continue;
709 :
710 : /* Ignore anything not suffixed with .ready */
711 3148 : if (strcmp(rlde->d_name + basenamelen, ".ready") != 0)
712 2422 : continue;
713 :
714 : /* Truncate off the .ready */
715 726 : memcpy(basename, rlde->d_name, basenamelen);
716 726 : basename[basenamelen] = '\0';
717 :
718 : /*
719 : * Store the file in our max-heap if it has a high enough priority.
720 : */
721 726 : if (arch_files->arch_heap->bh_size < NUM_FILES_PER_DIRECTORY_SCAN)
722 : {
723 : /* If the heap isn't full yet, quickly add it. */
724 726 : arch_file = arch_files->arch_filenames[arch_files->arch_heap->bh_size];
725 726 : strcpy(arch_file, basename);
726 726 : binaryheap_add_unordered(arch_files->arch_heap, CStringGetDatum(arch_file));
727 :
728 : /* If we just filled the heap, make it a valid one. */
729 726 : if (arch_files->arch_heap->bh_size == NUM_FILES_PER_DIRECTORY_SCAN)
730 0 : binaryheap_build(arch_files->arch_heap);
731 : }
732 0 : else if (ready_file_comparator(binaryheap_first(arch_files->arch_heap),
733 : CStringGetDatum(basename), NULL) > 0)
734 : {
735 : /*
736 : * Remove the lowest priority file and add the current one to the
737 : * heap.
738 : */
739 0 : arch_file = DatumGetCString(binaryheap_remove_first(arch_files->arch_heap));
740 0 : strcpy(arch_file, basename);
741 0 : binaryheap_add(arch_files->arch_heap, CStringGetDatum(arch_file));
742 : }
743 : }
744 1178 : FreeDir(rldir);
745 :
746 : /* If no files were found, simply return. */
747 1178 : if (arch_files->arch_heap->bh_size == 0)
748 606 : return false;
749 :
750 : /*
751 : * If we didn't fill the heap, we didn't make it a valid one. Do that
752 : * now.
753 : */
754 572 : if (arch_files->arch_heap->bh_size < NUM_FILES_PER_DIRECTORY_SCAN)
755 572 : binaryheap_build(arch_files->arch_heap);
756 :
757 : /*
758 : * Fill arch_files array with the files to archive in ascending order of
759 : * priority.
760 : */
761 572 : arch_files->arch_files_size = arch_files->arch_heap->bh_size;
762 1298 : for (int i = 0; i < arch_files->arch_files_size; i++)
763 726 : arch_files->arch_files[i] = DatumGetCString(binaryheap_remove_first(arch_files->arch_heap));
764 :
765 : /* Return the highest priority file. */
766 572 : arch_files->arch_files_size--;
767 572 : strcpy(xlog, arch_files->arch_files[arch_files->arch_files_size]);
768 :
769 572 : return true;
770 : }
771 :
772 : /*
773 : * ready_file_comparator
774 : *
775 : * Compares the archival priority of the given files to archive. If "a"
776 : * has a higher priority than "b", a negative value will be returned. If
777 : * "b" has a higher priority than "a", a positive value will be returned.
778 : * If "a" and "b" have equivalent values, 0 will be returned.
779 : */
780 : static int
781 912 : ready_file_comparator(Datum a, Datum b, void *arg)
782 : {
783 912 : char *a_str = DatumGetCString(a);
784 912 : char *b_str = DatumGetCString(b);
785 912 : bool a_history = IsTLHistoryFileName(a_str);
786 912 : bool b_history = IsTLHistoryFileName(b_str);
787 :
788 : /* Timeline history files always have the highest priority. */
789 912 : if (a_history != b_history)
790 0 : return a_history ? -1 : 1;
791 :
792 : /* Priority is given to older files. */
793 912 : return strcmp(a_str, b_str);
794 : }
795 :
796 : /*
797 : * PgArchForceDirScan
798 : *
799 : * When called, the next call to pgarch_readyXlog() will perform a
800 : * directory scan. This is useful for ensuring that important files such
801 : * as timeline history files are archived as quickly as possible.
802 : */
803 : void
804 28 : PgArchForceDirScan(void)
805 : {
806 28 : pg_atomic_write_membarrier_u32(&PgArch->force_dir_scan, 1);
807 28 : }
808 :
809 : /*
810 : * pgarch_archiveDone
811 : *
812 : * Emit notification that an xlog file has been successfully archived.
813 : * We do this by renaming the status file from NNN.ready to NNN.done.
814 : * Eventually, a checkpoint process will notice this and delete both the
815 : * NNN.done file and the xlog file itself.
816 : */
817 : static void
818 716 : pgarch_archiveDone(char *xlog)
819 : {
820 : char rlogready[MAXPGPATH];
821 : char rlogdone[MAXPGPATH];
822 :
823 716 : StatusFilePath(rlogready, xlog, ".ready");
824 716 : StatusFilePath(rlogdone, xlog, ".done");
825 :
826 : /*
827 : * To avoid extra overhead, we don't durably rename the .ready file to
828 : * .done. Archive commands and libraries must gracefully handle attempts
829 : * to re-archive files (e.g., if the server crashes just before this
830 : * function is called), so it should be okay if the .ready file reappears
831 : * after a crash.
832 : */
833 716 : if (rename(rlogready, rlogdone) < 0)
834 0 : ereport(WARNING,
835 : (errcode_for_file_access(),
836 : errmsg("could not rename file \"%s\" to \"%s\": %m",
837 : rlogready, rlogdone)));
838 716 : }
839 :
840 :
841 : /*
842 : * pgarch_die
843 : *
844 : * Exit-time cleanup handler
845 : */
846 : static void
847 30 : pgarch_die(int code, Datum arg)
848 : {
849 30 : PgArch->pgprocno = INVALID_PROC_NUMBER;
850 30 : }
851 :
852 : /*
853 : * Interrupt handler for WAL archiver process.
854 : *
855 : * This is called in the loops pgarch_MainLoop and pgarch_ArchiverCopyLoop.
856 : * It checks for barrier events, config update and request for logging of
857 : * memory contexts, but not shutdown request because how to handle
858 : * shutdown request is different between those loops.
859 : */
860 : static void
861 1348 : ProcessPgArchInterrupts(void)
862 : {
863 1348 : if (ProcSignalBarrierPending)
864 2 : ProcessProcSignalBarrier();
865 :
866 : /* Perform logging of memory contexts of this process */
867 1348 : if (LogMemoryContextPending)
868 0 : ProcessLogMemoryContextInterrupt();
869 :
870 : /* Publish memory contexts of this process */
871 1348 : if (PublishMemoryContextPending)
872 0 : ProcessGetMemoryContextInterrupt();
873 :
874 1348 : if (ConfigReloadPending)
875 : {
876 6 : char *archiveLib = pstrdup(XLogArchiveLibrary);
877 : bool archiveLibChanged;
878 :
879 6 : ConfigReloadPending = false;
880 6 : ProcessConfigFile(PGC_SIGHUP);
881 :
882 6 : if (XLogArchiveLibrary[0] != '\0' && XLogArchiveCommand[0] != '\0')
883 0 : ereport(ERROR,
884 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
885 : errmsg("both \"archive_command\" and \"archive_library\" set"),
886 : errdetail("Only one of \"archive_command\", \"archive_library\" may be set.")));
887 :
888 6 : archiveLibChanged = strcmp(XLogArchiveLibrary, archiveLib) != 0;
889 6 : pfree(archiveLib);
890 :
891 6 : if (archiveLibChanged)
892 : {
893 : /*
894 : * Ideally, we would simply unload the previous archive module and
895 : * load the new one, but there is presently no mechanism for
896 : * unloading a library (see the comment above
897 : * internal_load_library()). To deal with this, we simply restart
898 : * the archiver. The new archive module will be loaded when the
899 : * new archiver process starts up. Note that this triggers the
900 : * module's shutdown callback, if defined.
901 : */
902 0 : ereport(LOG,
903 : (errmsg("restarting archiver process because value of "
904 : "\"archive_library\" was changed")));
905 :
906 0 : proc_exit(0);
907 : }
908 : }
909 1348 : }
910 :
911 : /*
912 : * LoadArchiveLibrary
913 : *
914 : * Loads the archiving callbacks into our local ArchiveCallbacks.
915 : */
916 : static void
917 30 : LoadArchiveLibrary(void)
918 : {
919 : ArchiveModuleInit archive_init;
920 :
921 30 : if (XLogArchiveLibrary[0] != '\0' && XLogArchiveCommand[0] != '\0')
922 0 : ereport(ERROR,
923 : (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
924 : errmsg("both \"archive_command\" and \"archive_library\" set"),
925 : errdetail("Only one of \"archive_command\", \"archive_library\" may be set.")));
926 :
927 : /*
928 : * If shell archiving is enabled, use our special initialization function.
929 : * Otherwise, load the library and call its _PG_archive_module_init().
930 : */
931 30 : if (XLogArchiveLibrary[0] == '\0')
932 28 : archive_init = shell_archive_init;
933 : else
934 2 : archive_init = (ArchiveModuleInit)
935 2 : load_external_function(XLogArchiveLibrary,
936 : "_PG_archive_module_init", false, NULL);
937 :
938 30 : if (archive_init == NULL)
939 0 : ereport(ERROR,
940 : (errmsg("archive modules have to define the symbol %s", "_PG_archive_module_init")));
941 :
942 30 : ArchiveCallbacks = (*archive_init) ();
943 :
944 30 : if (ArchiveCallbacks->archive_file_cb == NULL)
945 0 : ereport(ERROR,
946 : (errmsg("archive modules must register an archive callback")));
947 :
948 30 : archive_module_state = (ArchiveModuleState *) palloc0(sizeof(ArchiveModuleState));
949 30 : if (ArchiveCallbacks->startup_cb != NULL)
950 0 : ArchiveCallbacks->startup_cb(archive_module_state);
951 :
952 30 : before_shmem_exit(pgarch_call_module_shutdown_cb, 0);
953 30 : }
954 :
955 : /*
956 : * Call the shutdown callback of the loaded archive module, if defined.
957 : */
958 : static void
959 30 : pgarch_call_module_shutdown_cb(int code, Datum arg)
960 : {
961 30 : if (ArchiveCallbacks->shutdown_cb != NULL)
962 28 : ArchiveCallbacks->shutdown_cb(archive_module_state);
963 30 : }
|