Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * walreceiver.c
4 : *
5 : * The WAL receiver process (walreceiver) is new as of Postgres 9.0. It
6 : * is the process in the standby server that takes charge of receiving
7 : * XLOG records from a primary server during streaming replication.
8 : *
9 : * When the startup process determines that it's time to start streaming,
10 : * it instructs postmaster to start walreceiver. Walreceiver first connects
11 : * to the primary server (it will be served by a walsender process
12 : * in the primary server), and then keeps receiving XLOG records and
13 : * writing them to the disk as long as the connection is alive. As XLOG
14 : * records are received and flushed to disk, it updates the
15 : * WalRcv->flushedUpto variable in shared memory, to inform the startup
16 : * process of how far it can proceed with XLOG replay.
17 : *
18 : * A WAL receiver cannot directly load GUC parameters used when establishing
19 : * its connection to the primary. Instead it relies on parameter values
20 : * that are passed down by the startup process when streaming is requested.
21 : * This applies, for example, to the replication slot and the connection
22 : * string to be used for the connection with the primary.
23 : *
24 : * If the primary server ends streaming, but doesn't disconnect, walreceiver
25 : * goes into "waiting" mode, and waits for the startup process to give new
26 : * instructions. The startup process will treat that the same as
27 : * disconnection, and will rescan the archive/pg_wal directory. But when the
28 : * startup process wants to try streaming replication again, it will just
29 : * nudge the existing walreceiver process that's waiting, instead of launching
30 : * a new one.
31 : *
32 : * Normal termination is by SIGTERM, which instructs the walreceiver to
33 : * ereport(FATAL). Emergency termination is by SIGQUIT; like any postmaster
34 : * child process, the walreceiver will simply abort and exit on SIGQUIT. A
35 : * close of the connection and a FATAL error are treated not as a crash but as
36 : * normal operation.
37 : *
38 : * This file contains the server-facing parts of walreceiver. The libpq-
39 : * specific parts are in the libpqwalreceiver module. It's loaded
40 : * dynamically to avoid linking the server with libpq.
41 : *
42 : * Portions Copyright (c) 2010-2026, PostgreSQL Global Development Group
43 : *
44 : *
45 : * IDENTIFICATION
46 : * src/backend/replication/walreceiver.c
47 : *
48 : *-------------------------------------------------------------------------
49 : */
50 : #include "postgres.h"
51 :
52 : #include <unistd.h>
53 :
54 : #include "access/htup_details.h"
55 : #include "access/timeline.h"
56 : #include "access/transam.h"
57 : #include "access/xlog_internal.h"
58 : #include "access/xlogarchive.h"
59 : #include "access/xlogrecovery.h"
60 : #include "access/xlogwait.h"
61 : #include "catalog/pg_authid.h"
62 : #include "funcapi.h"
63 : #include "libpq/pqformat.h"
64 : #include "libpq/pqsignal.h"
65 : #include "miscadmin.h"
66 : #include "pgstat.h"
67 : #include "postmaster/auxprocess.h"
68 : #include "postmaster/interrupt.h"
69 : #include "replication/walreceiver.h"
70 : #include "replication/walsender.h"
71 : #include "storage/ipc.h"
72 : #include "storage/proc.h"
73 : #include "storage/procarray.h"
74 : #include "storage/procsignal.h"
75 : #include "tcop/tcopprot.h"
76 : #include "utils/acl.h"
77 : #include "utils/builtins.h"
78 : #include "utils/guc.h"
79 : #include "utils/pg_lsn.h"
80 : #include "utils/ps_status.h"
81 : #include "utils/timestamp.h"
82 : #include "utils/wait_event.h"
83 :
84 :
85 : /*
86 : * GUC variables. (Other variables that affect walreceiver are in xlog.c
87 : * because they're passed down from the startup process, for better
88 : * synchronization.)
89 : */
90 : int wal_receiver_status_interval;
91 : int wal_receiver_timeout;
92 : bool hot_standby_feedback;
93 :
94 : /* libpqwalreceiver connection */
95 : static WalReceiverConn *wrconn = NULL;
96 : WalReceiverFunctionsType *WalReceiverFunctions = NULL;
97 :
98 : /*
99 : * These variables are used similarly to openLogFile/SegNo,
100 : * but for walreceiver to write the XLOG. recvFileTLI is the TimeLineID
101 : * corresponding the filename of recvFile.
102 : */
103 : static int recvFile = -1;
104 : static TimeLineID recvFileTLI = 0;
105 : static XLogSegNo recvSegNo = 0;
106 :
107 : /*
108 : * LogstreamResult indicates the byte positions that we have already
109 : * written/fsynced.
110 : */
111 : static struct
112 : {
113 : XLogRecPtr Write; /* last byte + 1 written out in the standby */
114 : XLogRecPtr Flush; /* last byte + 1 flushed in the standby */
115 : } LogstreamResult;
116 :
117 : /*
118 : * Reasons to wake up and perform periodic tasks.
119 : */
120 : typedef enum WalRcvWakeupReason
121 : {
122 : WALRCV_WAKEUP_TERMINATE,
123 : WALRCV_WAKEUP_PING,
124 : WALRCV_WAKEUP_REPLY,
125 : WALRCV_WAKEUP_HSFEEDBACK,
126 : #define NUM_WALRCV_WAKEUPS (WALRCV_WAKEUP_HSFEEDBACK + 1)
127 : } WalRcvWakeupReason;
128 :
129 : /*
130 : * Wake up times for periodic tasks.
131 : */
132 : static TimestampTz wakeup[NUM_WALRCV_WAKEUPS];
133 :
134 : static StringInfoData reply_message;
135 :
136 : /* Prototypes for private functions */
137 : static void WalRcvFetchTimeLineHistoryFiles(TimeLineID first, TimeLineID last);
138 : static void WalRcvWaitForStartPosition(XLogRecPtr *startpoint, TimeLineID *startpointTLI);
139 : static void WalRcvDie(int code, Datum arg);
140 : static void XLogWalRcvProcessMsg(unsigned char type, char *buf, Size len,
141 : TimeLineID tli);
142 : static void XLogWalRcvWrite(char *buf, Size nbytes, XLogRecPtr recptr,
143 : TimeLineID tli);
144 : static void XLogWalRcvFlush(bool dying, TimeLineID tli);
145 : static void XLogWalRcvClose(XLogRecPtr recptr, TimeLineID tli);
146 : static void XLogWalRcvSendReply(bool force, bool requestReply, bool checkApply);
147 : static void XLogWalRcvSendHSFeedback(bool immed);
148 : static void ProcessWalSndrMessage(XLogRecPtr walEnd, TimestampTz sendTime);
149 : static void WalRcvComputeNextWakeup(WalRcvWakeupReason reason, TimestampTz now);
150 :
151 :
152 : /* Main entry point for walreceiver process */
153 : void
154 268 : WalReceiverMain(const void *startup_data, size_t startup_data_len)
155 : {
156 : char conninfo[MAXCONNINFO];
157 : char *tmp_conninfo;
158 : char slotname[NAMEDATALEN];
159 : bool is_temp_slot;
160 : XLogRecPtr startpoint;
161 : TimeLineID startpointTLI;
162 : TimeLineID primaryTLI;
163 : bool first_stream;
164 : WalRcvData *walrcv;
165 : TimestampTz now;
166 : char *err;
167 268 : char *sender_host = NULL;
168 268 : int sender_port = 0;
169 : char *appname;
170 :
171 : Assert(startup_data_len == 0);
172 :
173 268 : AuxiliaryProcessMainCommon();
174 :
175 : /*
176 : * WalRcv should be set up already (if we are a backend, we inherit this
177 : * by fork() or EXEC_BACKEND mechanism from the postmaster).
178 : */
179 268 : walrcv = WalRcv;
180 : Assert(walrcv != NULL);
181 :
182 : /*
183 : * Mark walreceiver as running in shared memory.
184 : *
185 : * Do this as early as possible, so that if we fail later on, we'll set
186 : * state to STOPPED. If we die before this, the startup process will keep
187 : * waiting for us to start up, until it times out.
188 : */
189 268 : SpinLockAcquire(&walrcv->mutex);
190 : Assert(walrcv->pid == 0);
191 268 : switch (walrcv->walRcvState)
192 : {
193 0 : case WALRCV_STOPPING:
194 : /* If we've already been requested to stop, don't start up. */
195 0 : walrcv->walRcvState = WALRCV_STOPPED;
196 : pg_fallthrough;
197 :
198 3 : case WALRCV_STOPPED:
199 3 : SpinLockRelease(&walrcv->mutex);
200 3 : ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
201 3 : proc_exit(1);
202 : break;
203 :
204 265 : case WALRCV_STARTING:
205 : /* The usual case */
206 265 : break;
207 :
208 0 : case WALRCV_CONNECTING:
209 : case WALRCV_WAITING:
210 : case WALRCV_STREAMING:
211 : case WALRCV_RESTARTING:
212 : default:
213 : /* Shouldn't happen */
214 0 : SpinLockRelease(&walrcv->mutex);
215 0 : elog(PANIC, "walreceiver still running according to shared memory state");
216 : }
217 : /* Advertise our PID so that the startup process can kill us */
218 265 : walrcv->pid = MyProcPid;
219 265 : walrcv->walRcvState = WALRCV_CONNECTING;
220 :
221 : /* Fetch information required to start streaming */
222 265 : walrcv->ready_to_display = false;
223 265 : strlcpy(conninfo, walrcv->conninfo, MAXCONNINFO);
224 265 : strlcpy(slotname, walrcv->slotname, NAMEDATALEN);
225 265 : is_temp_slot = walrcv->is_temp_slot;
226 265 : startpoint = walrcv->receiveStart;
227 265 : startpointTLI = walrcv->receiveStartTLI;
228 :
229 : /*
230 : * At most one of is_temp_slot and slotname can be set; otherwise,
231 : * RequestXLogStreaming messed up.
232 : */
233 : Assert(!is_temp_slot || (slotname[0] == '\0'));
234 :
235 : /* Initialise to a sanish value */
236 265 : now = GetCurrentTimestamp();
237 265 : walrcv->lastMsgSendTime =
238 265 : walrcv->lastMsgReceiptTime = walrcv->latestWalEndTime = now;
239 :
240 : /* Report our proc number so that others can wake us up */
241 265 : walrcv->procno = MyProcNumber;
242 :
243 265 : SpinLockRelease(&walrcv->mutex);
244 :
245 : /* Arrange to clean up at walreceiver exit */
246 265 : on_shmem_exit(WalRcvDie, PointerGetDatum(&startpointTLI));
247 :
248 : /* Properly accept or ignore signals the postmaster might send us */
249 265 : pqsignal(SIGHUP, SignalHandlerForConfigReload); /* set flag to read config
250 : * file */
251 265 : pqsignal(SIGINT, PG_SIG_IGN);
252 265 : pqsignal(SIGTERM, die); /* request shutdown */
253 : /* SIGQUIT handler was already set up by InitPostmasterChild */
254 265 : pqsignal(SIGALRM, PG_SIG_IGN);
255 265 : pqsignal(SIGPIPE, PG_SIG_IGN);
256 265 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
257 265 : pqsignal(SIGUSR2, PG_SIG_IGN);
258 :
259 : /* Reset some signals that are accepted by postmaster but not here */
260 265 : pqsignal(SIGCHLD, PG_SIG_DFL);
261 :
262 : /* Load the libpq-specific functions */
263 265 : load_file("libpqwalreceiver", false);
264 265 : if (WalReceiverFunctions == NULL)
265 0 : elog(ERROR, "libpqwalreceiver didn't initialize correctly");
266 :
267 : /* Unblock signals (they were blocked when the postmaster forked us) */
268 265 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
269 :
270 : /* Establish the connection to the primary for XLOG streaming */
271 265 : appname = cluster_name[0] ? cluster_name : "walreceiver";
272 265 : wrconn = walrcv_connect(conninfo, true, false, false, appname, &err);
273 265 : if (!wrconn)
274 112 : ereport(ERROR,
275 : (errcode(ERRCODE_CONNECTION_FAILURE),
276 : errmsg("streaming replication receiver \"%s\" could not connect to the primary server: %s",
277 : appname, err)));
278 :
279 : /*
280 : * Save user-visible connection string. This clobbers the original
281 : * conninfo, for security. Also save host and port of the sender server
282 : * this walreceiver is connected to.
283 : */
284 153 : tmp_conninfo = walrcv_get_conninfo(wrconn);
285 153 : walrcv_get_senderinfo(wrconn, &sender_host, &sender_port);
286 153 : SpinLockAcquire(&walrcv->mutex);
287 153 : memset(walrcv->conninfo, 0, MAXCONNINFO);
288 153 : if (tmp_conninfo)
289 153 : strlcpy(walrcv->conninfo, tmp_conninfo, MAXCONNINFO);
290 :
291 153 : memset(walrcv->sender_host, 0, NI_MAXHOST);
292 153 : if (sender_host)
293 153 : strlcpy(walrcv->sender_host, sender_host, NI_MAXHOST);
294 :
295 153 : walrcv->sender_port = sender_port;
296 153 : walrcv->ready_to_display = true;
297 153 : SpinLockRelease(&walrcv->mutex);
298 :
299 153 : if (tmp_conninfo)
300 153 : pfree(tmp_conninfo);
301 :
302 153 : if (sender_host)
303 153 : pfree(sender_host);
304 :
305 : /* Initialize buffers for processing messages */
306 153 : initStringInfo(&reply_message);
307 :
308 153 : first_stream = true;
309 : for (;;)
310 13 : {
311 : char *primary_sysid;
312 : char standby_sysid[32];
313 : WalRcvStreamOptions options;
314 :
315 : /*
316 : * Check that we're connected to a valid server using the
317 : * IDENTIFY_SYSTEM replication command.
318 : */
319 166 : primary_sysid = walrcv_identify_system(wrconn, &primaryTLI);
320 :
321 166 : snprintf(standby_sysid, sizeof(standby_sysid), UINT64_FORMAT,
322 : GetSystemIdentifier());
323 166 : if (strcmp(primary_sysid, standby_sysid) != 0)
324 : {
325 0 : ereport(ERROR,
326 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
327 : errmsg("database system identifier differs between the primary and standby"),
328 : errdetail("The primary's identifier is %s, the standby's identifier is %s.",
329 : primary_sysid, standby_sysid)));
330 : }
331 166 : pfree(primary_sysid);
332 :
333 : /*
334 : * Confirm that the current timeline of the primary is the same or
335 : * ahead of ours.
336 : */
337 166 : if (primaryTLI < startpointTLI)
338 0 : ereport(ERROR,
339 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
340 : errmsg("highest timeline %u of the primary is behind recovery timeline %u",
341 : primaryTLI, startpointTLI)));
342 :
343 : /*
344 : * Get any missing history files. We do this always, even when we're
345 : * not interested in that timeline, so that if we're promoted to
346 : * become the primary later on, we don't select the same timeline that
347 : * was already used in the current primary. This isn't bullet-proof -
348 : * you'll need some external software to manage your cluster if you
349 : * need to ensure that a unique timeline id is chosen in every case,
350 : * but let's avoid the confusion of timeline id collisions where we
351 : * can.
352 : */
353 166 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
354 :
355 : /*
356 : * Create temporary replication slot if requested, and update slot
357 : * name in shared memory. (Note the slot name cannot already be set
358 : * in this case.)
359 : */
360 166 : if (is_temp_slot)
361 : {
362 0 : snprintf(slotname, sizeof(slotname),
363 : "pg_walreceiver_%lld",
364 0 : (long long int) walrcv_get_backend_pid(wrconn));
365 :
366 0 : walrcv_create_slot(wrconn, slotname, true, false, false, 0, NULL);
367 :
368 0 : SpinLockAcquire(&walrcv->mutex);
369 0 : strlcpy(walrcv->slotname, slotname, NAMEDATALEN);
370 0 : SpinLockRelease(&walrcv->mutex);
371 : }
372 :
373 : /*
374 : * Start streaming.
375 : *
376 : * We'll try to start at the requested starting point and timeline,
377 : * even if it's different from the server's latest timeline. In case
378 : * we've already reached the end of the old timeline, the server will
379 : * finish the streaming immediately, and we will go back to await
380 : * orders from the startup process. If recovery_target_timeline is
381 : * 'latest', the startup process will scan pg_wal and find the new
382 : * history file, bump recovery target timeline, and ask us to restart
383 : * on the new timeline.
384 : */
385 166 : options.logical = false;
386 166 : options.startpoint = startpoint;
387 166 : options.slotname = slotname[0] != '\0' ? slotname : NULL;
388 166 : options.proto.physical.startpointTLI = startpointTLI;
389 166 : if (walrcv_startstreaming(wrconn, &options))
390 : {
391 165 : if (first_stream)
392 152 : ereport(LOG,
393 : errmsg("started streaming WAL from primary at %X/%08X on timeline %u",
394 : LSN_FORMAT_ARGS(startpoint), startpointTLI));
395 : else
396 13 : ereport(LOG,
397 : errmsg("restarted WAL streaming at %X/%08X on timeline %u",
398 : LSN_FORMAT_ARGS(startpoint), startpointTLI));
399 165 : first_stream = false;
400 :
401 : /*
402 : * Switch to STREAMING after a successful connection if current
403 : * state is CONNECTING. This switch happens after an initial
404 : * startup, or after a restart as determined by
405 : * WalRcvWaitForStartPosition().
406 : */
407 165 : SpinLockAcquire(&walrcv->mutex);
408 165 : if (walrcv->walRcvState == WALRCV_CONNECTING)
409 165 : walrcv->walRcvState = WALRCV_STREAMING;
410 165 : SpinLockRelease(&walrcv->mutex);
411 :
412 : /* Initialize LogstreamResult for processing messages */
413 165 : LogstreamResult.Write = LogstreamResult.Flush = GetXLogReplayRecPtr(NULL);
414 :
415 : /* Initialize nap wakeup times. */
416 165 : now = GetCurrentTimestamp();
417 825 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
418 660 : WalRcvComputeNextWakeup(i, now);
419 :
420 : /* Send initial reply/feedback messages. */
421 165 : XLogWalRcvSendReply(true, false, false);
422 165 : XLogWalRcvSendHSFeedback(true);
423 :
424 : /* Loop until end-of-streaming or error */
425 : for (;;)
426 57634 : {
427 : char *buf;
428 : int len;
429 57799 : bool endofwal = false;
430 57799 : pgsocket wait_fd = PGINVALID_SOCKET;
431 : int rc;
432 : TimestampTz nextWakeup;
433 : long nap;
434 :
435 : /*
436 : * Exit walreceiver if we're not in recovery. This should not
437 : * happen, but cross-check the status here.
438 : */
439 57799 : if (!RecoveryInProgress())
440 0 : ereport(FATAL,
441 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
442 : errmsg("cannot continue WAL streaming, recovery has already ended")));
443 :
444 : /* Process any requests or signals received recently */
445 57799 : CHECK_FOR_INTERRUPTS();
446 :
447 57799 : if (ConfigReloadPending)
448 : {
449 28 : ConfigReloadPending = false;
450 28 : ProcessConfigFile(PGC_SIGHUP);
451 : /* recompute wakeup times */
452 28 : now = GetCurrentTimestamp();
453 140 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
454 112 : WalRcvComputeNextWakeup(i, now);
455 28 : XLogWalRcvSendHSFeedback(true);
456 : }
457 :
458 : /* See if we can read data immediately */
459 57799 : len = walrcv_receive(wrconn, &buf, &wait_fd);
460 57771 : if (len != 0)
461 : {
462 : /*
463 : * Process the received data, and any subsequent data we
464 : * can read without blocking.
465 : */
466 : for (;;)
467 : {
468 148312 : if (len > 0)
469 : {
470 : /*
471 : * Something was received from primary, so adjust
472 : * the ping and terminate wakeup times.
473 : */
474 107314 : now = GetCurrentTimestamp();
475 107314 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_TERMINATE,
476 : now);
477 107314 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_PING, now);
478 107314 : XLogWalRcvProcessMsg(buf[0], &buf[1], len - 1,
479 : startpointTLI);
480 : }
481 40998 : else if (len == 0)
482 40952 : break;
483 46 : else if (len < 0)
484 : {
485 46 : ereport(LOG,
486 : (errmsg("replication terminated by primary server"),
487 : errdetail("End of WAL reached on timeline %u at %X/%08X.",
488 : startpointTLI,
489 : LSN_FORMAT_ARGS(LogstreamResult.Write))));
490 46 : endofwal = true;
491 46 : break;
492 : }
493 107314 : len = walrcv_receive(wrconn, &buf, &wait_fd);
494 : }
495 :
496 : /* Let the primary know that we received some data. */
497 40998 : XLogWalRcvSendReply(false, false, false);
498 :
499 : /*
500 : * If we've written some records, flush them to disk and
501 : * let the startup process and primary server know about
502 : * them.
503 : */
504 40997 : XLogWalRcvFlush(false, startpointTLI);
505 : }
506 :
507 : /* Check if we need to exit the streaming loop. */
508 57769 : if (endofwal)
509 45 : break;
510 :
511 : /* Find the soonest wakeup time, to limit our nap. */
512 57724 : nextWakeup = TIMESTAMP_INFINITY;
513 288620 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
514 230896 : nextWakeup = Min(wakeup[i], nextWakeup);
515 :
516 : /* Calculate the nap time, clamping as necessary. */
517 57724 : now = GetCurrentTimestamp();
518 57724 : nap = TimestampDifferenceMilliseconds(now, nextWakeup);
519 :
520 : /*
521 : * Ideally we would reuse a WaitEventSet object repeatedly
522 : * here to avoid the overheads of WaitLatchOrSocket on epoll
523 : * systems, but we can't be sure that libpq (or any other
524 : * walreceiver implementation) has the same socket (even if
525 : * the fd is the same number, it may have been closed and
526 : * reopened since the last time). In future, if there is a
527 : * function for removing sockets from WaitEventSet, then we
528 : * could add and remove just the socket each time, potentially
529 : * avoiding some system calls.
530 : */
531 : Assert(wait_fd != PGINVALID_SOCKET);
532 57724 : rc = WaitLatchOrSocket(MyLatch,
533 : WL_EXIT_ON_PM_DEATH | WL_SOCKET_READABLE |
534 : WL_TIMEOUT | WL_LATCH_SET,
535 : wait_fd,
536 : nap,
537 : WAIT_EVENT_WAL_RECEIVER_MAIN);
538 57724 : if (rc & WL_LATCH_SET)
539 : {
540 13648 : ResetLatch(MyLatch);
541 13648 : CHECK_FOR_INTERRUPTS();
542 :
543 13558 : if (walrcv->apply_reply_requested)
544 : {
545 : /*
546 : * The recovery process has asked us to send apply
547 : * feedback now. Make sure the flag is really set to
548 : * false in shared memory before sending the reply, so
549 : * we don't miss a new request for a reply.
550 : */
551 13504 : walrcv->apply_reply_requested = false;
552 13504 : pg_memory_barrier();
553 13504 : XLogWalRcvSendReply(false, false, true);
554 : }
555 : }
556 57634 : if (rc & WL_TIMEOUT)
557 : {
558 : /*
559 : * We didn't receive anything new. If we haven't heard
560 : * anything from the server for more than
561 : * wal_receiver_timeout / 2, ping the server. Also, if
562 : * it's been longer than wal_receiver_status_interval
563 : * since the last update we sent, send a status update to
564 : * the primary anyway, to report any progress in applying
565 : * WAL.
566 : */
567 6 : bool requestReply = false;
568 :
569 : /*
570 : * Report pending statistics to the cumulative stats
571 : * system. This location is useful for the report as it
572 : * is not within a tight loop in the WAL receiver, to
573 : * avoid bloating pgstats with requests, while also making
574 : * sure that the reports happen each time a status update
575 : * is sent.
576 : */
577 6 : pgstat_report_wal(false);
578 :
579 : /*
580 : * Check if time since last receive from primary has
581 : * reached the configured limit.
582 : */
583 6 : now = GetCurrentTimestamp();
584 6 : if (now >= wakeup[WALRCV_WAKEUP_TERMINATE])
585 0 : ereport(ERROR,
586 : (errcode(ERRCODE_CONNECTION_FAILURE),
587 : errmsg("terminating walreceiver due to timeout")));
588 :
589 : /*
590 : * If we didn't receive anything new for half of receiver
591 : * replication timeout, then ping the server.
592 : */
593 6 : if (now >= wakeup[WALRCV_WAKEUP_PING])
594 : {
595 0 : requestReply = true;
596 0 : wakeup[WALRCV_WAKEUP_PING] = TIMESTAMP_INFINITY;
597 : }
598 :
599 6 : XLogWalRcvSendReply(requestReply, requestReply, false);
600 6 : XLogWalRcvSendHSFeedback(false);
601 : }
602 : }
603 :
604 : /*
605 : * The backend finished streaming. Exit streaming COPY-mode from
606 : * our side, too.
607 : */
608 45 : walrcv_endstreaming(wrconn, &primaryTLI);
609 :
610 : /*
611 : * If the server had switched to a new timeline that we didn't
612 : * know about when we began streaming, fetch its timeline history
613 : * file now.
614 : */
615 13 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
616 : }
617 : else
618 0 : ereport(LOG,
619 : (errmsg("primary server contains no more WAL on requested timeline %u",
620 : startpointTLI)));
621 :
622 : /*
623 : * End of WAL reached on the requested timeline. Close the last
624 : * segment, and await for new orders from the startup process.
625 : */
626 13 : if (recvFile >= 0)
627 : {
628 : char xlogfname[MAXFNAMELEN];
629 :
630 12 : XLogWalRcvFlush(false, startpointTLI);
631 12 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
632 12 : if (close(recvFile) != 0)
633 0 : ereport(PANIC,
634 : (errcode_for_file_access(),
635 : errmsg("could not close WAL segment %s: %m",
636 : xlogfname)));
637 :
638 : /*
639 : * Create .done file forcibly to prevent the streamed segment from
640 : * being archived later.
641 : */
642 12 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
643 12 : XLogArchiveForceDone(xlogfname);
644 : else
645 0 : XLogArchiveNotify(xlogfname);
646 : }
647 13 : recvFile = -1;
648 :
649 13 : elog(DEBUG1, "walreceiver ended streaming and awaits new instructions");
650 13 : WalRcvWaitForStartPosition(&startpoint, &startpointTLI);
651 : }
652 : /* not reached */
653 : }
654 :
655 : /*
656 : * Wait for startup process to set receiveStart and receiveStartTLI.
657 : */
658 : static void
659 13 : WalRcvWaitForStartPosition(XLogRecPtr *startpoint, TimeLineID *startpointTLI)
660 : {
661 13 : WalRcvData *walrcv = WalRcv;
662 : int state;
663 :
664 13 : SpinLockAcquire(&walrcv->mutex);
665 13 : state = walrcv->walRcvState;
666 13 : if (state != WALRCV_STREAMING && state != WALRCV_CONNECTING)
667 : {
668 0 : SpinLockRelease(&walrcv->mutex);
669 0 : if (state == WALRCV_STOPPING)
670 0 : proc_exit(0);
671 : else
672 0 : elog(FATAL, "unexpected walreceiver state");
673 : }
674 13 : walrcv->walRcvState = WALRCV_WAITING;
675 13 : walrcv->receiveStart = InvalidXLogRecPtr;
676 13 : walrcv->receiveStartTLI = 0;
677 13 : SpinLockRelease(&walrcv->mutex);
678 :
679 13 : set_ps_display("idle");
680 :
681 : /*
682 : * nudge startup process to notice that we've stopped streaming and are
683 : * now waiting for instructions.
684 : */
685 13 : WakeupRecovery();
686 : for (;;)
687 : {
688 26 : ResetLatch(MyLatch);
689 :
690 26 : CHECK_FOR_INTERRUPTS();
691 :
692 26 : SpinLockAcquire(&walrcv->mutex);
693 : Assert(walrcv->walRcvState == WALRCV_RESTARTING ||
694 : walrcv->walRcvState == WALRCV_WAITING ||
695 : walrcv->walRcvState == WALRCV_STOPPING);
696 26 : if (walrcv->walRcvState == WALRCV_RESTARTING)
697 : {
698 : /*
699 : * No need to handle changes in primary_conninfo or
700 : * primary_slot_name here. Startup process will signal us to
701 : * terminate in case those change.
702 : */
703 13 : *startpoint = walrcv->receiveStart;
704 13 : *startpointTLI = walrcv->receiveStartTLI;
705 13 : walrcv->walRcvState = WALRCV_CONNECTING;
706 13 : SpinLockRelease(&walrcv->mutex);
707 13 : break;
708 : }
709 13 : if (walrcv->walRcvState == WALRCV_STOPPING)
710 : {
711 : /*
712 : * We should've received SIGTERM if the startup process wants us
713 : * to die, but might as well check it here too.
714 : */
715 0 : SpinLockRelease(&walrcv->mutex);
716 0 : proc_exit(1);
717 : }
718 13 : SpinLockRelease(&walrcv->mutex);
719 :
720 13 : (void) WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, 0,
721 : WAIT_EVENT_WAL_RECEIVER_WAIT_START);
722 : }
723 :
724 13 : if (update_process_title)
725 : {
726 : char activitymsg[50];
727 :
728 13 : snprintf(activitymsg, sizeof(activitymsg), "restarting at %X/%08X",
729 13 : LSN_FORMAT_ARGS(*startpoint));
730 13 : set_ps_display(activitymsg);
731 : }
732 13 : }
733 :
734 : /*
735 : * Fetch any missing timeline history files between 'first' and 'last'
736 : * (inclusive) from the server.
737 : */
738 : static void
739 179 : WalRcvFetchTimeLineHistoryFiles(TimeLineID first, TimeLineID last)
740 : {
741 : TimeLineID tli;
742 :
743 381 : for (tli = first; tli <= last; tli++)
744 : {
745 : /* there's no history file for timeline 1 */
746 202 : if (tli != 1 && !existsTimeLineHistory(tli))
747 : {
748 : char *fname;
749 : char *content;
750 : int len;
751 : char expectedfname[MAXFNAMELEN];
752 :
753 12 : ereport(LOG,
754 : (errmsg("fetching timeline history file for timeline %u from primary server",
755 : tli)));
756 :
757 12 : walrcv_readtimelinehistoryfile(wrconn, tli, &fname, &content, &len);
758 :
759 : /*
760 : * Check that the filename on the primary matches what we
761 : * calculated ourselves. This is just a sanity check, it should
762 : * always match.
763 : */
764 12 : TLHistoryFileName(expectedfname, tli);
765 12 : if (strcmp(fname, expectedfname) != 0)
766 0 : ereport(ERROR,
767 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
768 : errmsg_internal("primary reported unexpected file name for timeline history file of timeline %u",
769 : tli)));
770 :
771 : /*
772 : * Write the file to pg_wal.
773 : */
774 12 : writeTimeLineHistoryFile(tli, content, len);
775 :
776 : /*
777 : * Mark the streamed history file as ready for archiving if
778 : * archive_mode is always.
779 : */
780 12 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
781 12 : XLogArchiveForceDone(fname);
782 : else
783 0 : XLogArchiveNotify(fname);
784 :
785 12 : pfree(fname);
786 12 : pfree(content);
787 : }
788 : }
789 179 : }
790 :
791 : /*
792 : * Mark us as STOPPED in shared memory at exit.
793 : */
794 : static void
795 265 : WalRcvDie(int code, Datum arg)
796 : {
797 265 : WalRcvData *walrcv = WalRcv;
798 265 : TimeLineID *startpointTLI_p = (TimeLineID *) DatumGetPointer(arg);
799 :
800 : Assert(*startpointTLI_p != 0);
801 :
802 : /* Ensure that all WAL records received are flushed to disk */
803 265 : XLogWalRcvFlush(true, *startpointTLI_p);
804 :
805 : /* Mark ourselves inactive in shared memory */
806 265 : SpinLockAcquire(&walrcv->mutex);
807 : Assert(walrcv->walRcvState == WALRCV_STREAMING ||
808 : walrcv->walRcvState == WALRCV_CONNECTING ||
809 : walrcv->walRcvState == WALRCV_RESTARTING ||
810 : walrcv->walRcvState == WALRCV_STARTING ||
811 : walrcv->walRcvState == WALRCV_WAITING ||
812 : walrcv->walRcvState == WALRCV_STOPPING);
813 : Assert(walrcv->pid == MyProcPid);
814 265 : walrcv->walRcvState = WALRCV_STOPPED;
815 265 : walrcv->pid = 0;
816 265 : walrcv->procno = INVALID_PROC_NUMBER;
817 265 : walrcv->ready_to_display = false;
818 265 : SpinLockRelease(&walrcv->mutex);
819 :
820 265 : ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
821 :
822 : /* Terminate the connection gracefully. */
823 265 : if (wrconn != NULL)
824 153 : walrcv_disconnect(wrconn);
825 :
826 : /* Wake up the startup process to notice promptly that we're gone */
827 265 : WakeupRecovery();
828 265 : }
829 :
830 : /*
831 : * Accept the message from XLOG stream, and process it.
832 : */
833 : static void
834 107314 : XLogWalRcvProcessMsg(unsigned char type, char *buf, Size len, TimeLineID tli)
835 : {
836 : int hdrlen;
837 : XLogRecPtr dataStart;
838 : XLogRecPtr walEnd;
839 : TimestampTz sendTime;
840 : bool replyRequested;
841 :
842 107314 : switch (type)
843 : {
844 106765 : case PqReplMsg_WALData:
845 : {
846 : StringInfoData incoming_message;
847 :
848 106765 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(int64);
849 106765 : if (len < hdrlen)
850 0 : ereport(ERROR,
851 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
852 : errmsg_internal("invalid WAL message received from primary")));
853 :
854 : /* initialize a StringInfo with the given buffer */
855 106765 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
856 :
857 : /* read the fields */
858 106765 : dataStart = pq_getmsgint64(&incoming_message);
859 106765 : walEnd = pq_getmsgint64(&incoming_message);
860 106765 : sendTime = pq_getmsgint64(&incoming_message);
861 106765 : ProcessWalSndrMessage(walEnd, sendTime);
862 :
863 106765 : buf += hdrlen;
864 106765 : len -= hdrlen;
865 106765 : XLogWalRcvWrite(buf, len, dataStart, tli);
866 106765 : break;
867 : }
868 549 : case PqReplMsg_Keepalive:
869 : {
870 : StringInfoData incoming_message;
871 :
872 549 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(char);
873 549 : if (len != hdrlen)
874 0 : ereport(ERROR,
875 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
876 : errmsg_internal("invalid keepalive message received from primary")));
877 :
878 : /* initialize a StringInfo with the given buffer */
879 549 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
880 :
881 : /* read the fields */
882 549 : walEnd = pq_getmsgint64(&incoming_message);
883 549 : sendTime = pq_getmsgint64(&incoming_message);
884 549 : replyRequested = pq_getmsgbyte(&incoming_message);
885 :
886 549 : ProcessWalSndrMessage(walEnd, sendTime);
887 :
888 : /* If the primary requested a reply, send one immediately */
889 549 : if (replyRequested)
890 549 : XLogWalRcvSendReply(true, false, false);
891 549 : break;
892 : }
893 0 : default:
894 0 : ereport(ERROR,
895 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
896 : errmsg_internal("invalid replication message type %d",
897 : type)));
898 : }
899 107314 : }
900 :
901 : /*
902 : * Write XLOG data to disk.
903 : */
904 : static void
905 106765 : XLogWalRcvWrite(char *buf, Size nbytes, XLogRecPtr recptr, TimeLineID tli)
906 : {
907 : int startoff;
908 : int byteswritten;
909 : instr_time start;
910 :
911 : Assert(tli != 0);
912 :
913 213964 : while (nbytes > 0)
914 : {
915 : int segbytes;
916 :
917 : /* Close the current segment if it's completed */
918 107199 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
919 434 : XLogWalRcvClose(recptr, tli);
920 :
921 107199 : if (recvFile < 0)
922 : {
923 : /* Create/use new log file */
924 892 : XLByteToSeg(recptr, recvSegNo, wal_segment_size);
925 892 : recvFile = XLogFileInit(recvSegNo, tli);
926 892 : recvFileTLI = tli;
927 : }
928 :
929 : /* Calculate the start offset of the received logs */
930 107199 : startoff = XLogSegmentOffset(recptr, wal_segment_size);
931 :
932 107199 : if (startoff + nbytes > wal_segment_size)
933 434 : segbytes = wal_segment_size - startoff;
934 : else
935 106765 : segbytes = nbytes;
936 :
937 : /* OK to write the logs */
938 107199 : errno = 0;
939 :
940 : /*
941 : * Measure I/O timing to write WAL data, for pg_stat_io.
942 : */
943 107199 : start = pgstat_prepare_io_time(track_wal_io_timing);
944 :
945 107199 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
946 107199 : byteswritten = pg_pwrite(recvFile, buf, segbytes, (pgoff_t) startoff);
947 107199 : pgstat_report_wait_end();
948 :
949 107199 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
950 : IOOP_WRITE, start, 1, byteswritten);
951 :
952 107199 : if (byteswritten <= 0)
953 : {
954 : char xlogfname[MAXFNAMELEN];
955 : int save_errno;
956 :
957 : /* if write didn't set errno, assume no disk space */
958 0 : if (errno == 0)
959 0 : errno = ENOSPC;
960 :
961 0 : save_errno = errno;
962 0 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
963 0 : errno = save_errno;
964 0 : ereport(PANIC,
965 : (errcode_for_file_access(),
966 : errmsg("could not write to WAL segment %s "
967 : "at offset %d, length %d: %m",
968 : xlogfname, startoff, segbytes)));
969 : }
970 :
971 : /* Update state for write */
972 107199 : recptr += byteswritten;
973 :
974 107199 : nbytes -= byteswritten;
975 107199 : buf += byteswritten;
976 :
977 107199 : LogstreamResult.Write = recptr;
978 : }
979 :
980 : /* Update shared-memory status */
981 106765 : pg_atomic_write_membarrier_u64(&WalRcv->writtenUpto, LogstreamResult.Write);
982 :
983 : /*
984 : * Wake up processes waiting for standby write LSN to reach current write
985 : * position.
986 : */
987 106765 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_WRITE, LogstreamResult.Write);
988 :
989 : /*
990 : * Close the current segment if it's fully written up in the last cycle of
991 : * the loop, to create its archive notification file soon. Otherwise WAL
992 : * archiving of the segment will be delayed until any data in the next
993 : * segment is received and written.
994 : */
995 106765 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
996 317 : XLogWalRcvClose(recptr, tli);
997 106765 : }
998 :
999 : /*
1000 : * Flush the log to disk.
1001 : *
1002 : * If we're in the midst of dying, it's unwise to do anything that might throw
1003 : * an error, so we skip sending a reply in that case.
1004 : */
1005 : static void
1006 42025 : XLogWalRcvFlush(bool dying, TimeLineID tli)
1007 : {
1008 : Assert(tli != 0);
1009 :
1010 42025 : if (LogstreamResult.Flush < LogstreamResult.Write)
1011 : {
1012 41311 : WalRcvData *walrcv = WalRcv;
1013 :
1014 41311 : issue_xlog_fsync(recvFile, recvSegNo, tli);
1015 :
1016 41311 : LogstreamResult.Flush = LogstreamResult.Write;
1017 :
1018 : /* Update shared-memory status */
1019 41311 : SpinLockAcquire(&walrcv->mutex);
1020 41311 : if (walrcv->flushedUpto < LogstreamResult.Flush)
1021 : {
1022 41311 : walrcv->latestChunkStart = walrcv->flushedUpto;
1023 41311 : walrcv->flushedUpto = LogstreamResult.Flush;
1024 41311 : walrcv->receivedTLI = tli;
1025 : }
1026 41311 : SpinLockRelease(&walrcv->mutex);
1027 :
1028 : /*
1029 : * Wake up processes waiting for standby flush LSN to reach current
1030 : * flush position.
1031 : */
1032 41311 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_FLUSH, LogstreamResult.Flush);
1033 :
1034 : /* Signal the startup process and walsender that new WAL has arrived */
1035 41311 : WakeupRecovery();
1036 41311 : if (AllowCascadeReplication())
1037 41311 : WalSndWakeup(true, false);
1038 :
1039 : /* Report XLOG streaming progress in PS display */
1040 41311 : if (update_process_title)
1041 : {
1042 : char activitymsg[50];
1043 :
1044 41311 : snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%08X",
1045 41311 : LSN_FORMAT_ARGS(LogstreamResult.Write));
1046 41311 : set_ps_display(activitymsg);
1047 : }
1048 :
1049 : /* Also let the primary know that we made some progress */
1050 41311 : if (!dying)
1051 : {
1052 41309 : XLogWalRcvSendReply(false, false, false);
1053 41309 : XLogWalRcvSendHSFeedback(false);
1054 : }
1055 : }
1056 42025 : }
1057 :
1058 : /*
1059 : * Close the current segment.
1060 : *
1061 : * Flush the segment to disk before closing it. Otherwise we have to
1062 : * reopen and fsync it later.
1063 : *
1064 : * Create an archive notification file since the segment is known completed.
1065 : */
1066 : static void
1067 751 : XLogWalRcvClose(XLogRecPtr recptr, TimeLineID tli)
1068 : {
1069 : char xlogfname[MAXFNAMELEN];
1070 :
1071 : Assert(recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size));
1072 : Assert(tli != 0);
1073 :
1074 : /*
1075 : * fsync() and close current file before we switch to next one. We would
1076 : * otherwise have to reopen this file to fsync it later
1077 : */
1078 751 : XLogWalRcvFlush(false, tli);
1079 :
1080 751 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
1081 :
1082 : /*
1083 : * XLOG segment files will be re-read by recovery in startup process soon,
1084 : * so we don't advise the OS to release cache pages associated with the
1085 : * file like XLogFileClose() does.
1086 : */
1087 751 : if (close(recvFile) != 0)
1088 0 : ereport(PANIC,
1089 : (errcode_for_file_access(),
1090 : errmsg("could not close WAL segment %s: %m",
1091 : xlogfname)));
1092 :
1093 : /*
1094 : * Create .done file forcibly to prevent the streamed segment from being
1095 : * archived later.
1096 : */
1097 751 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
1098 751 : XLogArchiveForceDone(xlogfname);
1099 : else
1100 0 : XLogArchiveNotify(xlogfname);
1101 :
1102 751 : recvFile = -1;
1103 751 : }
1104 :
1105 : /*
1106 : * Send reply message to primary, indicating our current WAL locations and
1107 : * time.
1108 : *
1109 : * The message is sent if 'force' is set, if enough time has passed since the
1110 : * last update to reach wal_receiver_status_interval, or if WAL locations have
1111 : * advanced since the previous status update. If wal_receiver_status_interval
1112 : * is disabled and 'force' is false, this function does nothing. Set 'force' to
1113 : * send the message unconditionally.
1114 : *
1115 : * Whether WAL locations are considered "advanced" depends on 'checkApply'.
1116 : * If 'checkApply' is false, only the write and flush locations are checked.
1117 : * This should be used when the call is triggered by write/flush activity
1118 : * (e.g., after walreceiver writes or flushes WAL), and avoids the
1119 : * apply-location check, which requires a spinlock. If 'checkApply' is true,
1120 : * the apply location is also considered. This should be used when the apply
1121 : * location is expected to advance (e.g., when the startup process requests
1122 : * an apply notification).
1123 : *
1124 : * If 'requestReply' is true, requests the server to reply immediately upon
1125 : * receiving this message. This is used for heartbeats, when approaching
1126 : * wal_receiver_timeout.
1127 : */
1128 : static void
1129 96531 : XLogWalRcvSendReply(bool force, bool requestReply, bool checkApply)
1130 : {
1131 : static XLogRecPtr writePtr = InvalidXLogRecPtr;
1132 : static XLogRecPtr flushPtr = InvalidXLogRecPtr;
1133 : static XLogRecPtr applyPtr = InvalidXLogRecPtr;
1134 96531 : XLogRecPtr latestApplyPtr = InvalidXLogRecPtr;
1135 : TimestampTz now;
1136 :
1137 : /*
1138 : * If the user doesn't want status to be reported to the primary, be sure
1139 : * to exit before doing anything at all.
1140 : */
1141 96531 : if (!force && wal_receiver_status_interval <= 0)
1142 0 : return;
1143 :
1144 : /* Get current timestamp. */
1145 96531 : now = GetCurrentTimestamp();
1146 :
1147 : /*
1148 : * We can compare the write and flush positions to the last message we
1149 : * sent without taking any lock, but the apply position requires a spin
1150 : * lock, so we don't check that unless it is expected to advance since the
1151 : * previous update, i.e., when 'checkApply' is true.
1152 : */
1153 96531 : if (!force && now < wakeup[WALRCV_WAKEUP_REPLY])
1154 : {
1155 95816 : if (checkApply)
1156 13504 : latestApplyPtr = GetXLogReplayRecPtr(NULL);
1157 :
1158 95816 : if (writePtr == LogstreamResult.Write
1159 54377 : && flushPtr == LogstreamResult.Flush
1160 13819 : && (!checkApply || applyPtr == latestApplyPtr))
1161 3585 : return;
1162 : }
1163 :
1164 : /* Make sure we wake up when it's time to send another reply. */
1165 92946 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_REPLY, now);
1166 :
1167 : /* Construct a new message */
1168 92946 : writePtr = LogstreamResult.Write;
1169 92946 : flushPtr = LogstreamResult.Flush;
1170 92946 : applyPtr = XLogRecPtrIsValid(latestApplyPtr) ?
1171 92946 : latestApplyPtr : GetXLogReplayRecPtr(NULL);
1172 :
1173 92946 : resetStringInfo(&reply_message);
1174 92946 : pq_sendbyte(&reply_message, PqReplMsg_StandbyStatusUpdate);
1175 92946 : pq_sendint64(&reply_message, writePtr);
1176 92946 : pq_sendint64(&reply_message, flushPtr);
1177 92946 : pq_sendint64(&reply_message, applyPtr);
1178 92946 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1179 92946 : pq_sendbyte(&reply_message, requestReply ? 1 : 0);
1180 :
1181 : /* Send it */
1182 92946 : elog(DEBUG2, "sending write %X/%08X flush %X/%08X apply %X/%08X%s",
1183 : LSN_FORMAT_ARGS(writePtr),
1184 : LSN_FORMAT_ARGS(flushPtr),
1185 : LSN_FORMAT_ARGS(applyPtr),
1186 : requestReply ? " (reply requested)" : "");
1187 :
1188 92946 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1189 : }
1190 :
1191 : /*
1192 : * Send hot standby feedback message to primary, plus the current time,
1193 : * in case they don't have a watch.
1194 : *
1195 : * If the user disables feedback, send one final message to tell sender
1196 : * to forget about the xmin on this standby. We also send this message
1197 : * on first connect because a previous connection might have set xmin
1198 : * on a replication slot. (If we're not using a slot it's harmless to
1199 : * send a feedback message explicitly setting InvalidTransactionId).
1200 : */
1201 : static void
1202 41508 : XLogWalRcvSendHSFeedback(bool immed)
1203 : {
1204 : TimestampTz now;
1205 : FullTransactionId nextFullXid;
1206 : TransactionId nextXid;
1207 : uint32 xmin_epoch,
1208 : catalog_xmin_epoch;
1209 : TransactionId xmin,
1210 : catalog_xmin;
1211 :
1212 : /* initially true so we always send at least one feedback message */
1213 : static bool primary_has_standby_xmin = true;
1214 :
1215 : /*
1216 : * If the user doesn't want status to be reported to the primary, be sure
1217 : * to exit before doing anything at all.
1218 : */
1219 41508 : if ((wal_receiver_status_interval <= 0 || !hot_standby_feedback) &&
1220 40908 : !primary_has_standby_xmin)
1221 41331 : return;
1222 :
1223 : /* Get current timestamp. */
1224 782 : now = GetCurrentTimestamp();
1225 :
1226 : /* Send feedback at most once per wal_receiver_status_interval. */
1227 782 : if (!immed && now < wakeup[WALRCV_WAKEUP_HSFEEDBACK])
1228 604 : return;
1229 :
1230 : /* Make sure we wake up when it's time to send feedback again. */
1231 178 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_HSFEEDBACK, now);
1232 :
1233 : /*
1234 : * If Hot Standby is not yet accepting connections there is nothing to
1235 : * send. Check this after the interval has expired to reduce number of
1236 : * calls.
1237 : *
1238 : * Bailing out here also ensures that we don't send feedback until we've
1239 : * read our own replication slot state, so we don't tell the primary to
1240 : * discard needed xmin or catalog_xmin from any slots that may exist on
1241 : * this replica.
1242 : */
1243 178 : if (!HotStandbyActive())
1244 1 : return;
1245 :
1246 : /*
1247 : * Make the expensive call to get the oldest xmin once we are certain
1248 : * everything else has been checked.
1249 : */
1250 177 : if (hot_standby_feedback)
1251 : {
1252 53 : GetReplicationHorizons(&xmin, &catalog_xmin);
1253 : }
1254 : else
1255 : {
1256 124 : xmin = InvalidTransactionId;
1257 124 : catalog_xmin = InvalidTransactionId;
1258 : }
1259 :
1260 : /*
1261 : * Get epoch and adjust if nextXid and oldestXmin are different sides of
1262 : * the epoch boundary.
1263 : */
1264 177 : nextFullXid = ReadNextFullTransactionId();
1265 177 : nextXid = XidFromFullTransactionId(nextFullXid);
1266 177 : xmin_epoch = EpochFromFullTransactionId(nextFullXid);
1267 177 : catalog_xmin_epoch = xmin_epoch;
1268 177 : if (nextXid < xmin)
1269 0 : xmin_epoch--;
1270 177 : if (nextXid < catalog_xmin)
1271 0 : catalog_xmin_epoch--;
1272 :
1273 177 : elog(DEBUG2, "sending hot standby feedback xmin %u epoch %u catalog_xmin %u catalog_xmin_epoch %u",
1274 : xmin, xmin_epoch, catalog_xmin, catalog_xmin_epoch);
1275 :
1276 : /* Construct the message and send it. */
1277 177 : resetStringInfo(&reply_message);
1278 177 : pq_sendbyte(&reply_message, PqReplMsg_HotStandbyFeedback);
1279 177 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1280 177 : pq_sendint32(&reply_message, xmin);
1281 177 : pq_sendint32(&reply_message, xmin_epoch);
1282 177 : pq_sendint32(&reply_message, catalog_xmin);
1283 177 : pq_sendint32(&reply_message, catalog_xmin_epoch);
1284 177 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1285 177 : if (TransactionIdIsValid(xmin) || TransactionIdIsValid(catalog_xmin))
1286 53 : primary_has_standby_xmin = true;
1287 : else
1288 124 : primary_has_standby_xmin = false;
1289 : }
1290 :
1291 : /*
1292 : * Update shared memory status upon receiving a message from primary.
1293 : *
1294 : * 'walEnd' and 'sendTime' are the end-of-WAL and timestamp of the latest
1295 : * message, reported by primary.
1296 : */
1297 : static void
1298 107314 : ProcessWalSndrMessage(XLogRecPtr walEnd, TimestampTz sendTime)
1299 : {
1300 107314 : WalRcvData *walrcv = WalRcv;
1301 107314 : TimestampTz lastMsgReceiptTime = GetCurrentTimestamp();
1302 :
1303 : /* Update shared-memory status */
1304 107314 : SpinLockAcquire(&walrcv->mutex);
1305 107314 : if (walrcv->latestWalEnd < walEnd)
1306 22646 : walrcv->latestWalEndTime = sendTime;
1307 107314 : walrcv->latestWalEnd = walEnd;
1308 107314 : walrcv->lastMsgSendTime = sendTime;
1309 107314 : walrcv->lastMsgReceiptTime = lastMsgReceiptTime;
1310 107314 : SpinLockRelease(&walrcv->mutex);
1311 :
1312 107314 : if (message_level_is_interesting(DEBUG2))
1313 : {
1314 : char *sendtime;
1315 : char *receipttime;
1316 : int applyDelay;
1317 :
1318 : /* Copy because timestamptz_to_str returns a static buffer */
1319 406 : sendtime = pstrdup(timestamptz_to_str(sendTime));
1320 406 : receipttime = pstrdup(timestamptz_to_str(lastMsgReceiptTime));
1321 406 : applyDelay = GetReplicationApplyDelay();
1322 :
1323 : /* apply delay is not available */
1324 406 : if (applyDelay == -1)
1325 1 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay (N/A) transfer latency %d ms",
1326 : sendtime,
1327 : receipttime,
1328 : GetReplicationTransferLatency());
1329 : else
1330 405 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay %d ms transfer latency %d ms",
1331 : sendtime,
1332 : receipttime,
1333 : applyDelay,
1334 : GetReplicationTransferLatency());
1335 :
1336 406 : pfree(sendtime);
1337 406 : pfree(receipttime);
1338 : }
1339 107314 : }
1340 :
1341 : /*
1342 : * Compute the next wakeup time for a given wakeup reason. Can be called to
1343 : * initialize a wakeup time, to adjust it for the next wakeup, or to
1344 : * reinitialize it when GUCs have changed. We ask the caller to pass in the
1345 : * value of "now" because this frequently avoids multiple calls of
1346 : * GetCurrentTimestamp(). It had better be a reasonably up-to-date value
1347 : * though.
1348 : */
1349 : static void
1350 308524 : WalRcvComputeNextWakeup(WalRcvWakeupReason reason, TimestampTz now)
1351 : {
1352 308524 : switch (reason)
1353 : {
1354 107507 : case WALRCV_WAKEUP_TERMINATE:
1355 107507 : if (wal_receiver_timeout <= 0)
1356 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1357 : else
1358 107507 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout);
1359 107507 : break;
1360 107507 : case WALRCV_WAKEUP_PING:
1361 107507 : if (wal_receiver_timeout <= 0)
1362 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1363 : else
1364 107507 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout / 2);
1365 107507 : break;
1366 371 : case WALRCV_WAKEUP_HSFEEDBACK:
1367 371 : if (!hot_standby_feedback || wal_receiver_status_interval <= 0)
1368 270 : wakeup[reason] = TIMESTAMP_INFINITY;
1369 : else
1370 101 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1371 371 : break;
1372 93139 : case WALRCV_WAKEUP_REPLY:
1373 93139 : if (wal_receiver_status_interval <= 0)
1374 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1375 : else
1376 93139 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1377 93139 : break;
1378 : /* there's intentionally no default: here */
1379 : }
1380 308524 : }
1381 :
1382 : /*
1383 : * Wake up the walreceiver main loop.
1384 : *
1385 : * This is called by the startup process whenever interesting xlog records
1386 : * are applied, so that walreceiver can check if it needs to send an apply
1387 : * notification back to the primary which may be waiting in a COMMIT with
1388 : * synchronous_commit = remote_apply.
1389 : */
1390 : void
1391 13432 : WalRcvRequestApplyReply(void)
1392 : {
1393 : ProcNumber procno;
1394 :
1395 13432 : WalRcv->apply_reply_requested = true;
1396 : /* fetching the proc number is probably atomic, but don't rely on it */
1397 13432 : SpinLockAcquire(&WalRcv->mutex);
1398 13432 : procno = WalRcv->procno;
1399 13432 : SpinLockRelease(&WalRcv->mutex);
1400 13432 : if (procno != INVALID_PROC_NUMBER)
1401 13236 : SetLatch(&GetPGProcByNumber(procno)->procLatch);
1402 13432 : }
1403 :
1404 : /*
1405 : * Return a string constant representing the state. This is used
1406 : * in system functions and views, and should *not* be translated.
1407 : */
1408 : static const char *
1409 18 : WalRcvGetStateString(WalRcvState state)
1410 : {
1411 18 : switch (state)
1412 : {
1413 0 : case WALRCV_STOPPED:
1414 0 : return "stopped";
1415 0 : case WALRCV_STARTING:
1416 0 : return "starting";
1417 0 : case WALRCV_CONNECTING:
1418 0 : return "connecting";
1419 18 : case WALRCV_STREAMING:
1420 18 : return "streaming";
1421 0 : case WALRCV_WAITING:
1422 0 : return "waiting";
1423 0 : case WALRCV_RESTARTING:
1424 0 : return "restarting";
1425 0 : case WALRCV_STOPPING:
1426 0 : return "stopping";
1427 : }
1428 0 : return "UNKNOWN";
1429 : }
1430 :
1431 : /*
1432 : * Returns activity of WAL receiver, including pid, state and xlog locations
1433 : * received from the WAL sender of another server.
1434 : */
1435 : Datum
1436 29 : pg_stat_get_wal_receiver(PG_FUNCTION_ARGS)
1437 : {
1438 : TupleDesc tupdesc;
1439 : Datum *values;
1440 : bool *nulls;
1441 : int pid;
1442 : bool ready_to_display;
1443 : WalRcvState state;
1444 : XLogRecPtr receive_start_lsn;
1445 : TimeLineID receive_start_tli;
1446 : XLogRecPtr written_lsn;
1447 : XLogRecPtr flushed_lsn;
1448 : TimeLineID received_tli;
1449 : TimestampTz last_send_time;
1450 : TimestampTz last_receipt_time;
1451 : XLogRecPtr latest_end_lsn;
1452 : TimestampTz latest_end_time;
1453 : char sender_host[NI_MAXHOST];
1454 29 : int sender_port = 0;
1455 : char slotname[NAMEDATALEN];
1456 : char conninfo[MAXCONNINFO];
1457 :
1458 : /* Take a lock to ensure value consistency */
1459 29 : SpinLockAcquire(&WalRcv->mutex);
1460 29 : pid = (int) WalRcv->pid;
1461 29 : ready_to_display = WalRcv->ready_to_display;
1462 29 : state = WalRcv->walRcvState;
1463 29 : receive_start_lsn = WalRcv->receiveStart;
1464 29 : receive_start_tli = WalRcv->receiveStartTLI;
1465 29 : flushed_lsn = WalRcv->flushedUpto;
1466 29 : received_tli = WalRcv->receivedTLI;
1467 29 : last_send_time = WalRcv->lastMsgSendTime;
1468 29 : last_receipt_time = WalRcv->lastMsgReceiptTime;
1469 29 : latest_end_lsn = WalRcv->latestWalEnd;
1470 29 : latest_end_time = WalRcv->latestWalEndTime;
1471 29 : strlcpy(slotname, WalRcv->slotname, sizeof(slotname));
1472 29 : strlcpy(sender_host, WalRcv->sender_host, sizeof(sender_host));
1473 29 : sender_port = WalRcv->sender_port;
1474 29 : strlcpy(conninfo, WalRcv->conninfo, sizeof(conninfo));
1475 29 : SpinLockRelease(&WalRcv->mutex);
1476 :
1477 : /*
1478 : * No WAL receiver (or not ready yet), just return a tuple with NULL
1479 : * values
1480 : */
1481 29 : if (pid == 0 || !ready_to_display)
1482 11 : PG_RETURN_NULL();
1483 :
1484 : /*
1485 : * Read "writtenUpto" without holding a spinlock. Note that it may not be
1486 : * consistent with the other shared variables of the WAL receiver
1487 : * protected by a spinlock, but this should not be used for data integrity
1488 : * checks.
1489 : */
1490 18 : written_lsn = pg_atomic_read_u64(&WalRcv->writtenUpto);
1491 :
1492 : /* determine result type */
1493 18 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1494 0 : elog(ERROR, "return type must be a row type");
1495 :
1496 18 : values = palloc0_array(Datum, tupdesc->natts);
1497 18 : nulls = palloc0_array(bool, tupdesc->natts);
1498 :
1499 : /* Fetch values */
1500 18 : values[0] = Int32GetDatum(pid);
1501 :
1502 18 : if (!has_privs_of_role(GetUserId(), ROLE_PG_READ_ALL_STATS))
1503 : {
1504 : /*
1505 : * Only superusers and roles with privileges of pg_read_all_stats can
1506 : * see details. Other users only get the pid value to know whether it
1507 : * is a WAL receiver, but no details.
1508 : */
1509 0 : memset(&nulls[1], true, sizeof(bool) * (tupdesc->natts - 1));
1510 : }
1511 : else
1512 : {
1513 18 : values[1] = CStringGetTextDatum(WalRcvGetStateString(state));
1514 :
1515 18 : if (!XLogRecPtrIsValid(receive_start_lsn))
1516 0 : nulls[2] = true;
1517 : else
1518 18 : values[2] = LSNGetDatum(receive_start_lsn);
1519 18 : values[3] = Int32GetDatum(receive_start_tli);
1520 18 : if (!XLogRecPtrIsValid(written_lsn))
1521 0 : nulls[4] = true;
1522 : else
1523 18 : values[4] = LSNGetDatum(written_lsn);
1524 18 : if (!XLogRecPtrIsValid(flushed_lsn))
1525 0 : nulls[5] = true;
1526 : else
1527 18 : values[5] = LSNGetDatum(flushed_lsn);
1528 18 : values[6] = Int32GetDatum(received_tli);
1529 18 : if (last_send_time == 0)
1530 0 : nulls[7] = true;
1531 : else
1532 18 : values[7] = TimestampTzGetDatum(last_send_time);
1533 18 : if (last_receipt_time == 0)
1534 0 : nulls[8] = true;
1535 : else
1536 18 : values[8] = TimestampTzGetDatum(last_receipt_time);
1537 18 : if (!XLogRecPtrIsValid(latest_end_lsn))
1538 0 : nulls[9] = true;
1539 : else
1540 18 : values[9] = LSNGetDatum(latest_end_lsn);
1541 18 : if (latest_end_time == 0)
1542 0 : nulls[10] = true;
1543 : else
1544 18 : values[10] = TimestampTzGetDatum(latest_end_time);
1545 18 : if (*slotname == '\0')
1546 16 : nulls[11] = true;
1547 : else
1548 2 : values[11] = CStringGetTextDatum(slotname);
1549 18 : if (*sender_host == '\0')
1550 0 : nulls[12] = true;
1551 : else
1552 18 : values[12] = CStringGetTextDatum(sender_host);
1553 18 : if (sender_port == 0)
1554 0 : nulls[13] = true;
1555 : else
1556 18 : values[13] = Int32GetDatum(sender_port);
1557 18 : if (*conninfo == '\0')
1558 0 : nulls[14] = true;
1559 : else
1560 18 : values[14] = CStringGetTextDatum(conninfo);
1561 : }
1562 :
1563 : /* Returns the record as Datum */
1564 18 : PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
1565 : }
|
