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 : * exit(0). Emergency termination is by SIGQUIT; like any postmaster child
34 : * process, the walreceiver will simply abort and exit on SIGQUIT. A close
35 : * 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) 2010-2025, 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 "catalog/pg_authid.h"
61 : #include "funcapi.h"
62 : #include "libpq/pqformat.h"
63 : #include "libpq/pqsignal.h"
64 : #include "miscadmin.h"
65 : #include "pgstat.h"
66 : #include "postmaster/auxprocess.h"
67 : #include "postmaster/interrupt.h"
68 : #include "replication/walreceiver.h"
69 : #include "replication/walsender.h"
70 : #include "storage/ipc.h"
71 : #include "storage/proc.h"
72 : #include "storage/procarray.h"
73 : #include "storage/procsignal.h"
74 : #include "tcop/tcopprot.h"
75 : #include "utils/acl.h"
76 : #include "utils/builtins.h"
77 : #include "utils/guc.h"
78 : #include "utils/pg_lsn.h"
79 : #include "utils/ps_status.h"
80 : #include "utils/timestamp.h"
81 :
82 :
83 : /*
84 : * GUC variables. (Other variables that affect walreceiver are in xlog.c
85 : * because they're passed down from the startup process, for better
86 : * synchronization.)
87 : */
88 : int wal_receiver_status_interval;
89 : int wal_receiver_timeout;
90 : bool hot_standby_feedback;
91 :
92 : /* libpqwalreceiver connection */
93 : static WalReceiverConn *wrconn = NULL;
94 : WalReceiverFunctionsType *WalReceiverFunctions = NULL;
95 :
96 : /*
97 : * These variables are used similarly to openLogFile/SegNo,
98 : * but for walreceiver to write the XLOG. recvFileTLI is the TimeLineID
99 : * corresponding the filename of recvFile.
100 : */
101 : static int recvFile = -1;
102 : static TimeLineID recvFileTLI = 0;
103 : static XLogSegNo recvSegNo = 0;
104 :
105 : /*
106 : * LogstreamResult indicates the byte positions that we have already
107 : * written/fsynced.
108 : */
109 : static struct
110 : {
111 : XLogRecPtr Write; /* last byte + 1 written out in the standby */
112 : XLogRecPtr Flush; /* last byte + 1 flushed in the standby */
113 : } LogstreamResult;
114 :
115 : /*
116 : * Reasons to wake up and perform periodic tasks.
117 : */
118 : typedef enum WalRcvWakeupReason
119 : {
120 : WALRCV_WAKEUP_TERMINATE,
121 : WALRCV_WAKEUP_PING,
122 : WALRCV_WAKEUP_REPLY,
123 : WALRCV_WAKEUP_HSFEEDBACK,
124 : #define NUM_WALRCV_WAKEUPS (WALRCV_WAKEUP_HSFEEDBACK + 1)
125 : } WalRcvWakeupReason;
126 :
127 : /*
128 : * Wake up times for periodic tasks.
129 : */
130 : static TimestampTz wakeup[NUM_WALRCV_WAKEUPS];
131 :
132 : static StringInfoData reply_message;
133 :
134 : /* Prototypes for private functions */
135 : static void WalRcvFetchTimeLineHistoryFiles(TimeLineID first, TimeLineID last);
136 : static void WalRcvWaitForStartPosition(XLogRecPtr *startpoint, TimeLineID *startpointTLI);
137 : static void WalRcvDie(int code, Datum arg);
138 : static void XLogWalRcvProcessMsg(unsigned char type, char *buf, Size len,
139 : TimeLineID tli);
140 : static void XLogWalRcvWrite(char *buf, Size nbytes, XLogRecPtr recptr,
141 : TimeLineID tli);
142 : static void XLogWalRcvFlush(bool dying, TimeLineID tli);
143 : static void XLogWalRcvClose(XLogRecPtr recptr, TimeLineID tli);
144 : static void XLogWalRcvSendReply(bool force, bool requestReply);
145 : static void XLogWalRcvSendHSFeedback(bool immed);
146 : static void ProcessWalSndrMessage(XLogRecPtr walEnd, TimestampTz sendTime);
147 : static void WalRcvComputeNextWakeup(WalRcvWakeupReason reason, TimestampTz now);
148 :
149 :
150 : /* Main entry point for walreceiver process */
151 : void
152 454 : WalReceiverMain(const void *startup_data, size_t startup_data_len)
153 : {
154 : char conninfo[MAXCONNINFO];
155 : char *tmp_conninfo;
156 : char slotname[NAMEDATALEN];
157 : bool is_temp_slot;
158 : XLogRecPtr startpoint;
159 : TimeLineID startpointTLI;
160 : TimeLineID primaryTLI;
161 : bool first_stream;
162 : WalRcvData *walrcv;
163 : TimestampTz now;
164 : char *err;
165 454 : char *sender_host = NULL;
166 454 : int sender_port = 0;
167 : char *appname;
168 :
169 : Assert(startup_data_len == 0);
170 :
171 454 : MyBackendType = B_WAL_RECEIVER;
172 454 : AuxiliaryProcessMainCommon();
173 :
174 : /*
175 : * WalRcv should be set up already (if we are a backend, we inherit this
176 : * by fork() or EXEC_BACKEND mechanism from the postmaster).
177 : */
178 454 : walrcv = WalRcv;
179 : Assert(walrcv != NULL);
180 :
181 : /*
182 : * Mark walreceiver as running in shared memory.
183 : *
184 : * Do this as early as possible, so that if we fail later on, we'll set
185 : * state to STOPPED. If we die before this, the startup process will keep
186 : * waiting for us to start up, until it times out.
187 : */
188 454 : SpinLockAcquire(&walrcv->mutex);
189 : Assert(walrcv->pid == 0);
190 454 : switch (walrcv->walRcvState)
191 : {
192 0 : case WALRCV_STOPPING:
193 : /* If we've already been requested to stop, don't start up. */
194 0 : walrcv->walRcvState = WALRCV_STOPPED;
195 : /* fall through */
196 :
197 8 : case WALRCV_STOPPED:
198 8 : SpinLockRelease(&walrcv->mutex);
199 8 : ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
200 8 : proc_exit(1);
201 : break;
202 :
203 446 : case WALRCV_STARTING:
204 : /* The usual case */
205 446 : break;
206 :
207 0 : case WALRCV_WAITING:
208 : case WALRCV_STREAMING:
209 : case WALRCV_RESTARTING:
210 : default:
211 : /* Shouldn't happen */
212 0 : SpinLockRelease(&walrcv->mutex);
213 0 : elog(PANIC, "walreceiver still running according to shared memory state");
214 : }
215 : /* Advertise our PID so that the startup process can kill us */
216 446 : walrcv->pid = MyProcPid;
217 446 : walrcv->walRcvState = WALRCV_STREAMING;
218 :
219 : /* Fetch information required to start streaming */
220 446 : walrcv->ready_to_display = false;
221 446 : strlcpy(conninfo, walrcv->conninfo, MAXCONNINFO);
222 446 : strlcpy(slotname, walrcv->slotname, NAMEDATALEN);
223 446 : is_temp_slot = walrcv->is_temp_slot;
224 446 : startpoint = walrcv->receiveStart;
225 446 : startpointTLI = walrcv->receiveStartTLI;
226 :
227 : /*
228 : * At most one of is_temp_slot and slotname can be set; otherwise,
229 : * RequestXLogStreaming messed up.
230 : */
231 : Assert(!is_temp_slot || (slotname[0] == '\0'));
232 :
233 : /* Initialise to a sanish value */
234 446 : now = GetCurrentTimestamp();
235 446 : walrcv->lastMsgSendTime =
236 446 : walrcv->lastMsgReceiptTime = walrcv->latestWalEndTime = now;
237 :
238 : /* Report our proc number so that others can wake us up */
239 446 : walrcv->procno = MyProcNumber;
240 :
241 446 : SpinLockRelease(&walrcv->mutex);
242 :
243 446 : pg_atomic_write_u64(&WalRcv->writtenUpto, 0);
244 :
245 : /* Arrange to clean up at walreceiver exit */
246 446 : on_shmem_exit(WalRcvDie, PointerGetDatum(&startpointTLI));
247 :
248 : /* Properly accept or ignore signals the postmaster might send us */
249 446 : pqsignal(SIGHUP, SignalHandlerForConfigReload); /* set flag to read config
250 : * file */
251 446 : pqsignal(SIGINT, SIG_IGN);
252 446 : pqsignal(SIGTERM, die); /* request shutdown */
253 : /* SIGQUIT handler was already set up by InitPostmasterChild */
254 446 : pqsignal(SIGALRM, SIG_IGN);
255 446 : pqsignal(SIGPIPE, SIG_IGN);
256 446 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
257 446 : pqsignal(SIGUSR2, SIG_IGN);
258 :
259 : /* Reset some signals that are accepted by postmaster but not here */
260 446 : pqsignal(SIGCHLD, SIG_DFL);
261 :
262 : /* Load the libpq-specific functions */
263 446 : load_file("libpqwalreceiver", false);
264 446 : if (WalReceiverFunctions == NULL)
265 0 : elog(ERROR, "libpqwalreceiver didn't initialize correctly");
266 :
267 : /* Unblock signals (they were blocked when the postmaster forked us) */
268 446 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
269 :
270 : /* Establish the connection to the primary for XLOG streaming */
271 446 : appname = cluster_name[0] ? cluster_name : "walreceiver";
272 446 : wrconn = walrcv_connect(conninfo, true, false, false, appname, &err);
273 442 : if (!wrconn)
274 166 : 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 : */
284 276 : tmp_conninfo = walrcv_get_conninfo(wrconn);
285 276 : walrcv_get_senderinfo(wrconn, &sender_host, &sender_port);
286 276 : SpinLockAcquire(&walrcv->mutex);
287 276 : memset(walrcv->conninfo, 0, MAXCONNINFO);
288 276 : if (tmp_conninfo)
289 276 : strlcpy(walrcv->conninfo, tmp_conninfo, MAXCONNINFO);
290 :
291 276 : memset(walrcv->sender_host, 0, NI_MAXHOST);
292 276 : if (sender_host)
293 276 : strlcpy(walrcv->sender_host, sender_host, NI_MAXHOST);
294 :
295 276 : walrcv->sender_port = sender_port;
296 276 : walrcv->ready_to_display = true;
297 276 : SpinLockRelease(&walrcv->mutex);
298 :
299 276 : if (tmp_conninfo)
300 276 : pfree(tmp_conninfo);
301 :
302 276 : if (sender_host)
303 276 : pfree(sender_host);
304 :
305 276 : first_stream = true;
306 : for (;;)
307 0 : {
308 : char *primary_sysid;
309 : char standby_sysid[32];
310 : WalRcvStreamOptions options;
311 :
312 : /*
313 : * Check that we're connected to a valid server using the
314 : * IDENTIFY_SYSTEM replication command.
315 : */
316 276 : primary_sysid = walrcv_identify_system(wrconn, &primaryTLI);
317 :
318 276 : snprintf(standby_sysid, sizeof(standby_sysid), UINT64_FORMAT,
319 : GetSystemIdentifier());
320 276 : if (strcmp(primary_sysid, standby_sysid) != 0)
321 : {
322 0 : ereport(ERROR,
323 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
324 : errmsg("database system identifier differs between the primary and standby"),
325 : errdetail("The primary's identifier is %s, the standby's identifier is %s.",
326 : primary_sysid, standby_sysid)));
327 : }
328 :
329 : /*
330 : * Confirm that the current timeline of the primary is the same or
331 : * ahead of ours.
332 : */
333 276 : if (primaryTLI < startpointTLI)
334 0 : ereport(ERROR,
335 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
336 : errmsg("highest timeline %u of the primary is behind recovery timeline %u",
337 : primaryTLI, startpointTLI)));
338 :
339 : /*
340 : * Get any missing history files. We do this always, even when we're
341 : * not interested in that timeline, so that if we're promoted to
342 : * become the primary later on, we don't select the same timeline that
343 : * was already used in the current primary. This isn't bullet-proof -
344 : * you'll need some external software to manage your cluster if you
345 : * need to ensure that a unique timeline id is chosen in every case,
346 : * but let's avoid the confusion of timeline id collisions where we
347 : * can.
348 : */
349 276 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
350 :
351 : /*
352 : * Create temporary replication slot if requested, and update slot
353 : * name in shared memory. (Note the slot name cannot already be set
354 : * in this case.)
355 : */
356 276 : if (is_temp_slot)
357 : {
358 0 : snprintf(slotname, sizeof(slotname),
359 : "pg_walreceiver_%lld",
360 0 : (long long int) walrcv_get_backend_pid(wrconn));
361 :
362 0 : walrcv_create_slot(wrconn, slotname, true, false, false, 0, NULL);
363 :
364 0 : SpinLockAcquire(&walrcv->mutex);
365 0 : strlcpy(walrcv->slotname, slotname, NAMEDATALEN);
366 0 : SpinLockRelease(&walrcv->mutex);
367 : }
368 :
369 : /*
370 : * Start streaming.
371 : *
372 : * We'll try to start at the requested starting point and timeline,
373 : * even if it's different from the server's latest timeline. In case
374 : * we've already reached the end of the old timeline, the server will
375 : * finish the streaming immediately, and we will go back to await
376 : * orders from the startup process. If recovery_target_timeline is
377 : * 'latest', the startup process will scan pg_wal and find the new
378 : * history file, bump recovery target timeline, and ask us to restart
379 : * on the new timeline.
380 : */
381 276 : options.logical = false;
382 276 : options.startpoint = startpoint;
383 276 : options.slotname = slotname[0] != '\0' ? slotname : NULL;
384 276 : options.proto.physical.startpointTLI = startpointTLI;
385 276 : if (walrcv_startstreaming(wrconn, &options))
386 : {
387 272 : if (first_stream)
388 272 : ereport(LOG,
389 : (errmsg("started streaming WAL from primary at %X/%X on timeline %u",
390 : LSN_FORMAT_ARGS(startpoint), startpointTLI)));
391 : else
392 0 : ereport(LOG,
393 : (errmsg("restarted WAL streaming at %X/%X on timeline %u",
394 : LSN_FORMAT_ARGS(startpoint), startpointTLI)));
395 272 : first_stream = false;
396 :
397 : /* Initialize LogstreamResult and buffers for processing messages */
398 272 : LogstreamResult.Write = LogstreamResult.Flush = GetXLogReplayRecPtr(NULL);
399 272 : initStringInfo(&reply_message);
400 :
401 : /* Initialize nap wakeup times. */
402 272 : now = GetCurrentTimestamp();
403 1360 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
404 1088 : WalRcvComputeNextWakeup(i, now);
405 :
406 : /* Send initial reply/feedback messages. */
407 272 : XLogWalRcvSendReply(true, false);
408 272 : XLogWalRcvSendHSFeedback(true);
409 :
410 : /* Loop until end-of-streaming or error */
411 : for (;;)
412 30360 : {
413 : char *buf;
414 : int len;
415 30632 : bool endofwal = false;
416 30632 : pgsocket wait_fd = PGINVALID_SOCKET;
417 : int rc;
418 : TimestampTz nextWakeup;
419 : long nap;
420 :
421 : /*
422 : * Exit walreceiver if we're not in recovery. This should not
423 : * happen, but cross-check the status here.
424 : */
425 30632 : if (!RecoveryInProgress())
426 0 : ereport(FATAL,
427 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
428 : errmsg("cannot continue WAL streaming, recovery has already ended")));
429 :
430 : /* Process any requests or signals received recently */
431 30632 : CHECK_FOR_INTERRUPTS();
432 :
433 30632 : if (ConfigReloadPending)
434 : {
435 48 : ConfigReloadPending = false;
436 48 : ProcessConfigFile(PGC_SIGHUP);
437 : /* recompute wakeup times */
438 48 : now = GetCurrentTimestamp();
439 240 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
440 192 : WalRcvComputeNextWakeup(i, now);
441 48 : XLogWalRcvSendHSFeedback(true);
442 : }
443 :
444 : /* See if we can read data immediately */
445 30632 : len = walrcv_receive(wrconn, &buf, &wait_fd);
446 30572 : if (len != 0)
447 : {
448 : /*
449 : * Process the received data, and any subsequent data we
450 : * can read without blocking.
451 : */
452 : for (;;)
453 : {
454 207922 : if (len > 0)
455 : {
456 : /*
457 : * Something was received from primary, so adjust
458 : * the ping and terminate wakeup times.
459 : */
460 187394 : now = GetCurrentTimestamp();
461 187394 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_TERMINATE,
462 : now);
463 187394 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_PING, now);
464 187394 : XLogWalRcvProcessMsg(buf[0], &buf[1], len - 1,
465 : startpointTLI);
466 : }
467 20528 : else if (len == 0)
468 20452 : break;
469 76 : else if (len < 0)
470 : {
471 76 : ereport(LOG,
472 : (errmsg("replication terminated by primary server"),
473 : errdetail("End of WAL reached on timeline %u at %X/%X.",
474 : startpointTLI,
475 : LSN_FORMAT_ARGS(LogstreamResult.Write))));
476 76 : endofwal = true;
477 76 : break;
478 : }
479 187394 : len = walrcv_receive(wrconn, &buf, &wait_fd);
480 : }
481 :
482 : /* Let the primary know that we received some data. */
483 20528 : XLogWalRcvSendReply(false, false);
484 :
485 : /*
486 : * If we've written some records, flush them to disk and
487 : * let the startup process and primary server know about
488 : * them.
489 : */
490 20528 : XLogWalRcvFlush(false, startpointTLI);
491 : }
492 :
493 : /* Check if we need to exit the streaming loop. */
494 30568 : if (endofwal)
495 76 : break;
496 :
497 : /* Find the soonest wakeup time, to limit our nap. */
498 30492 : nextWakeup = TIMESTAMP_INFINITY;
499 152460 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
500 121968 : nextWakeup = Min(wakeup[i], nextWakeup);
501 :
502 : /* Calculate the nap time, clamping as necessary. */
503 30492 : now = GetCurrentTimestamp();
504 30492 : nap = TimestampDifferenceMilliseconds(now, nextWakeup);
505 :
506 : /*
507 : * Ideally we would reuse a WaitEventSet object repeatedly
508 : * here to avoid the overheads of WaitLatchOrSocket on epoll
509 : * systems, but we can't be sure that libpq (or any other
510 : * walreceiver implementation) has the same socket (even if
511 : * the fd is the same number, it may have been closed and
512 : * reopened since the last time). In future, if there is a
513 : * function for removing sockets from WaitEventSet, then we
514 : * could add and remove just the socket each time, potentially
515 : * avoiding some system calls.
516 : */
517 : Assert(wait_fd != PGINVALID_SOCKET);
518 30492 : rc = WaitLatchOrSocket(MyLatch,
519 : WL_EXIT_ON_PM_DEATH | WL_SOCKET_READABLE |
520 : WL_TIMEOUT | WL_LATCH_SET,
521 : wait_fd,
522 : nap,
523 : WAIT_EVENT_WAL_RECEIVER_MAIN);
524 30492 : if (rc & WL_LATCH_SET)
525 : {
526 8946 : ResetLatch(MyLatch);
527 8946 : CHECK_FOR_INTERRUPTS();
528 :
529 8814 : if (walrcv->force_reply)
530 : {
531 : /*
532 : * The recovery process has asked us to send apply
533 : * feedback now. Make sure the flag is really set to
534 : * false in shared memory before sending the reply, so
535 : * we don't miss a new request for a reply.
536 : */
537 8732 : walrcv->force_reply = false;
538 8732 : pg_memory_barrier();
539 8732 : XLogWalRcvSendReply(true, false);
540 : }
541 : }
542 30360 : if (rc & WL_TIMEOUT)
543 : {
544 : /*
545 : * We didn't receive anything new. If we haven't heard
546 : * anything from the server for more than
547 : * wal_receiver_timeout / 2, ping the server. Also, if
548 : * it's been longer than wal_receiver_status_interval
549 : * since the last update we sent, send a status update to
550 : * the primary anyway, to report any progress in applying
551 : * WAL.
552 : */
553 14 : bool requestReply = false;
554 :
555 : /*
556 : * Report pending statistics to the cumulative stats
557 : * system. This location is useful for the report as it
558 : * is not within a tight loop in the WAL receiver, to
559 : * avoid bloating pgstats with requests, while also making
560 : * sure that the reports happen each time a status update
561 : * is sent.
562 : */
563 14 : pgstat_report_wal(false);
564 :
565 : /*
566 : * Check if time since last receive from primary has
567 : * reached the configured limit.
568 : */
569 14 : now = GetCurrentTimestamp();
570 14 : if (now >= wakeup[WALRCV_WAKEUP_TERMINATE])
571 0 : ereport(ERROR,
572 : (errcode(ERRCODE_CONNECTION_FAILURE),
573 : errmsg("terminating walreceiver due to timeout")));
574 :
575 : /*
576 : * If we didn't receive anything new for half of receiver
577 : * replication timeout, then ping the server.
578 : */
579 14 : if (now >= wakeup[WALRCV_WAKEUP_PING])
580 : {
581 0 : requestReply = true;
582 0 : wakeup[WALRCV_WAKEUP_PING] = TIMESTAMP_INFINITY;
583 : }
584 :
585 14 : XLogWalRcvSendReply(requestReply, requestReply);
586 14 : XLogWalRcvSendHSFeedback(false);
587 : }
588 : }
589 :
590 : /*
591 : * The backend finished streaming. Exit streaming COPY-mode from
592 : * our side, too.
593 : */
594 76 : walrcv_endstreaming(wrconn, &primaryTLI);
595 :
596 : /*
597 : * If the server had switched to a new timeline that we didn't
598 : * know about when we began streaming, fetch its timeline history
599 : * file now.
600 : */
601 24 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
602 : }
603 : else
604 0 : ereport(LOG,
605 : (errmsg("primary server contains no more WAL on requested timeline %u",
606 : startpointTLI)));
607 :
608 : /*
609 : * End of WAL reached on the requested timeline. Close the last
610 : * segment, and await for new orders from the startup process.
611 : */
612 24 : if (recvFile >= 0)
613 : {
614 : char xlogfname[MAXFNAMELEN];
615 :
616 22 : XLogWalRcvFlush(false, startpointTLI);
617 22 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
618 22 : if (close(recvFile) != 0)
619 0 : ereport(PANIC,
620 : (errcode_for_file_access(),
621 : errmsg("could not close WAL segment %s: %m",
622 : xlogfname)));
623 :
624 : /*
625 : * Create .done file forcibly to prevent the streamed segment from
626 : * being archived later.
627 : */
628 22 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
629 22 : XLogArchiveForceDone(xlogfname);
630 : else
631 0 : XLogArchiveNotify(xlogfname);
632 : }
633 24 : recvFile = -1;
634 :
635 24 : elog(DEBUG1, "walreceiver ended streaming and awaits new instructions");
636 24 : WalRcvWaitForStartPosition(&startpoint, &startpointTLI);
637 : }
638 : /* not reached */
639 : }
640 :
641 : /*
642 : * Wait for startup process to set receiveStart and receiveStartTLI.
643 : */
644 : static void
645 24 : WalRcvWaitForStartPosition(XLogRecPtr *startpoint, TimeLineID *startpointTLI)
646 : {
647 24 : WalRcvData *walrcv = WalRcv;
648 : int state;
649 :
650 24 : SpinLockAcquire(&walrcv->mutex);
651 24 : state = walrcv->walRcvState;
652 24 : if (state != WALRCV_STREAMING)
653 : {
654 0 : SpinLockRelease(&walrcv->mutex);
655 0 : if (state == WALRCV_STOPPING)
656 0 : proc_exit(0);
657 : else
658 0 : elog(FATAL, "unexpected walreceiver state");
659 : }
660 24 : walrcv->walRcvState = WALRCV_WAITING;
661 24 : walrcv->receiveStart = InvalidXLogRecPtr;
662 24 : walrcv->receiveStartTLI = 0;
663 24 : SpinLockRelease(&walrcv->mutex);
664 :
665 24 : set_ps_display("idle");
666 :
667 : /*
668 : * nudge startup process to notice that we've stopped streaming and are
669 : * now waiting for instructions.
670 : */
671 24 : WakeupRecovery();
672 : for (;;)
673 : {
674 46 : ResetLatch(MyLatch);
675 :
676 46 : CHECK_FOR_INTERRUPTS();
677 :
678 22 : SpinLockAcquire(&walrcv->mutex);
679 : Assert(walrcv->walRcvState == WALRCV_RESTARTING ||
680 : walrcv->walRcvState == WALRCV_WAITING ||
681 : walrcv->walRcvState == WALRCV_STOPPING);
682 22 : if (walrcv->walRcvState == WALRCV_RESTARTING)
683 : {
684 : /*
685 : * No need to handle changes in primary_conninfo or
686 : * primary_slot_name here. Startup process will signal us to
687 : * terminate in case those change.
688 : */
689 0 : *startpoint = walrcv->receiveStart;
690 0 : *startpointTLI = walrcv->receiveStartTLI;
691 0 : walrcv->walRcvState = WALRCV_STREAMING;
692 0 : SpinLockRelease(&walrcv->mutex);
693 0 : break;
694 : }
695 22 : if (walrcv->walRcvState == WALRCV_STOPPING)
696 : {
697 : /*
698 : * We should've received SIGTERM if the startup process wants us
699 : * to die, but might as well check it here too.
700 : */
701 0 : SpinLockRelease(&walrcv->mutex);
702 0 : exit(1);
703 : }
704 22 : SpinLockRelease(&walrcv->mutex);
705 :
706 22 : (void) WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, 0,
707 : WAIT_EVENT_WAL_RECEIVER_WAIT_START);
708 : }
709 :
710 0 : if (update_process_title)
711 : {
712 : char activitymsg[50];
713 :
714 0 : snprintf(activitymsg, sizeof(activitymsg), "restarting at %X/%X",
715 0 : LSN_FORMAT_ARGS(*startpoint));
716 0 : set_ps_display(activitymsg);
717 : }
718 0 : }
719 :
720 : /*
721 : * Fetch any missing timeline history files between 'first' and 'last'
722 : * (inclusive) from the server.
723 : */
724 : static void
725 300 : WalRcvFetchTimeLineHistoryFiles(TimeLineID first, TimeLineID last)
726 : {
727 : TimeLineID tli;
728 :
729 646 : for (tli = first; tli <= last; tli++)
730 : {
731 : /* there's no history file for timeline 1 */
732 346 : if (tli != 1 && !existsTimeLineHistory(tli))
733 : {
734 : char *fname;
735 : char *content;
736 : int len;
737 : char expectedfname[MAXFNAMELEN];
738 :
739 22 : ereport(LOG,
740 : (errmsg("fetching timeline history file for timeline %u from primary server",
741 : tli)));
742 :
743 22 : walrcv_readtimelinehistoryfile(wrconn, tli, &fname, &content, &len);
744 :
745 : /*
746 : * Check that the filename on the primary matches what we
747 : * calculated ourselves. This is just a sanity check, it should
748 : * always match.
749 : */
750 22 : TLHistoryFileName(expectedfname, tli);
751 22 : if (strcmp(fname, expectedfname) != 0)
752 0 : ereport(ERROR,
753 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
754 : errmsg_internal("primary reported unexpected file name for timeline history file of timeline %u",
755 : tli)));
756 :
757 : /*
758 : * Write the file to pg_wal.
759 : */
760 22 : writeTimeLineHistoryFile(tli, content, len);
761 :
762 : /*
763 : * Mark the streamed history file as ready for archiving if
764 : * archive_mode is always.
765 : */
766 22 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
767 22 : XLogArchiveForceDone(fname);
768 : else
769 0 : XLogArchiveNotify(fname);
770 :
771 22 : pfree(fname);
772 22 : pfree(content);
773 : }
774 : }
775 300 : }
776 :
777 : /*
778 : * Mark us as STOPPED in shared memory at exit.
779 : */
780 : static void
781 446 : WalRcvDie(int code, Datum arg)
782 : {
783 446 : WalRcvData *walrcv = WalRcv;
784 446 : TimeLineID *startpointTLI_p = (TimeLineID *) DatumGetPointer(arg);
785 :
786 : Assert(*startpointTLI_p != 0);
787 :
788 : /* Ensure that all WAL records received are flushed to disk */
789 446 : XLogWalRcvFlush(true, *startpointTLI_p);
790 :
791 : /* Mark ourselves inactive in shared memory */
792 446 : SpinLockAcquire(&walrcv->mutex);
793 : Assert(walrcv->walRcvState == WALRCV_STREAMING ||
794 : walrcv->walRcvState == WALRCV_RESTARTING ||
795 : walrcv->walRcvState == WALRCV_STARTING ||
796 : walrcv->walRcvState == WALRCV_WAITING ||
797 : walrcv->walRcvState == WALRCV_STOPPING);
798 : Assert(walrcv->pid == MyProcPid);
799 446 : walrcv->walRcvState = WALRCV_STOPPED;
800 446 : walrcv->pid = 0;
801 446 : walrcv->procno = INVALID_PROC_NUMBER;
802 446 : walrcv->ready_to_display = false;
803 446 : SpinLockRelease(&walrcv->mutex);
804 :
805 446 : ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
806 :
807 : /* Terminate the connection gracefully. */
808 446 : if (wrconn != NULL)
809 276 : walrcv_disconnect(wrconn);
810 :
811 : /* Wake up the startup process to notice promptly that we're gone */
812 446 : WakeupRecovery();
813 446 : }
814 :
815 : /*
816 : * Accept the message from XLOG stream, and process it.
817 : */
818 : static void
819 187394 : XLogWalRcvProcessMsg(unsigned char type, char *buf, Size len, TimeLineID tli)
820 : {
821 : int hdrlen;
822 : XLogRecPtr dataStart;
823 : XLogRecPtr walEnd;
824 : TimestampTz sendTime;
825 : bool replyRequested;
826 :
827 187394 : switch (type)
828 : {
829 184854 : case 'w': /* WAL records */
830 : {
831 : StringInfoData incoming_message;
832 :
833 184854 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(int64);
834 184854 : if (len < hdrlen)
835 0 : ereport(ERROR,
836 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
837 : errmsg_internal("invalid WAL message received from primary")));
838 :
839 : /* initialize a StringInfo with the given buffer */
840 184854 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
841 :
842 : /* read the fields */
843 184854 : dataStart = pq_getmsgint64(&incoming_message);
844 184854 : walEnd = pq_getmsgint64(&incoming_message);
845 184854 : sendTime = pq_getmsgint64(&incoming_message);
846 184854 : ProcessWalSndrMessage(walEnd, sendTime);
847 :
848 184854 : buf += hdrlen;
849 184854 : len -= hdrlen;
850 184854 : XLogWalRcvWrite(buf, len, dataStart, tli);
851 184854 : break;
852 : }
853 2540 : case 'k': /* Keepalive */
854 : {
855 : StringInfoData incoming_message;
856 :
857 2540 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(char);
858 2540 : if (len != hdrlen)
859 0 : ereport(ERROR,
860 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
861 : errmsg_internal("invalid keepalive message received from primary")));
862 :
863 : /* initialize a StringInfo with the given buffer */
864 2540 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
865 :
866 : /* read the fields */
867 2540 : walEnd = pq_getmsgint64(&incoming_message);
868 2540 : sendTime = pq_getmsgint64(&incoming_message);
869 2540 : replyRequested = pq_getmsgbyte(&incoming_message);
870 :
871 2540 : ProcessWalSndrMessage(walEnd, sendTime);
872 :
873 : /* If the primary requested a reply, send one immediately */
874 2540 : if (replyRequested)
875 2540 : XLogWalRcvSendReply(true, false);
876 2540 : break;
877 : }
878 0 : default:
879 0 : ereport(ERROR,
880 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
881 : errmsg_internal("invalid replication message type %d",
882 : type)));
883 : }
884 187394 : }
885 :
886 : /*
887 : * Write XLOG data to disk.
888 : */
889 : static void
890 184854 : XLogWalRcvWrite(char *buf, Size nbytes, XLogRecPtr recptr, TimeLineID tli)
891 : {
892 : int startoff;
893 : int byteswritten;
894 : instr_time start;
895 :
896 : Assert(tli != 0);
897 :
898 370922 : while (nbytes > 0)
899 : {
900 : int segbytes;
901 :
902 : /* Close the current segment if it's completed */
903 186068 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
904 1214 : XLogWalRcvClose(recptr, tli);
905 :
906 186068 : if (recvFile < 0)
907 : {
908 : /* Create/use new log file */
909 1636 : XLByteToSeg(recptr, recvSegNo, wal_segment_size);
910 1636 : recvFile = XLogFileInit(recvSegNo, tli);
911 1636 : recvFileTLI = tli;
912 : }
913 :
914 : /* Calculate the start offset of the received logs */
915 186068 : startoff = XLogSegmentOffset(recptr, wal_segment_size);
916 :
917 186068 : if (startoff + nbytes > wal_segment_size)
918 1214 : segbytes = wal_segment_size - startoff;
919 : else
920 184854 : segbytes = nbytes;
921 :
922 : /* OK to write the logs */
923 186068 : errno = 0;
924 :
925 : /*
926 : * Measure I/O timing to write WAL data, for pg_stat_io.
927 : */
928 186068 : start = pgstat_prepare_io_time(track_wal_io_timing);
929 :
930 186068 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
931 186068 : byteswritten = pg_pwrite(recvFile, buf, segbytes, (off_t) startoff);
932 186068 : pgstat_report_wait_end();
933 :
934 186068 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
935 : IOOP_WRITE, start, 1, byteswritten);
936 :
937 186068 : if (byteswritten <= 0)
938 : {
939 : char xlogfname[MAXFNAMELEN];
940 : int save_errno;
941 :
942 : /* if write didn't set errno, assume no disk space */
943 0 : if (errno == 0)
944 0 : errno = ENOSPC;
945 :
946 0 : save_errno = errno;
947 0 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
948 0 : errno = save_errno;
949 0 : ereport(PANIC,
950 : (errcode_for_file_access(),
951 : errmsg("could not write to WAL segment %s "
952 : "at offset %d, length %lu: %m",
953 : xlogfname, startoff, (unsigned long) segbytes)));
954 : }
955 :
956 : /* Update state for write */
957 186068 : recptr += byteswritten;
958 :
959 186068 : nbytes -= byteswritten;
960 186068 : buf += byteswritten;
961 :
962 186068 : LogstreamResult.Write = recptr;
963 : }
964 :
965 : /* Update shared-memory status */
966 184854 : pg_atomic_write_u64(&WalRcv->writtenUpto, LogstreamResult.Write);
967 :
968 : /*
969 : * Close the current segment if it's fully written up in the last cycle of
970 : * the loop, to create its archive notification file soon. Otherwise WAL
971 : * archiving of the segment will be delayed until any data in the next
972 : * segment is received and written.
973 : */
974 184854 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
975 190 : XLogWalRcvClose(recptr, tli);
976 184854 : }
977 :
978 : /*
979 : * Flush the log to disk.
980 : *
981 : * If we're in the midst of dying, it's unwise to do anything that might throw
982 : * an error, so we skip sending a reply in that case.
983 : */
984 : static void
985 22400 : XLogWalRcvFlush(bool dying, TimeLineID tli)
986 : {
987 : Assert(tli != 0);
988 :
989 22400 : if (LogstreamResult.Flush < LogstreamResult.Write)
990 : {
991 21678 : WalRcvData *walrcv = WalRcv;
992 :
993 21678 : issue_xlog_fsync(recvFile, recvSegNo, tli);
994 :
995 21678 : LogstreamResult.Flush = LogstreamResult.Write;
996 :
997 : /* Update shared-memory status */
998 21678 : SpinLockAcquire(&walrcv->mutex);
999 21678 : if (walrcv->flushedUpto < LogstreamResult.Flush)
1000 : {
1001 21678 : walrcv->latestChunkStart = walrcv->flushedUpto;
1002 21678 : walrcv->flushedUpto = LogstreamResult.Flush;
1003 21678 : walrcv->receivedTLI = tli;
1004 : }
1005 21678 : SpinLockRelease(&walrcv->mutex);
1006 :
1007 : /* Signal the startup process and walsender that new WAL has arrived */
1008 21678 : WakeupRecovery();
1009 21678 : if (AllowCascadeReplication())
1010 21678 : WalSndWakeup(true, false);
1011 :
1012 : /* Report XLOG streaming progress in PS display */
1013 21678 : if (update_process_title)
1014 : {
1015 : char activitymsg[50];
1016 :
1017 21678 : snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%X",
1018 21678 : LSN_FORMAT_ARGS(LogstreamResult.Write));
1019 21678 : set_ps_display(activitymsg);
1020 : }
1021 :
1022 : /* Also let the primary know that we made some progress */
1023 21678 : if (!dying)
1024 : {
1025 21674 : XLogWalRcvSendReply(false, false);
1026 21674 : XLogWalRcvSendHSFeedback(false);
1027 : }
1028 : }
1029 22400 : }
1030 :
1031 : /*
1032 : * Close the current segment.
1033 : *
1034 : * Flush the segment to disk before closing it. Otherwise we have to
1035 : * reopen and fsync it later.
1036 : *
1037 : * Create an archive notification file since the segment is known completed.
1038 : */
1039 : static void
1040 1404 : XLogWalRcvClose(XLogRecPtr recptr, TimeLineID tli)
1041 : {
1042 : char xlogfname[MAXFNAMELEN];
1043 :
1044 : Assert(recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size));
1045 : Assert(tli != 0);
1046 :
1047 : /*
1048 : * fsync() and close current file before we switch to next one. We would
1049 : * otherwise have to reopen this file to fsync it later
1050 : */
1051 1404 : XLogWalRcvFlush(false, tli);
1052 :
1053 1404 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
1054 :
1055 : /*
1056 : * XLOG segment files will be re-read by recovery in startup process soon,
1057 : * so we don't advise the OS to release cache pages associated with the
1058 : * file like XLogFileClose() does.
1059 : */
1060 1404 : if (close(recvFile) != 0)
1061 0 : ereport(PANIC,
1062 : (errcode_for_file_access(),
1063 : errmsg("could not close WAL segment %s: %m",
1064 : xlogfname)));
1065 :
1066 : /*
1067 : * Create .done file forcibly to prevent the streamed segment from being
1068 : * archived later.
1069 : */
1070 1404 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
1071 1404 : XLogArchiveForceDone(xlogfname);
1072 : else
1073 0 : XLogArchiveNotify(xlogfname);
1074 :
1075 1404 : recvFile = -1;
1076 1404 : }
1077 :
1078 : /*
1079 : * Send reply message to primary, indicating our current WAL locations, oldest
1080 : * xmin and the current time.
1081 : *
1082 : * If 'force' is not set, the message is only sent if enough time has
1083 : * passed since last status update to reach wal_receiver_status_interval.
1084 : * If wal_receiver_status_interval is disabled altogether and 'force' is
1085 : * false, this is a no-op.
1086 : *
1087 : * If 'requestReply' is true, requests the server to reply immediately upon
1088 : * receiving this message. This is used for heartbeats, when approaching
1089 : * wal_receiver_timeout.
1090 : */
1091 : static void
1092 53760 : XLogWalRcvSendReply(bool force, bool requestReply)
1093 : {
1094 : static XLogRecPtr writePtr = 0;
1095 : static XLogRecPtr flushPtr = 0;
1096 : XLogRecPtr applyPtr;
1097 : TimestampTz now;
1098 :
1099 : /*
1100 : * If the user doesn't want status to be reported to the primary, be sure
1101 : * to exit before doing anything at all.
1102 : */
1103 53760 : if (!force && wal_receiver_status_interval <= 0)
1104 0 : return;
1105 :
1106 : /* Get current timestamp. */
1107 53760 : now = GetCurrentTimestamp();
1108 :
1109 : /*
1110 : * We can compare the write and flush positions to the last message we
1111 : * sent without taking any lock, but the apply position requires a spin
1112 : * lock, so we don't check that unless something else has changed or 10
1113 : * seconds have passed. This means that the apply WAL location will
1114 : * appear, from the primary's point of view, to lag slightly, but since
1115 : * this is only for reporting purposes and only on idle systems, that's
1116 : * probably OK.
1117 : */
1118 53760 : if (!force
1119 42216 : && writePtr == LogstreamResult.Write
1120 20496 : && flushPtr == LogstreamResult.Flush
1121 226 : && now < wakeup[WALRCV_WAKEUP_REPLY])
1122 222 : return;
1123 :
1124 : /* Make sure we wake up when it's time to send another reply. */
1125 53538 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_REPLY, now);
1126 :
1127 : /* Construct a new message */
1128 53538 : writePtr = LogstreamResult.Write;
1129 53538 : flushPtr = LogstreamResult.Flush;
1130 53538 : applyPtr = GetXLogReplayRecPtr(NULL);
1131 :
1132 53538 : resetStringInfo(&reply_message);
1133 53538 : pq_sendbyte(&reply_message, 'r');
1134 53538 : pq_sendint64(&reply_message, writePtr);
1135 53538 : pq_sendint64(&reply_message, flushPtr);
1136 53538 : pq_sendint64(&reply_message, applyPtr);
1137 53538 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1138 53538 : pq_sendbyte(&reply_message, requestReply ? 1 : 0);
1139 :
1140 : /* Send it */
1141 53538 : elog(DEBUG2, "sending write %X/%X flush %X/%X apply %X/%X%s",
1142 : LSN_FORMAT_ARGS(writePtr),
1143 : LSN_FORMAT_ARGS(flushPtr),
1144 : LSN_FORMAT_ARGS(applyPtr),
1145 : requestReply ? " (reply requested)" : "");
1146 :
1147 53538 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1148 : }
1149 :
1150 : /*
1151 : * Send hot standby feedback message to primary, plus the current time,
1152 : * in case they don't have a watch.
1153 : *
1154 : * If the user disables feedback, send one final message to tell sender
1155 : * to forget about the xmin on this standby. We also send this message
1156 : * on first connect because a previous connection might have set xmin
1157 : * on a replication slot. (If we're not using a slot it's harmless to
1158 : * send a feedback message explicitly setting InvalidTransactionId).
1159 : */
1160 : static void
1161 22008 : XLogWalRcvSendHSFeedback(bool immed)
1162 : {
1163 : TimestampTz now;
1164 : FullTransactionId nextFullXid;
1165 : TransactionId nextXid;
1166 : uint32 xmin_epoch,
1167 : catalog_xmin_epoch;
1168 : TransactionId xmin,
1169 : catalog_xmin;
1170 :
1171 : /* initially true so we always send at least one feedback message */
1172 : static bool primary_has_standby_xmin = true;
1173 :
1174 : /*
1175 : * If the user doesn't want status to be reported to the primary, be sure
1176 : * to exit before doing anything at all.
1177 : */
1178 22008 : if ((wal_receiver_status_interval <= 0 || !hot_standby_feedback) &&
1179 21560 : !primary_has_standby_xmin)
1180 21692 : return;
1181 :
1182 : /* Get current timestamp. */
1183 676 : now = GetCurrentTimestamp();
1184 :
1185 : /* Send feedback at most once per wal_receiver_status_interval. */
1186 676 : if (!immed && now < wakeup[WALRCV_WAKEUP_HSFEEDBACK])
1187 358 : return;
1188 :
1189 : /* Make sure we wake up when it's time to send feedback again. */
1190 318 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_HSFEEDBACK, now);
1191 :
1192 : /*
1193 : * If Hot Standby is not yet accepting connections there is nothing to
1194 : * send. Check this after the interval has expired to reduce number of
1195 : * calls.
1196 : *
1197 : * Bailing out here also ensures that we don't send feedback until we've
1198 : * read our own replication slot state, so we don't tell the primary to
1199 : * discard needed xmin or catalog_xmin from any slots that may exist on
1200 : * this replica.
1201 : */
1202 318 : if (!HotStandbyActive())
1203 2 : return;
1204 :
1205 : /*
1206 : * Make the expensive call to get the oldest xmin once we are certain
1207 : * everything else has been checked.
1208 : */
1209 316 : if (hot_standby_feedback)
1210 : {
1211 92 : GetReplicationHorizons(&xmin, &catalog_xmin);
1212 : }
1213 : else
1214 : {
1215 224 : xmin = InvalidTransactionId;
1216 224 : catalog_xmin = InvalidTransactionId;
1217 : }
1218 :
1219 : /*
1220 : * Get epoch and adjust if nextXid and oldestXmin are different sides of
1221 : * the epoch boundary.
1222 : */
1223 316 : nextFullXid = ReadNextFullTransactionId();
1224 316 : nextXid = XidFromFullTransactionId(nextFullXid);
1225 316 : xmin_epoch = EpochFromFullTransactionId(nextFullXid);
1226 316 : catalog_xmin_epoch = xmin_epoch;
1227 316 : if (nextXid < xmin)
1228 0 : xmin_epoch--;
1229 316 : if (nextXid < catalog_xmin)
1230 0 : catalog_xmin_epoch--;
1231 :
1232 316 : elog(DEBUG2, "sending hot standby feedback xmin %u epoch %u catalog_xmin %u catalog_xmin_epoch %u",
1233 : xmin, xmin_epoch, catalog_xmin, catalog_xmin_epoch);
1234 :
1235 : /* Construct the message and send it. */
1236 316 : resetStringInfo(&reply_message);
1237 316 : pq_sendbyte(&reply_message, 'h');
1238 316 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1239 316 : pq_sendint32(&reply_message, xmin);
1240 316 : pq_sendint32(&reply_message, xmin_epoch);
1241 316 : pq_sendint32(&reply_message, catalog_xmin);
1242 316 : pq_sendint32(&reply_message, catalog_xmin_epoch);
1243 316 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1244 316 : if (TransactionIdIsValid(xmin) || TransactionIdIsValid(catalog_xmin))
1245 92 : primary_has_standby_xmin = true;
1246 : else
1247 224 : primary_has_standby_xmin = false;
1248 : }
1249 :
1250 : /*
1251 : * Update shared memory status upon receiving a message from primary.
1252 : *
1253 : * 'walEnd' and 'sendTime' are the end-of-WAL and timestamp of the latest
1254 : * message, reported by primary.
1255 : */
1256 : static void
1257 187394 : ProcessWalSndrMessage(XLogRecPtr walEnd, TimestampTz sendTime)
1258 : {
1259 187394 : WalRcvData *walrcv = WalRcv;
1260 187394 : TimestampTz lastMsgReceiptTime = GetCurrentTimestamp();
1261 :
1262 : /* Update shared-memory status */
1263 187394 : SpinLockAcquire(&walrcv->mutex);
1264 187394 : if (walrcv->latestWalEnd < walEnd)
1265 16462 : walrcv->latestWalEndTime = sendTime;
1266 187394 : walrcv->latestWalEnd = walEnd;
1267 187394 : walrcv->lastMsgSendTime = sendTime;
1268 187394 : walrcv->lastMsgReceiptTime = lastMsgReceiptTime;
1269 187394 : SpinLockRelease(&walrcv->mutex);
1270 :
1271 187394 : if (message_level_is_interesting(DEBUG2))
1272 : {
1273 : char *sendtime;
1274 : char *receipttime;
1275 : int applyDelay;
1276 :
1277 : /* Copy because timestamptz_to_str returns a static buffer */
1278 772 : sendtime = pstrdup(timestamptz_to_str(sendTime));
1279 772 : receipttime = pstrdup(timestamptz_to_str(lastMsgReceiptTime));
1280 772 : applyDelay = GetReplicationApplyDelay();
1281 :
1282 : /* apply delay is not available */
1283 772 : if (applyDelay == -1)
1284 2 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay (N/A) transfer latency %d ms",
1285 : sendtime,
1286 : receipttime,
1287 : GetReplicationTransferLatency());
1288 : else
1289 770 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay %d ms transfer latency %d ms",
1290 : sendtime,
1291 : receipttime,
1292 : applyDelay,
1293 : GetReplicationTransferLatency());
1294 :
1295 772 : pfree(sendtime);
1296 772 : pfree(receipttime);
1297 : }
1298 187394 : }
1299 :
1300 : /*
1301 : * Compute the next wakeup time for a given wakeup reason. Can be called to
1302 : * initialize a wakeup time, to adjust it for the next wakeup, or to
1303 : * reinitialize it when GUCs have changed. We ask the caller to pass in the
1304 : * value of "now" because this frequently avoids multiple calls of
1305 : * GetCurrentTimestamp(). It had better be a reasonably up-to-date value
1306 : * though.
1307 : */
1308 : static void
1309 429924 : WalRcvComputeNextWakeup(WalRcvWakeupReason reason, TimestampTz now)
1310 : {
1311 429924 : switch (reason)
1312 : {
1313 187714 : case WALRCV_WAKEUP_TERMINATE:
1314 187714 : if (wal_receiver_timeout <= 0)
1315 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1316 : else
1317 187714 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout);
1318 187714 : break;
1319 187714 : case WALRCV_WAKEUP_PING:
1320 187714 : if (wal_receiver_timeout <= 0)
1321 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1322 : else
1323 187714 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout / 2);
1324 187714 : break;
1325 638 : case WALRCV_WAKEUP_HSFEEDBACK:
1326 638 : if (!hot_standby_feedback || wal_receiver_status_interval <= 0)
1327 464 : wakeup[reason] = TIMESTAMP_INFINITY;
1328 : else
1329 174 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1330 638 : break;
1331 53858 : case WALRCV_WAKEUP_REPLY:
1332 53858 : if (wal_receiver_status_interval <= 0)
1333 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1334 : else
1335 53858 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1336 53858 : break;
1337 : /* there's intentionally no default: here */
1338 : }
1339 429924 : }
1340 :
1341 : /*
1342 : * Wake up the walreceiver main loop.
1343 : *
1344 : * This is called by the startup process whenever interesting xlog records
1345 : * are applied, so that walreceiver can check if it needs to send an apply
1346 : * notification back to the primary which may be waiting in a COMMIT with
1347 : * synchronous_commit = remote_apply.
1348 : */
1349 : void
1350 8650 : WalRcvForceReply(void)
1351 : {
1352 : ProcNumber procno;
1353 :
1354 8650 : WalRcv->force_reply = true;
1355 : /* fetching the proc number is probably atomic, but don't rely on it */
1356 8650 : SpinLockAcquire(&WalRcv->mutex);
1357 8650 : procno = WalRcv->procno;
1358 8650 : SpinLockRelease(&WalRcv->mutex);
1359 8650 : if (procno != INVALID_PROC_NUMBER)
1360 8340 : SetLatch(&GetPGProcByNumber(procno)->procLatch);
1361 8650 : }
1362 :
1363 : /*
1364 : * Return a string constant representing the state. This is used
1365 : * in system functions and views, and should *not* be translated.
1366 : */
1367 : static const char *
1368 0 : WalRcvGetStateString(WalRcvState state)
1369 : {
1370 0 : switch (state)
1371 : {
1372 0 : case WALRCV_STOPPED:
1373 0 : return "stopped";
1374 0 : case WALRCV_STARTING:
1375 0 : return "starting";
1376 0 : case WALRCV_STREAMING:
1377 0 : return "streaming";
1378 0 : case WALRCV_WAITING:
1379 0 : return "waiting";
1380 0 : case WALRCV_RESTARTING:
1381 0 : return "restarting";
1382 0 : case WALRCV_STOPPING:
1383 0 : return "stopping";
1384 : }
1385 0 : return "UNKNOWN";
1386 : }
1387 :
1388 : /*
1389 : * Returns activity of WAL receiver, including pid, state and xlog locations
1390 : * received from the WAL sender of another server.
1391 : */
1392 : Datum
1393 6 : pg_stat_get_wal_receiver(PG_FUNCTION_ARGS)
1394 : {
1395 : TupleDesc tupdesc;
1396 : Datum *values;
1397 : bool *nulls;
1398 : int pid;
1399 : bool ready_to_display;
1400 : WalRcvState state;
1401 : XLogRecPtr receive_start_lsn;
1402 : TimeLineID receive_start_tli;
1403 : XLogRecPtr written_lsn;
1404 : XLogRecPtr flushed_lsn;
1405 : TimeLineID received_tli;
1406 : TimestampTz last_send_time;
1407 : TimestampTz last_receipt_time;
1408 : XLogRecPtr latest_end_lsn;
1409 : TimestampTz latest_end_time;
1410 : char sender_host[NI_MAXHOST];
1411 6 : int sender_port = 0;
1412 : char slotname[NAMEDATALEN];
1413 : char conninfo[MAXCONNINFO];
1414 :
1415 : /* Take a lock to ensure value consistency */
1416 6 : SpinLockAcquire(&WalRcv->mutex);
1417 6 : pid = (int) WalRcv->pid;
1418 6 : ready_to_display = WalRcv->ready_to_display;
1419 6 : state = WalRcv->walRcvState;
1420 6 : receive_start_lsn = WalRcv->receiveStart;
1421 6 : receive_start_tli = WalRcv->receiveStartTLI;
1422 6 : flushed_lsn = WalRcv->flushedUpto;
1423 6 : received_tli = WalRcv->receivedTLI;
1424 6 : last_send_time = WalRcv->lastMsgSendTime;
1425 6 : last_receipt_time = WalRcv->lastMsgReceiptTime;
1426 6 : latest_end_lsn = WalRcv->latestWalEnd;
1427 6 : latest_end_time = WalRcv->latestWalEndTime;
1428 6 : strlcpy(slotname, WalRcv->slotname, sizeof(slotname));
1429 6 : strlcpy(sender_host, WalRcv->sender_host, sizeof(sender_host));
1430 6 : sender_port = WalRcv->sender_port;
1431 6 : strlcpy(conninfo, WalRcv->conninfo, sizeof(conninfo));
1432 6 : SpinLockRelease(&WalRcv->mutex);
1433 :
1434 : /*
1435 : * No WAL receiver (or not ready yet), just return a tuple with NULL
1436 : * values
1437 : */
1438 6 : if (pid == 0 || !ready_to_display)
1439 6 : PG_RETURN_NULL();
1440 :
1441 : /*
1442 : * Read "writtenUpto" without holding a spinlock. Note that it may not be
1443 : * consistent with the other shared variables of the WAL receiver
1444 : * protected by a spinlock, but this should not be used for data integrity
1445 : * checks.
1446 : */
1447 0 : written_lsn = pg_atomic_read_u64(&WalRcv->writtenUpto);
1448 :
1449 : /* determine result type */
1450 0 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1451 0 : elog(ERROR, "return type must be a row type");
1452 :
1453 0 : values = palloc0(sizeof(Datum) * tupdesc->natts);
1454 0 : nulls = palloc0(sizeof(bool) * tupdesc->natts);
1455 :
1456 : /* Fetch values */
1457 0 : values[0] = Int32GetDatum(pid);
1458 :
1459 0 : if (!has_privs_of_role(GetUserId(), ROLE_PG_READ_ALL_STATS))
1460 : {
1461 : /*
1462 : * Only superusers and roles with privileges of pg_read_all_stats can
1463 : * see details. Other users only get the pid value to know whether it
1464 : * is a WAL receiver, but no details.
1465 : */
1466 0 : memset(&nulls[1], true, sizeof(bool) * (tupdesc->natts - 1));
1467 : }
1468 : else
1469 : {
1470 0 : values[1] = CStringGetTextDatum(WalRcvGetStateString(state));
1471 :
1472 0 : if (XLogRecPtrIsInvalid(receive_start_lsn))
1473 0 : nulls[2] = true;
1474 : else
1475 0 : values[2] = LSNGetDatum(receive_start_lsn);
1476 0 : values[3] = Int32GetDatum(receive_start_tli);
1477 0 : if (XLogRecPtrIsInvalid(written_lsn))
1478 0 : nulls[4] = true;
1479 : else
1480 0 : values[4] = LSNGetDatum(written_lsn);
1481 0 : if (XLogRecPtrIsInvalid(flushed_lsn))
1482 0 : nulls[5] = true;
1483 : else
1484 0 : values[5] = LSNGetDatum(flushed_lsn);
1485 0 : values[6] = Int32GetDatum(received_tli);
1486 0 : if (last_send_time == 0)
1487 0 : nulls[7] = true;
1488 : else
1489 0 : values[7] = TimestampTzGetDatum(last_send_time);
1490 0 : if (last_receipt_time == 0)
1491 0 : nulls[8] = true;
1492 : else
1493 0 : values[8] = TimestampTzGetDatum(last_receipt_time);
1494 0 : if (XLogRecPtrIsInvalid(latest_end_lsn))
1495 0 : nulls[9] = true;
1496 : else
1497 0 : values[9] = LSNGetDatum(latest_end_lsn);
1498 0 : if (latest_end_time == 0)
1499 0 : nulls[10] = true;
1500 : else
1501 0 : values[10] = TimestampTzGetDatum(latest_end_time);
1502 0 : if (*slotname == '\0')
1503 0 : nulls[11] = true;
1504 : else
1505 0 : values[11] = CStringGetTextDatum(slotname);
1506 0 : if (*sender_host == '\0')
1507 0 : nulls[12] = true;
1508 : else
1509 0 : values[12] = CStringGetTextDatum(sender_host);
1510 0 : if (sender_port == 0)
1511 0 : nulls[13] = true;
1512 : else
1513 0 : values[13] = Int32GetDatum(sender_port);
1514 0 : if (*conninfo == '\0')
1515 0 : nulls[14] = true;
1516 : else
1517 0 : values[14] = CStringGetTextDatum(conninfo);
1518 : }
1519 :
1520 : /* Returns the record as Datum */
1521 0 : PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
1522 : }
|
