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
154 261 : 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;
167 261 : char *sender_host = NULL;
168 261 : int sender_port = 0;
169 : char *appname;
170 :
171 : Assert(startup_data_len == 0);
172 :
173 261 : 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 : */
179 261 : 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 : */
189 261 : SpinLockAcquire(&walrcv->mutex);
190 : Assert(walrcv->pid == 0);
191 261 : 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 :
204 258 : case WALRCV_STARTING:
205 : /* The usual case */
206 258 : 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 */
218 258 : walrcv->pid = MyProcPid;
219 258 : walrcv->walRcvState = WALRCV_CONNECTING;
220 :
221 : /* Fetch information required to start streaming */
222 258 : walrcv->ready_to_display = false;
223 258 : strlcpy(conninfo, walrcv->conninfo, MAXCONNINFO);
224 258 : strlcpy(slotname, walrcv->slotname, NAMEDATALEN);
225 258 : is_temp_slot = walrcv->is_temp_slot;
226 258 : startpoint = walrcv->receiveStart;
227 258 : 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 */
236 258 : now = GetCurrentTimestamp();
237 258 : walrcv->lastMsgSendTime =
238 258 : walrcv->lastMsgReceiptTime = walrcv->latestWalEndTime = now;
239 :
240 : /* Report our proc number so that others can wake us up */
241 258 : walrcv->procno = MyProcNumber;
242 :
243 258 : SpinLockRelease(&walrcv->mutex);
244 :
245 : /* Arrange to clean up at walreceiver exit */
246 258 : on_shmem_exit(WalRcvDie, PointerGetDatum(&startpointTLI));
247 :
248 : /* Properly accept or ignore signals the postmaster might send us */
249 258 : pqsignal(SIGHUP, SignalHandlerForConfigReload); /* set flag to read config
250 : * file */
251 258 : pqsignal(SIGINT, PG_SIG_IGN);
252 258 : pqsignal(SIGTERM, die); /* request shutdown */
253 : /* SIGQUIT handler was already set up by InitPostmasterChild */
254 258 : pqsignal(SIGALRM, PG_SIG_IGN);
255 258 : pqsignal(SIGPIPE, PG_SIG_IGN);
256 258 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
257 258 : pqsignal(SIGUSR2, PG_SIG_IGN);
258 :
259 : /* Reset some signals that are accepted by postmaster but not here */
260 258 : pqsignal(SIGCHLD, PG_SIG_DFL);
261 :
262 : /* Load the libpq-specific functions */
263 258 : load_file("libpqwalreceiver", false);
264 258 : if (WalReceiverFunctions == NULL)
265 0 : elog(ERROR, "libpqwalreceiver didn't initialize correctly");
266 :
267 : /* Unblock signals (they were blocked when the postmaster forked us) */
268 258 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
269 :
270 : /*
271 : * Switch the WAL receiver state as ready for display before doing a
272 : * connection attempt, so as its connecting state is visible before
273 : * attempting to contact the primary server. Note that this resets the
274 : * original conninfo, sender_port and sender_host, for security. These
275 : * fields are filled once the connection is fully established.
276 : */
277 258 : SpinLockAcquire(&walrcv->mutex);
278 258 : memset(walrcv->conninfo, 0, MAXCONNINFO);
279 258 : memset(walrcv->sender_host, 0, NI_MAXHOST);
280 258 : walrcv->sender_port = 0;
281 258 : walrcv->ready_to_display = true;
282 258 : SpinLockRelease(&walrcv->mutex);
283 :
284 : /* Establish the connection to the primary for XLOG streaming */
285 258 : appname = cluster_name[0] ? cluster_name : "walreceiver";
286 258 : wrconn = walrcv_connect(conninfo, true, false, false, appname, &err);
287 258 : if (!wrconn)
288 105 : ereport(ERROR,
289 : (errcode(ERRCODE_CONNECTION_FAILURE),
290 : errmsg("streaming replication receiver \"%s\" could not connect to the primary server: %s",
291 : appname, err)));
292 :
293 : /*
294 : * Save user-visible connection string, now that the connection has been
295 : * achieved.
296 : */
297 153 : tmp_conninfo = walrcv_get_conninfo(wrconn);
298 153 : walrcv_get_senderinfo(wrconn, &sender_host, &sender_port);
299 153 : SpinLockAcquire(&walrcv->mutex);
300 153 : if (tmp_conninfo)
301 153 : strlcpy(walrcv->conninfo, tmp_conninfo, MAXCONNINFO);
302 153 : if (sender_host)
303 153 : strlcpy(walrcv->sender_host, sender_host, NI_MAXHOST);
304 153 : walrcv->sender_port = sender_port;
305 153 : SpinLockRelease(&walrcv->mutex);
306 :
307 153 : if (tmp_conninfo)
308 153 : pfree(tmp_conninfo);
309 :
310 153 : if (sender_host)
311 153 : pfree(sender_host);
312 :
313 : /* Initialize buffers for processing messages */
314 153 : initStringInfo(&reply_message);
315 :
316 153 : first_stream = true;
317 : for (;;)
318 13 : {
319 : char *primary_sysid;
320 : char standby_sysid[32];
321 : WalRcvStreamOptions options;
322 :
323 : /*
324 : * Check that we're connected to a valid server using the
325 : * IDENTIFY_SYSTEM replication command.
326 : */
327 166 : primary_sysid = walrcv_identify_system(wrconn, &primaryTLI);
328 :
329 166 : snprintf(standby_sysid, sizeof(standby_sysid), UINT64_FORMAT,
330 : GetSystemIdentifier());
331 166 : if (strcmp(primary_sysid, standby_sysid) != 0)
332 : {
333 0 : ereport(ERROR,
334 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
335 : errmsg("database system identifier differs between the primary and standby"),
336 : errdetail("The primary's identifier is %s, the standby's identifier is %s.",
337 : primary_sysid, standby_sysid)));
338 : }
339 166 : pfree(primary_sysid);
340 :
341 : /*
342 : * Confirm that the current timeline of the primary is the same or
343 : * ahead of ours.
344 : */
345 166 : if (primaryTLI < startpointTLI)
346 0 : ereport(ERROR,
347 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
348 : errmsg("highest timeline %u of the primary is behind recovery timeline %u",
349 : primaryTLI, startpointTLI)));
350 :
351 : /*
352 : * Get any missing history files. We do this always, even when we're
353 : * not interested in that timeline, so that if we're promoted to
354 : * become the primary later on, we don't select the same timeline that
355 : * was already used in the current primary. This isn't bullet-proof -
356 : * you'll need some external software to manage your cluster if you
357 : * need to ensure that a unique timeline id is chosen in every case,
358 : * but let's avoid the confusion of timeline id collisions where we
359 : * can.
360 : */
361 166 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
362 :
363 : /*
364 : * Create temporary replication slot if requested, and update slot
365 : * name in shared memory. (Note the slot name cannot already be set
366 : * in this case.)
367 : */
368 166 : if (is_temp_slot)
369 : {
370 0 : snprintf(slotname, sizeof(slotname),
371 : "pg_walreceiver_%lld",
372 0 : (long long int) walrcv_get_backend_pid(wrconn));
373 :
374 0 : walrcv_create_slot(wrconn, slotname, true, false, false, 0, NULL);
375 :
376 0 : SpinLockAcquire(&walrcv->mutex);
377 0 : strlcpy(walrcv->slotname, slotname, NAMEDATALEN);
378 0 : SpinLockRelease(&walrcv->mutex);
379 : }
380 :
381 : /*
382 : * Start streaming.
383 : *
384 : * We'll try to start at the requested starting point and timeline,
385 : * even if it's different from the server's latest timeline. In case
386 : * we've already reached the end of the old timeline, the server will
387 : * finish the streaming immediately, and we will go back to await
388 : * orders from the startup process. If recovery_target_timeline is
389 : * 'latest', the startup process will scan pg_wal and find the new
390 : * history file, bump recovery target timeline, and ask us to restart
391 : * on the new timeline.
392 : */
393 166 : options.logical = false;
394 166 : options.startpoint = startpoint;
395 166 : options.slotname = slotname[0] != '\0' ? slotname : NULL;
396 166 : options.proto.physical.startpointTLI = startpointTLI;
397 166 : if (walrcv_startstreaming(wrconn, &options))
398 : {
399 165 : if (first_stream)
400 152 : ereport(LOG,
401 : errmsg("started streaming WAL from primary at %X/%08X on timeline %u",
402 : LSN_FORMAT_ARGS(startpoint), startpointTLI));
403 : else
404 13 : ereport(LOG,
405 : errmsg("restarted WAL streaming at %X/%08X on timeline %u",
406 : LSN_FORMAT_ARGS(startpoint), startpointTLI));
407 165 : first_stream = false;
408 :
409 : /*
410 : * Switch to STREAMING after a successful connection if current
411 : * state is CONNECTING. This switch happens after an initial
412 : * startup, or after a restart as determined by
413 : * WalRcvWaitForStartPosition().
414 : */
415 165 : SpinLockAcquire(&walrcv->mutex);
416 165 : if (walrcv->walRcvState == WALRCV_CONNECTING)
417 165 : walrcv->walRcvState = WALRCV_STREAMING;
418 165 : SpinLockRelease(&walrcv->mutex);
419 :
420 : /* Initialize LogstreamResult for processing messages */
421 165 : LogstreamResult.Write = LogstreamResult.Flush = GetXLogReplayRecPtr(NULL);
422 :
423 : /* Initialize nap wakeup times. */
424 165 : now = GetCurrentTimestamp();
425 825 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
426 660 : WalRcvComputeNextWakeup(i, now);
427 :
428 : /* Send initial reply/feedback messages. */
429 165 : XLogWalRcvSendReply(true, false, false);
430 165 : XLogWalRcvSendHSFeedback(true);
431 :
432 : /* Loop until end-of-streaming or error */
433 : for (;;)
434 56631 : {
435 : char *buf;
436 : int len;
437 56796 : bool endofwal = false;
438 56796 : pgsocket wait_fd = PGINVALID_SOCKET;
439 : int rc;
440 : TimestampTz nextWakeup;
441 : long nap;
442 :
443 : /*
444 : * Exit walreceiver if we're not in recovery. This should not
445 : * happen, but cross-check the status here.
446 : */
447 56796 : if (!RecoveryInProgress())
448 0 : ereport(FATAL,
449 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
450 : errmsg("cannot continue WAL streaming, recovery has already ended")));
451 :
452 : /* Process any requests or signals received recently */
453 56796 : CHECK_FOR_INTERRUPTS();
454 :
455 56796 : if (ConfigReloadPending)
456 : {
457 26 : ConfigReloadPending = false;
458 26 : ProcessConfigFile(PGC_SIGHUP);
459 : /* recompute wakeup times */
460 26 : now = GetCurrentTimestamp();
461 130 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
462 104 : WalRcvComputeNextWakeup(i, now);
463 26 : XLogWalRcvSendHSFeedback(true);
464 : }
465 :
466 : /* See if we can read data immediately */
467 56796 : len = walrcv_receive(wrconn, &buf, &wait_fd);
468 56767 : if (len != 0)
469 : {
470 : /*
471 : * Process the received data, and any subsequent data we
472 : * can read without blocking.
473 : */
474 : for (;;)
475 : {
476 147443 : if (len > 0)
477 : {
478 : /*
479 : * Something was received from primary, so adjust
480 : * the ping and terminate wakeup times.
481 : */
482 107472 : now = GetCurrentTimestamp();
483 107472 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_TERMINATE,
484 : now);
485 107472 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_PING, now);
486 107472 : XLogWalRcvProcessMsg(buf[0], &buf[1], len - 1,
487 : startpointTLI);
488 : }
489 39971 : else if (len == 0)
490 39924 : break;
491 47 : else if (len < 0)
492 : {
493 47 : ereport(LOG,
494 : (errmsg("replication terminated by primary server"),
495 : errdetail("End of WAL reached on timeline %u at %X/%08X.",
496 : startpointTLI,
497 : LSN_FORMAT_ARGS(LogstreamResult.Write))));
498 47 : endofwal = true;
499 47 : break;
500 : }
501 107472 : len = walrcv_receive(wrconn, &buf, &wait_fd);
502 : }
503 :
504 : /* Let the primary know that we received some data. */
505 39971 : XLogWalRcvSendReply(false, false, false);
506 :
507 : /*
508 : * If we've written some records, flush them to disk and
509 : * let the startup process and primary server know about
510 : * them.
511 : */
512 39970 : XLogWalRcvFlush(false, startpointTLI);
513 : }
514 :
515 : /* Check if we need to exit the streaming loop. */
516 56765 : if (endofwal)
517 46 : break;
518 :
519 : /* Find the soonest wakeup time, to limit our nap. */
520 56719 : nextWakeup = TIMESTAMP_INFINITY;
521 283595 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
522 226876 : nextWakeup = Min(wakeup[i], nextWakeup);
523 :
524 : /* Calculate the nap time, clamping as necessary. */
525 56719 : now = GetCurrentTimestamp();
526 56719 : nap = TimestampDifferenceMilliseconds(now, nextWakeup);
527 :
528 : /*
529 : * Ideally we would reuse a WaitEventSet object repeatedly
530 : * here to avoid the overheads of WaitLatchOrSocket on epoll
531 : * systems, but we can't be sure that libpq (or any other
532 : * walreceiver implementation) has the same socket (even if
533 : * the fd is the same number, it may have been closed and
534 : * reopened since the last time). In future, if there is a
535 : * function for removing sockets from WaitEventSet, then we
536 : * could add and remove just the socket each time, potentially
537 : * avoiding some system calls.
538 : */
539 : Assert(wait_fd != PGINVALID_SOCKET);
540 56719 : rc = WaitLatchOrSocket(MyLatch,
541 : WL_EXIT_ON_PM_DEATH | WL_SOCKET_READABLE |
542 : WL_TIMEOUT | WL_LATCH_SET,
543 : wait_fd,
544 : nap,
545 : WAIT_EVENT_WAL_RECEIVER_MAIN);
546 56719 : if (rc & WL_LATCH_SET)
547 : {
548 13453 : ResetLatch(MyLatch);
549 13453 : CHECK_FOR_INTERRUPTS();
550 :
551 13365 : if (walrcv->apply_reply_requested)
552 : {
553 : /*
554 : * The recovery process has asked us to send apply
555 : * feedback now. Make sure the flag is really set to
556 : * false in shared memory before sending the reply, so
557 : * we don't miss a new request for a reply.
558 : */
559 13309 : walrcv->apply_reply_requested = false;
560 13309 : pg_memory_barrier();
561 13309 : XLogWalRcvSendReply(false, false, true);
562 : }
563 : }
564 56631 : if (rc & WL_TIMEOUT)
565 : {
566 : /*
567 : * We didn't receive anything new. If we haven't heard
568 : * anything from the server for more than
569 : * wal_receiver_timeout / 2, ping the server. Also, if
570 : * it's been longer than wal_receiver_status_interval
571 : * since the last update we sent, send a status update to
572 : * the primary anyway, to report any progress in applying
573 : * WAL.
574 : */
575 4 : bool requestReply = false;
576 :
577 : /*
578 : * Report pending statistics to the cumulative stats
579 : * system. This location is useful for the report as it
580 : * is not within a tight loop in the WAL receiver, to
581 : * avoid bloating pgstats with requests, while also making
582 : * sure that the reports happen each time a status update
583 : * is sent.
584 : */
585 4 : pgstat_report_wal(false);
586 :
587 : /*
588 : * Check if time since last receive from primary has
589 : * reached the configured limit.
590 : */
591 4 : now = GetCurrentTimestamp();
592 4 : if (now >= wakeup[WALRCV_WAKEUP_TERMINATE])
593 0 : ereport(ERROR,
594 : (errcode(ERRCODE_CONNECTION_FAILURE),
595 : errmsg("terminating walreceiver due to timeout")));
596 :
597 : /*
598 : * If we didn't receive anything new for half of receiver
599 : * replication timeout, then ping the server.
600 : */
601 4 : if (now >= wakeup[WALRCV_WAKEUP_PING])
602 : {
603 0 : requestReply = true;
604 0 : wakeup[WALRCV_WAKEUP_PING] = TIMESTAMP_INFINITY;
605 : }
606 :
607 4 : XLogWalRcvSendReply(requestReply, requestReply, false);
608 4 : XLogWalRcvSendHSFeedback(false);
609 : }
610 : }
611 :
612 : /*
613 : * The backend finished streaming. Exit streaming COPY-mode from
614 : * our side, too.
615 : */
616 46 : walrcv_endstreaming(wrconn, &primaryTLI);
617 :
618 : /*
619 : * If the server had switched to a new timeline that we didn't
620 : * know about when we began streaming, fetch its timeline history
621 : * file now.
622 : */
623 13 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
624 : }
625 : else
626 0 : ereport(LOG,
627 : (errmsg("primary server contains no more WAL on requested timeline %u",
628 : startpointTLI)));
629 :
630 : /*
631 : * End of WAL reached on the requested timeline. Close the last
632 : * segment, and await for new orders from the startup process.
633 : */
634 13 : if (recvFile >= 0)
635 : {
636 : char xlogfname[MAXFNAMELEN];
637 :
638 12 : XLogWalRcvFlush(false, startpointTLI);
639 12 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
640 12 : if (close(recvFile) != 0)
641 0 : ereport(PANIC,
642 : (errcode_for_file_access(),
643 : errmsg("could not close WAL segment %s: %m",
644 : xlogfname)));
645 :
646 : /*
647 : * Create .done file forcibly to prevent the streamed segment from
648 : * being archived later.
649 : */
650 12 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
651 12 : XLogArchiveForceDone(xlogfname);
652 : else
653 0 : XLogArchiveNotify(xlogfname);
654 : }
655 13 : recvFile = -1;
656 :
657 13 : elog(DEBUG1, "walreceiver ended streaming and awaits new instructions");
658 13 : WalRcvWaitForStartPosition(&startpoint, &startpointTLI);
659 : }
660 : /* not reached */
661 : }
662 :
663 : /*
664 : * Wait for startup process to set receiveStart and receiveStartTLI.
665 : */
666 : static void
667 13 : WalRcvWaitForStartPosition(XLogRecPtr *startpoint, TimeLineID *startpointTLI)
668 : {
669 13 : WalRcvData *walrcv = WalRcv;
670 : int state;
671 :
672 13 : SpinLockAcquire(&walrcv->mutex);
673 13 : state = walrcv->walRcvState;
674 13 : if (state != WALRCV_STREAMING && state != WALRCV_CONNECTING)
675 : {
676 0 : SpinLockRelease(&walrcv->mutex);
677 0 : if (state == WALRCV_STOPPING)
678 0 : proc_exit(0);
679 : else
680 0 : elog(FATAL, "unexpected walreceiver state");
681 : }
682 13 : walrcv->walRcvState = WALRCV_WAITING;
683 13 : walrcv->receiveStart = InvalidXLogRecPtr;
684 13 : walrcv->receiveStartTLI = 0;
685 13 : SpinLockRelease(&walrcv->mutex);
686 :
687 13 : set_ps_display("idle");
688 :
689 : /*
690 : * nudge startup process to notice that we've stopped streaming and are
691 : * now waiting for instructions.
692 : */
693 13 : WakeupRecovery();
694 : for (;;)
695 : {
696 26 : ResetLatch(MyLatch);
697 :
698 26 : CHECK_FOR_INTERRUPTS();
699 :
700 26 : SpinLockAcquire(&walrcv->mutex);
701 : Assert(walrcv->walRcvState == WALRCV_RESTARTING ||
702 : walrcv->walRcvState == WALRCV_WAITING ||
703 : walrcv->walRcvState == WALRCV_STOPPING);
704 26 : if (walrcv->walRcvState == WALRCV_RESTARTING)
705 : {
706 : /*
707 : * No need to handle changes in primary_conninfo or
708 : * primary_slot_name here. Startup process will signal us to
709 : * terminate in case those change.
710 : */
711 13 : *startpoint = walrcv->receiveStart;
712 13 : *startpointTLI = walrcv->receiveStartTLI;
713 13 : walrcv->walRcvState = WALRCV_CONNECTING;
714 13 : SpinLockRelease(&walrcv->mutex);
715 13 : break;
716 : }
717 13 : if (walrcv->walRcvState == WALRCV_STOPPING)
718 : {
719 : /*
720 : * We should've received SIGTERM if the startup process wants us
721 : * to die, but might as well check it here too.
722 : */
723 0 : SpinLockRelease(&walrcv->mutex);
724 0 : proc_exit(1);
725 : }
726 13 : SpinLockRelease(&walrcv->mutex);
727 :
728 13 : (void) WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, 0,
729 : WAIT_EVENT_WAL_RECEIVER_WAIT_START);
730 : }
731 :
732 13 : if (update_process_title)
733 : {
734 : char activitymsg[50];
735 :
736 13 : snprintf(activitymsg, sizeof(activitymsg), "restarting at %X/%08X",
737 13 : LSN_FORMAT_ARGS(*startpoint));
738 13 : set_ps_display(activitymsg);
739 : }
740 13 : }
741 :
742 : /*
743 : * Fetch any missing timeline history files between 'first' and 'last'
744 : * (inclusive) from the server.
745 : */
746 : static void
747 179 : WalRcvFetchTimeLineHistoryFiles(TimeLineID first, TimeLineID last)
748 : {
749 : TimeLineID tli;
750 :
751 381 : for (tli = first; tli <= last; tli++)
752 : {
753 : /* there's no history file for timeline 1 */
754 202 : if (tli != 1 && !existsTimeLineHistory(tli))
755 : {
756 : char *fname;
757 : char *content;
758 : int len;
759 : char expectedfname[MAXFNAMELEN];
760 :
761 12 : ereport(LOG,
762 : (errmsg("fetching timeline history file for timeline %u from primary server",
763 : tli)));
764 :
765 12 : walrcv_readtimelinehistoryfile(wrconn, tli, &fname, &content, &len);
766 :
767 : /*
768 : * Check that the filename on the primary matches what we
769 : * calculated ourselves. This is just a sanity check, it should
770 : * always match.
771 : */
772 12 : TLHistoryFileName(expectedfname, tli);
773 12 : if (strcmp(fname, expectedfname) != 0)
774 0 : ereport(ERROR,
775 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
776 : errmsg_internal("primary reported unexpected file name for timeline history file of timeline %u",
777 : tli)));
778 :
779 : /*
780 : * Write the file to pg_wal.
781 : */
782 12 : writeTimeLineHistoryFile(tli, content, len);
783 :
784 : /*
785 : * Mark the streamed history file as ready for archiving if
786 : * archive_mode is always.
787 : */
788 12 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
789 12 : XLogArchiveForceDone(fname);
790 : else
791 0 : XLogArchiveNotify(fname);
792 :
793 12 : pfree(fname);
794 12 : pfree(content);
795 : }
796 : }
797 179 : }
798 :
799 : /*
800 : * Mark us as STOPPED in shared memory at exit.
801 : */
802 : static void
803 258 : WalRcvDie(int code, Datum arg)
804 : {
805 258 : WalRcvData *walrcv = WalRcv;
806 258 : TimeLineID *startpointTLI_p = (TimeLineID *) DatumGetPointer(arg);
807 :
808 : Assert(*startpointTLI_p != 0);
809 :
810 : /* Ensure that all WAL records received are flushed to disk */
811 258 : XLogWalRcvFlush(true, *startpointTLI_p);
812 :
813 : /* Mark ourselves inactive in shared memory */
814 258 : SpinLockAcquire(&walrcv->mutex);
815 : Assert(walrcv->walRcvState == WALRCV_STREAMING ||
816 : walrcv->walRcvState == WALRCV_CONNECTING ||
817 : walrcv->walRcvState == WALRCV_RESTARTING ||
818 : walrcv->walRcvState == WALRCV_STARTING ||
819 : walrcv->walRcvState == WALRCV_WAITING ||
820 : walrcv->walRcvState == WALRCV_STOPPING);
821 : Assert(walrcv->pid == MyProcPid);
822 258 : walrcv->walRcvState = WALRCV_STOPPED;
823 258 : walrcv->pid = 0;
824 258 : walrcv->procno = INVALID_PROC_NUMBER;
825 258 : walrcv->ready_to_display = false;
826 258 : SpinLockRelease(&walrcv->mutex);
827 :
828 258 : ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
829 :
830 : /* Terminate the connection gracefully. */
831 258 : if (wrconn != NULL)
832 153 : walrcv_disconnect(wrconn);
833 :
834 : /* Wake up the startup process to notice promptly that we're gone */
835 258 : WakeupRecovery();
836 258 : }
837 :
838 : /*
839 : * Accept the message from XLOG stream, and process it.
840 : */
841 : static void
842 107472 : XLogWalRcvProcessMsg(unsigned char type, char *buf, Size len, TimeLineID tli)
843 : {
844 : int hdrlen;
845 : XLogRecPtr dataStart;
846 : XLogRecPtr walEnd;
847 : TimestampTz sendTime;
848 : bool replyRequested;
849 :
850 107472 : switch (type)
851 : {
852 107357 : case PqReplMsg_WALData:
853 : {
854 : StringInfoData incoming_message;
855 :
856 107357 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(int64);
857 107357 : if (len < hdrlen)
858 0 : ereport(ERROR,
859 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
860 : errmsg_internal("invalid WAL message received from primary")));
861 :
862 : /* initialize a StringInfo with the given buffer */
863 107357 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
864 :
865 : /* read the fields */
866 107357 : dataStart = pq_getmsgint64(&incoming_message);
867 107357 : walEnd = pq_getmsgint64(&incoming_message);
868 107357 : sendTime = pq_getmsgint64(&incoming_message);
869 107357 : ProcessWalSndrMessage(walEnd, sendTime);
870 :
871 107357 : buf += hdrlen;
872 107357 : len -= hdrlen;
873 107357 : XLogWalRcvWrite(buf, len, dataStart, tli);
874 107357 : break;
875 : }
876 115 : case PqReplMsg_Keepalive:
877 : {
878 : StringInfoData incoming_message;
879 :
880 115 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(char);
881 115 : if (len != hdrlen)
882 0 : ereport(ERROR,
883 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
884 : errmsg_internal("invalid keepalive message received from primary")));
885 :
886 : /* initialize a StringInfo with the given buffer */
887 115 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
888 :
889 : /* read the fields */
890 115 : walEnd = pq_getmsgint64(&incoming_message);
891 115 : sendTime = pq_getmsgint64(&incoming_message);
892 115 : replyRequested = pq_getmsgbyte(&incoming_message);
893 :
894 115 : ProcessWalSndrMessage(walEnd, sendTime);
895 :
896 : /* If the primary requested a reply, send one immediately */
897 115 : if (replyRequested)
898 115 : XLogWalRcvSendReply(true, false, false);
899 115 : break;
900 : }
901 0 : default:
902 0 : ereport(ERROR,
903 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
904 : errmsg_internal("invalid replication message type %d",
905 : type)));
906 : }
907 107472 : }
908 :
909 : /*
910 : * Write XLOG data to disk.
911 : */
912 : static void
913 107357 : XLogWalRcvWrite(char *buf, Size nbytes, XLogRecPtr recptr, TimeLineID tli)
914 : {
915 : int startoff;
916 : int byteswritten;
917 : instr_time start;
918 :
919 : Assert(tli != 0);
920 :
921 215107 : while (nbytes > 0)
922 : {
923 : int segbytes;
924 :
925 : /* Close the current segment if it's completed */
926 107750 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
927 393 : XLogWalRcvClose(recptr, tli);
928 :
929 107750 : if (recvFile < 0)
930 : {
931 : /* Create/use new log file */
932 895 : XLByteToSeg(recptr, recvSegNo, wal_segment_size);
933 895 : recvFile = XLogFileInit(recvSegNo, tli);
934 895 : recvFileTLI = tli;
935 : }
936 :
937 : /* Calculate the start offset of the received logs */
938 107750 : startoff = XLogSegmentOffset(recptr, wal_segment_size);
939 :
940 107750 : if (startoff + nbytes > wal_segment_size)
941 393 : segbytes = wal_segment_size - startoff;
942 : else
943 107357 : segbytes = nbytes;
944 :
945 : /* OK to write the logs */
946 107750 : errno = 0;
947 :
948 : /*
949 : * Measure I/O timing to write WAL data, for pg_stat_io.
950 : */
951 107750 : start = pgstat_prepare_io_time(track_wal_io_timing);
952 :
953 107750 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
954 107750 : byteswritten = pg_pwrite(recvFile, buf, segbytes, (pgoff_t) startoff);
955 107750 : pgstat_report_wait_end();
956 :
957 107750 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
958 : IOOP_WRITE, start, 1, byteswritten);
959 :
960 107750 : if (byteswritten <= 0)
961 : {
962 : char xlogfname[MAXFNAMELEN];
963 : int save_errno;
964 :
965 : /* if write didn't set errno, assume no disk space */
966 0 : if (errno == 0)
967 0 : errno = ENOSPC;
968 :
969 0 : save_errno = errno;
970 0 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
971 0 : errno = save_errno;
972 0 : ereport(PANIC,
973 : (errcode_for_file_access(),
974 : errmsg("could not write to WAL segment %s "
975 : "at offset %d, length %d: %m",
976 : xlogfname, startoff, segbytes)));
977 : }
978 :
979 : /* Update state for write */
980 107750 : recptr += byteswritten;
981 :
982 107750 : nbytes -= byteswritten;
983 107750 : buf += byteswritten;
984 :
985 107750 : LogstreamResult.Write = recptr;
986 : }
987 :
988 : /* Update shared-memory status */
989 107357 : pg_atomic_write_membarrier_u64(&WalRcv->writtenUpto, LogstreamResult.Write);
990 :
991 : /*
992 : * Wake up processes waiting for standby write LSN to reach current write
993 : * position.
994 : */
995 107357 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_WRITE, LogstreamResult.Write);
996 :
997 : /*
998 : * Close the current segment if it's fully written up in the last cycle of
999 : * the loop, to create its archive notification file soon. Otherwise WAL
1000 : * archiving of the segment will be delayed until any data in the next
1001 : * segment is received and written.
1002 : */
1003 107357 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
1004 361 : XLogWalRcvClose(recptr, tli);
1005 107357 : }
1006 :
1007 : /*
1008 : * Flush the log to disk.
1009 : *
1010 : * If we're in the midst of dying, it's unwise to do anything that might throw
1011 : * an error, so we skip sending a reply in that case.
1012 : */
1013 : static void
1014 40994 : XLogWalRcvFlush(bool dying, TimeLineID tli)
1015 : {
1016 : Assert(tli != 0);
1017 :
1018 40994 : if (LogstreamResult.Flush < LogstreamResult.Write)
1019 : {
1020 40341 : WalRcvData *walrcv = WalRcv;
1021 :
1022 40341 : issue_xlog_fsync(recvFile, recvSegNo, tli);
1023 :
1024 40341 : LogstreamResult.Flush = LogstreamResult.Write;
1025 :
1026 : /* Update shared-memory status */
1027 40341 : SpinLockAcquire(&walrcv->mutex);
1028 40341 : if (walrcv->flushedUpto < LogstreamResult.Flush)
1029 : {
1030 40341 : walrcv->latestChunkStart = walrcv->flushedUpto;
1031 40341 : walrcv->flushedUpto = LogstreamResult.Flush;
1032 40341 : walrcv->receivedTLI = tli;
1033 : }
1034 40341 : SpinLockRelease(&walrcv->mutex);
1035 :
1036 : /*
1037 : * Wake up processes waiting for standby flush LSN to reach current
1038 : * flush position.
1039 : */
1040 40341 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_FLUSH, LogstreamResult.Flush);
1041 :
1042 : /* Signal the startup process and walsender that new WAL has arrived */
1043 40341 : WakeupRecovery();
1044 40341 : if (AllowCascadeReplication())
1045 40341 : WalSndWakeup(true, false);
1046 :
1047 : /* Report XLOG streaming progress in PS display */
1048 40341 : if (update_process_title)
1049 : {
1050 : char activitymsg[50];
1051 :
1052 40341 : snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%08X",
1053 40341 : LSN_FORMAT_ARGS(LogstreamResult.Write));
1054 40341 : set_ps_display(activitymsg);
1055 : }
1056 :
1057 : /* Also let the primary know that we made some progress */
1058 40341 : if (!dying)
1059 : {
1060 40339 : XLogWalRcvSendReply(false, false, false);
1061 40339 : XLogWalRcvSendHSFeedback(false);
1062 : }
1063 : }
1064 40994 : }
1065 :
1066 : /*
1067 : * Close the current segment.
1068 : *
1069 : * Flush the segment to disk before closing it. Otherwise we have to
1070 : * reopen and fsync it later.
1071 : *
1072 : * Create an archive notification file since the segment is known completed.
1073 : */
1074 : static void
1075 754 : XLogWalRcvClose(XLogRecPtr recptr, TimeLineID tli)
1076 : {
1077 : char xlogfname[MAXFNAMELEN];
1078 :
1079 : Assert(recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size));
1080 : Assert(tli != 0);
1081 :
1082 : /*
1083 : * fsync() and close current file before we switch to next one. We would
1084 : * otherwise have to reopen this file to fsync it later
1085 : */
1086 754 : XLogWalRcvFlush(false, tli);
1087 :
1088 754 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
1089 :
1090 : /*
1091 : * XLOG segment files will be re-read by recovery in startup process soon,
1092 : * so we don't advise the OS to release cache pages associated with the
1093 : * file like XLogFileClose() does.
1094 : */
1095 754 : if (close(recvFile) != 0)
1096 0 : ereport(PANIC,
1097 : (errcode_for_file_access(),
1098 : errmsg("could not close WAL segment %s: %m",
1099 : xlogfname)));
1100 :
1101 : /*
1102 : * Create .done file forcibly to prevent the streamed segment from being
1103 : * archived later.
1104 : */
1105 754 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
1106 754 : XLogArchiveForceDone(xlogfname);
1107 : else
1108 0 : XLogArchiveNotify(xlogfname);
1109 :
1110 754 : recvFile = -1;
1111 754 : }
1112 :
1113 : /*
1114 : * Send reply message to primary, indicating our current WAL locations and
1115 : * time.
1116 : *
1117 : * The message is sent if 'force' is set, if enough time has passed since the
1118 : * last update to reach wal_receiver_status_interval, or if WAL locations have
1119 : * advanced since the previous status update. If wal_receiver_status_interval
1120 : * is disabled and 'force' is false, this function does nothing. Set 'force' to
1121 : * send the message unconditionally.
1122 : *
1123 : * Whether WAL locations are considered "advanced" depends on 'checkApply'.
1124 : * If 'checkApply' is false, only the write and flush locations are checked.
1125 : * This should be used when the call is triggered by write/flush activity
1126 : * (e.g., after walreceiver writes or flushes WAL), and avoids the
1127 : * apply-location check, which requires a spinlock. If 'checkApply' is true,
1128 : * the apply location is also considered. This should be used when the apply
1129 : * location is expected to advance (e.g., when the startup process requests
1130 : * an apply notification).
1131 : *
1132 : * If 'requestReply' is true, requests the server to reply immediately upon
1133 : * receiving this message. This is used for heartbeats, when approaching
1134 : * wal_receiver_timeout.
1135 : */
1136 : static void
1137 93903 : XLogWalRcvSendReply(bool force, bool requestReply, bool checkApply)
1138 : {
1139 : static XLogRecPtr writePtr = InvalidXLogRecPtr;
1140 : static XLogRecPtr flushPtr = InvalidXLogRecPtr;
1141 : static XLogRecPtr applyPtr = InvalidXLogRecPtr;
1142 93903 : XLogRecPtr latestApplyPtr = InvalidXLogRecPtr;
1143 : TimestampTz now;
1144 :
1145 : /*
1146 : * If the user doesn't want status to be reported to the primary, be sure
1147 : * to exit before doing anything at all.
1148 : */
1149 93903 : if (!force && wal_receiver_status_interval <= 0)
1150 0 : return;
1151 :
1152 : /* Get current timestamp. */
1153 93903 : now = GetCurrentTimestamp();
1154 :
1155 : /*
1156 : * We can compare the write and flush positions to the last message we
1157 : * sent without taking any lock, but the apply position requires a spin
1158 : * lock, so we don't check that unless it is expected to advance since the
1159 : * previous update, i.e., when 'checkApply' is true.
1160 : */
1161 93903 : if (!force && now < wakeup[WALRCV_WAKEUP_REPLY])
1162 : {
1163 93623 : if (checkApply)
1164 13309 : latestApplyPtr = GetXLogReplayRecPtr(NULL);
1165 :
1166 93623 : if (writePtr == LogstreamResult.Write
1167 53201 : && flushPtr == LogstreamResult.Flush
1168 13616 : && (!checkApply || applyPtr == latestApplyPtr))
1169 3296 : return;
1170 : }
1171 :
1172 : /* Make sure we wake up when it's time to send another reply. */
1173 90607 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_REPLY, now);
1174 :
1175 : /* Construct a new message */
1176 90607 : writePtr = LogstreamResult.Write;
1177 90607 : flushPtr = LogstreamResult.Flush;
1178 90607 : applyPtr = XLogRecPtrIsValid(latestApplyPtr) ?
1179 90607 : latestApplyPtr : GetXLogReplayRecPtr(NULL);
1180 :
1181 90607 : resetStringInfo(&reply_message);
1182 90607 : pq_sendbyte(&reply_message, PqReplMsg_StandbyStatusUpdate);
1183 90607 : pq_sendint64(&reply_message, writePtr);
1184 90607 : pq_sendint64(&reply_message, flushPtr);
1185 90607 : pq_sendint64(&reply_message, applyPtr);
1186 90607 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1187 90607 : pq_sendbyte(&reply_message, requestReply ? 1 : 0);
1188 :
1189 : /* Send it */
1190 90607 : elog(DEBUG2, "sending write %X/%08X flush %X/%08X apply %X/%08X%s",
1191 : LSN_FORMAT_ARGS(writePtr),
1192 : LSN_FORMAT_ARGS(flushPtr),
1193 : LSN_FORMAT_ARGS(applyPtr),
1194 : requestReply ? " (reply requested)" : "");
1195 :
1196 90607 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1197 : }
1198 :
1199 : /*
1200 : * Send hot standby feedback message to primary, plus the current time,
1201 : * in case they don't have a watch.
1202 : *
1203 : * If the user disables feedback, send one final message to tell sender
1204 : * to forget about the xmin on this standby. We also send this message
1205 : * on first connect because a previous connection might have set xmin
1206 : * on a replication slot. (If we're not using a slot it's harmless to
1207 : * send a feedback message explicitly setting InvalidTransactionId).
1208 : */
1209 : static void
1210 40534 : XLogWalRcvSendHSFeedback(bool immed)
1211 : {
1212 : TimestampTz now;
1213 : FullTransactionId nextFullXid;
1214 : TransactionId nextXid;
1215 : uint32 xmin_epoch,
1216 : catalog_xmin_epoch;
1217 : TransactionId xmin,
1218 : catalog_xmin;
1219 :
1220 : /* initially true so we always send at least one feedback message */
1221 : static bool primary_has_standby_xmin = true;
1222 :
1223 : /*
1224 : * If the user doesn't want status to be reported to the primary, be sure
1225 : * to exit before doing anything at all.
1226 : */
1227 40534 : if ((wal_receiver_status_interval <= 0 || !hot_standby_feedback) &&
1228 39991 : !primary_has_standby_xmin)
1229 40359 : return;
1230 :
1231 : /* Get current timestamp. */
1232 670 : now = GetCurrentTimestamp();
1233 :
1234 : /* Send feedback at most once per wal_receiver_status_interval. */
1235 670 : if (!immed && now < wakeup[WALRCV_WAKEUP_HSFEEDBACK])
1236 494 : return;
1237 :
1238 : /* Make sure we wake up when it's time to send feedback again. */
1239 176 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_HSFEEDBACK, now);
1240 :
1241 : /*
1242 : * If Hot Standby is not yet accepting connections there is nothing to
1243 : * send. Check this after the interval has expired to reduce number of
1244 : * calls.
1245 : *
1246 : * Bailing out here also ensures that we don't send feedback until we've
1247 : * read our own replication slot state, so we don't tell the primary to
1248 : * discard needed xmin or catalog_xmin from any slots that may exist on
1249 : * this replica.
1250 : */
1251 176 : if (!HotStandbyActive())
1252 1 : return;
1253 :
1254 : /*
1255 : * Make the expensive call to get the oldest xmin once we are certain
1256 : * everything else has been checked.
1257 : */
1258 175 : if (hot_standby_feedback)
1259 : {
1260 53 : GetReplicationHorizons(&xmin, &catalog_xmin);
1261 : }
1262 : else
1263 : {
1264 122 : xmin = InvalidTransactionId;
1265 122 : catalog_xmin = InvalidTransactionId;
1266 : }
1267 :
1268 : /*
1269 : * Get epoch and adjust if nextXid and oldestXmin are different sides of
1270 : * the epoch boundary.
1271 : */
1272 175 : nextFullXid = ReadNextFullTransactionId();
1273 175 : nextXid = XidFromFullTransactionId(nextFullXid);
1274 175 : xmin_epoch = EpochFromFullTransactionId(nextFullXid);
1275 175 : catalog_xmin_epoch = xmin_epoch;
1276 175 : if (nextXid < xmin)
1277 0 : xmin_epoch--;
1278 175 : if (nextXid < catalog_xmin)
1279 0 : catalog_xmin_epoch--;
1280 :
1281 175 : elog(DEBUG2, "sending hot standby feedback xmin %u epoch %u catalog_xmin %u catalog_xmin_epoch %u",
1282 : xmin, xmin_epoch, catalog_xmin, catalog_xmin_epoch);
1283 :
1284 : /* Construct the message and send it. */
1285 175 : resetStringInfo(&reply_message);
1286 175 : pq_sendbyte(&reply_message, PqReplMsg_HotStandbyFeedback);
1287 175 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1288 175 : pq_sendint32(&reply_message, xmin);
1289 175 : pq_sendint32(&reply_message, xmin_epoch);
1290 175 : pq_sendint32(&reply_message, catalog_xmin);
1291 175 : pq_sendint32(&reply_message, catalog_xmin_epoch);
1292 175 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1293 175 : if (TransactionIdIsValid(xmin) || TransactionIdIsValid(catalog_xmin))
1294 53 : primary_has_standby_xmin = true;
1295 : else
1296 122 : primary_has_standby_xmin = false;
1297 : }
1298 :
1299 : /*
1300 : * Update shared memory status upon receiving a message from primary.
1301 : *
1302 : * 'walEnd' and 'sendTime' are the end-of-WAL and timestamp of the latest
1303 : * message, reported by primary.
1304 : */
1305 : static void
1306 107472 : ProcessWalSndrMessage(XLogRecPtr walEnd, TimestampTz sendTime)
1307 : {
1308 107472 : WalRcvData *walrcv = WalRcv;
1309 107472 : TimestampTz lastMsgReceiptTime = GetCurrentTimestamp();
1310 :
1311 : /* Update shared-memory status */
1312 107472 : SpinLockAcquire(&walrcv->mutex);
1313 107472 : if (walrcv->latestWalEnd < walEnd)
1314 22917 : walrcv->latestWalEndTime = sendTime;
1315 107472 : walrcv->latestWalEnd = walEnd;
1316 107472 : walrcv->lastMsgSendTime = sendTime;
1317 107472 : walrcv->lastMsgReceiptTime = lastMsgReceiptTime;
1318 107472 : SpinLockRelease(&walrcv->mutex);
1319 :
1320 107472 : if (message_level_is_interesting(DEBUG2))
1321 : {
1322 : char *sendtime;
1323 : char *receipttime;
1324 : int applyDelay;
1325 :
1326 : /* Copy because timestamptz_to_str returns a static buffer */
1327 404 : sendtime = pstrdup(timestamptz_to_str(sendTime));
1328 404 : receipttime = pstrdup(timestamptz_to_str(lastMsgReceiptTime));
1329 404 : applyDelay = GetReplicationApplyDelay();
1330 :
1331 : /* apply delay is not available */
1332 404 : if (applyDelay == -1)
1333 1 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay (N/A) transfer latency %d ms",
1334 : sendtime,
1335 : receipttime,
1336 : GetReplicationTransferLatency());
1337 : else
1338 403 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay %d ms transfer latency %d ms",
1339 : sendtime,
1340 : receipttime,
1341 : applyDelay,
1342 : GetReplicationTransferLatency());
1343 :
1344 404 : pfree(sendtime);
1345 404 : pfree(receipttime);
1346 : }
1347 107472 : }
1348 :
1349 : /*
1350 : * Compute the next wakeup time for a given wakeup reason. Can be called to
1351 : * initialize a wakeup time, to adjust it for the next wakeup, or to
1352 : * reinitialize it when GUCs have changed. We ask the caller to pass in the
1353 : * value of "now" because this frequently avoids multiple calls of
1354 : * GetCurrentTimestamp(). It had better be a reasonably up-to-date value
1355 : * though.
1356 : */
1357 : static void
1358 306491 : WalRcvComputeNextWakeup(WalRcvWakeupReason reason, TimestampTz now)
1359 : {
1360 306491 : switch (reason)
1361 : {
1362 107663 : case WALRCV_WAKEUP_TERMINATE:
1363 107663 : if (wal_receiver_timeout <= 0)
1364 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1365 : else
1366 107663 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout);
1367 107663 : break;
1368 107663 : case WALRCV_WAKEUP_PING:
1369 107663 : if (wal_receiver_timeout <= 0)
1370 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1371 : else
1372 107663 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout / 2);
1373 107663 : break;
1374 367 : case WALRCV_WAKEUP_HSFEEDBACK:
1375 367 : if (!hot_standby_feedback || wal_receiver_status_interval <= 0)
1376 265 : wakeup[reason] = TIMESTAMP_INFINITY;
1377 : else
1378 102 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1379 367 : break;
1380 90798 : case WALRCV_WAKEUP_REPLY:
1381 90798 : if (wal_receiver_status_interval <= 0)
1382 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1383 : else
1384 90798 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1385 90798 : break;
1386 : /* there's intentionally no default: here */
1387 : }
1388 306491 : }
1389 :
1390 : /*
1391 : * Wake up the walreceiver main loop.
1392 : *
1393 : * This is called by the startup process whenever interesting xlog records
1394 : * are applied, so that walreceiver can check if it needs to send an apply
1395 : * notification back to the primary which may be waiting in a COMMIT with
1396 : * synchronous_commit = remote_apply.
1397 : */
1398 : void
1399 13242 : WalRcvRequestApplyReply(void)
1400 : {
1401 : ProcNumber procno;
1402 :
1403 13242 : WalRcv->apply_reply_requested = true;
1404 : /* fetching the proc number is probably atomic, but don't rely on it */
1405 13242 : SpinLockAcquire(&WalRcv->mutex);
1406 13242 : procno = WalRcv->procno;
1407 13242 : SpinLockRelease(&WalRcv->mutex);
1408 13242 : if (procno != INVALID_PROC_NUMBER)
1409 13052 : SetLatch(&GetPGProcByNumber(procno)->procLatch);
1410 13242 : }
1411 :
1412 : /*
1413 : * Return a string constant representing the state. This is used
1414 : * in system functions and views, and should *not* be translated.
1415 : */
1416 : static const char *
1417 16 : WalRcvGetStateString(WalRcvState state)
1418 : {
1419 16 : switch (state)
1420 : {
1421 0 : case WALRCV_STOPPED:
1422 0 : return "stopped";
1423 0 : case WALRCV_STARTING:
1424 0 : return "starting";
1425 0 : case WALRCV_CONNECTING:
1426 0 : return "connecting";
1427 16 : case WALRCV_STREAMING:
1428 16 : return "streaming";
1429 0 : case WALRCV_WAITING:
1430 0 : return "waiting";
1431 0 : case WALRCV_RESTARTING:
1432 0 : return "restarting";
1433 0 : case WALRCV_STOPPING:
1434 0 : return "stopping";
1435 : }
1436 0 : return "UNKNOWN";
1437 : }
1438 :
1439 : /*
1440 : * Returns activity of WAL receiver, including pid, state and xlog locations
1441 : * received from the WAL sender of another server.
1442 : */
1443 : Datum
1444 27 : pg_stat_get_wal_receiver(PG_FUNCTION_ARGS)
1445 : {
1446 : TupleDesc tupdesc;
1447 : Datum *values;
1448 : bool *nulls;
1449 : int pid;
1450 : bool ready_to_display;
1451 : WalRcvState state;
1452 : XLogRecPtr receive_start_lsn;
1453 : TimeLineID receive_start_tli;
1454 : XLogRecPtr written_lsn;
1455 : XLogRecPtr flushed_lsn;
1456 : TimeLineID received_tli;
1457 : TimestampTz last_send_time;
1458 : TimestampTz last_receipt_time;
1459 : XLogRecPtr latest_end_lsn;
1460 : TimestampTz latest_end_time;
1461 : char sender_host[NI_MAXHOST];
1462 27 : int sender_port = 0;
1463 : char slotname[NAMEDATALEN];
1464 : char conninfo[MAXCONNINFO];
1465 :
1466 : /* Take a lock to ensure value consistency */
1467 27 : SpinLockAcquire(&WalRcv->mutex);
1468 27 : pid = (int) WalRcv->pid;
1469 27 : ready_to_display = WalRcv->ready_to_display;
1470 27 : state = WalRcv->walRcvState;
1471 27 : receive_start_lsn = WalRcv->receiveStart;
1472 27 : receive_start_tli = WalRcv->receiveStartTLI;
1473 27 : flushed_lsn = WalRcv->flushedUpto;
1474 27 : received_tli = WalRcv->receivedTLI;
1475 27 : last_send_time = WalRcv->lastMsgSendTime;
1476 27 : last_receipt_time = WalRcv->lastMsgReceiptTime;
1477 27 : latest_end_lsn = WalRcv->latestWalEnd;
1478 27 : latest_end_time = WalRcv->latestWalEndTime;
1479 27 : strlcpy(slotname, WalRcv->slotname, sizeof(slotname));
1480 27 : strlcpy(sender_host, WalRcv->sender_host, sizeof(sender_host));
1481 27 : sender_port = WalRcv->sender_port;
1482 27 : strlcpy(conninfo, WalRcv->conninfo, sizeof(conninfo));
1483 27 : SpinLockRelease(&WalRcv->mutex);
1484 :
1485 : /*
1486 : * No WAL receiver (or not ready yet), just return a tuple with NULL
1487 : * values
1488 : */
1489 27 : if (pid == 0 || !ready_to_display)
1490 11 : PG_RETURN_NULL();
1491 :
1492 : /*
1493 : * Read "writtenUpto" without holding a spinlock. Note that it may not be
1494 : * consistent with the other shared variables of the WAL receiver
1495 : * protected by a spinlock, but this should not be used for data integrity
1496 : * checks.
1497 : */
1498 16 : written_lsn = pg_atomic_read_u64(&WalRcv->writtenUpto);
1499 :
1500 : /* determine result type */
1501 16 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1502 0 : elog(ERROR, "return type must be a row type");
1503 :
1504 16 : values = palloc0_array(Datum, tupdesc->natts);
1505 16 : nulls = palloc0_array(bool, tupdesc->natts);
1506 :
1507 : /* Fetch values */
1508 16 : values[0] = Int32GetDatum(pid);
1509 :
1510 16 : if (!has_privs_of_role(GetUserId(), ROLE_PG_READ_ALL_STATS))
1511 : {
1512 : /*
1513 : * Only superusers and roles with privileges of pg_read_all_stats can
1514 : * see details. Other users only get the pid value to know whether it
1515 : * is a WAL receiver, but no details.
1516 : */
1517 0 : memset(&nulls[1], true, sizeof(bool) * (tupdesc->natts - 1));
1518 : }
1519 : else
1520 : {
1521 16 : values[1] = CStringGetTextDatum(WalRcvGetStateString(state));
1522 :
1523 16 : if (!XLogRecPtrIsValid(receive_start_lsn))
1524 0 : nulls[2] = true;
1525 : else
1526 16 : values[2] = LSNGetDatum(receive_start_lsn);
1527 16 : values[3] = Int32GetDatum(receive_start_tli);
1528 16 : if (!XLogRecPtrIsValid(written_lsn))
1529 0 : nulls[4] = true;
1530 : else
1531 16 : values[4] = LSNGetDatum(written_lsn);
1532 16 : if (!XLogRecPtrIsValid(flushed_lsn))
1533 0 : nulls[5] = true;
1534 : else
1535 16 : values[5] = LSNGetDatum(flushed_lsn);
1536 16 : values[6] = Int32GetDatum(received_tli);
1537 16 : if (last_send_time == 0)
1538 0 : nulls[7] = true;
1539 : else
1540 16 : values[7] = TimestampTzGetDatum(last_send_time);
1541 16 : if (last_receipt_time == 0)
1542 0 : nulls[8] = true;
1543 : else
1544 16 : values[8] = TimestampTzGetDatum(last_receipt_time);
1545 16 : if (!XLogRecPtrIsValid(latest_end_lsn))
1546 0 : nulls[9] = true;
1547 : else
1548 16 : values[9] = LSNGetDatum(latest_end_lsn);
1549 16 : if (latest_end_time == 0)
1550 0 : nulls[10] = true;
1551 : else
1552 16 : values[10] = TimestampTzGetDatum(latest_end_time);
1553 16 : if (*slotname == '\0')
1554 14 : nulls[11] = true;
1555 : else
1556 2 : values[11] = CStringGetTextDatum(slotname);
1557 16 : if (*sender_host == '\0')
1558 0 : nulls[12] = true;
1559 : else
1560 16 : values[12] = CStringGetTextDatum(sender_host);
1561 16 : if (sender_port == 0)
1562 0 : nulls[13] = true;
1563 : else
1564 16 : values[13] = Int32GetDatum(sender_port);
1565 16 : if (*conninfo == '\0')
1566 0 : nulls[14] = true;
1567 : else
1568 16 : values[14] = CStringGetTextDatum(conninfo);
1569 : }
1570 :
1571 : /* Returns the record as Datum */
1572 16 : PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
1573 : }
|
