Branch data 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) 2010-2026, 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 : :
83 : :
84 : : /*
85 : : * GUC variables. (Other variables that affect walreceiver are in xlog.c
86 : : * because they're passed down from the startup process, for better
87 : : * synchronization.)
88 : : */
89 : : int wal_receiver_status_interval;
90 : : int wal_receiver_timeout;
91 : : bool hot_standby_feedback;
92 : :
93 : : /* libpqwalreceiver connection */
94 : : static WalReceiverConn *wrconn = NULL;
95 : : WalReceiverFunctionsType *WalReceiverFunctions = NULL;
96 : :
97 : : /*
98 : : * These variables are used similarly to openLogFile/SegNo,
99 : : * but for walreceiver to write the XLOG. recvFileTLI is the TimeLineID
100 : : * corresponding the filename of recvFile.
101 : : */
102 : : static int recvFile = -1;
103 : : static TimeLineID recvFileTLI = 0;
104 : : static XLogSegNo recvSegNo = 0;
105 : :
106 : : /*
107 : : * LogstreamResult indicates the byte positions that we have already
108 : : * written/fsynced.
109 : : */
110 : : static struct
111 : : {
112 : : XLogRecPtr Write; /* last byte + 1 written out in the standby */
113 : : XLogRecPtr Flush; /* last byte + 1 flushed in the standby */
114 : : } LogstreamResult;
115 : :
116 : : /*
117 : : * Reasons to wake up and perform periodic tasks.
118 : : */
119 : : typedef enum WalRcvWakeupReason
120 : : {
121 : : WALRCV_WAKEUP_TERMINATE,
122 : : WALRCV_WAKEUP_PING,
123 : : WALRCV_WAKEUP_REPLY,
124 : : WALRCV_WAKEUP_HSFEEDBACK,
125 : : #define NUM_WALRCV_WAKEUPS (WALRCV_WAKEUP_HSFEEDBACK + 1)
126 : : } WalRcvWakeupReason;
127 : :
128 : : /*
129 : : * Wake up times for periodic tasks.
130 : : */
131 : : static TimestampTz wakeup[NUM_WALRCV_WAKEUPS];
132 : :
133 : : static StringInfoData reply_message;
134 : :
135 : : /* Prototypes for private functions */
136 : : static void WalRcvFetchTimeLineHistoryFiles(TimeLineID first, TimeLineID last);
137 : : static void WalRcvWaitForStartPosition(XLogRecPtr *startpoint, TimeLineID *startpointTLI);
138 : : static void WalRcvDie(int code, Datum arg);
139 : : static void XLogWalRcvProcessMsg(unsigned char type, char *buf, Size len,
140 : : TimeLineID tli);
141 : : static void XLogWalRcvWrite(char *buf, Size nbytes, XLogRecPtr recptr,
142 : : TimeLineID tli);
143 : : static void XLogWalRcvFlush(bool dying, TimeLineID tli);
144 : : static void XLogWalRcvClose(XLogRecPtr recptr, TimeLineID tli);
145 : : static void XLogWalRcvSendReply(bool force, bool requestReply);
146 : : static void XLogWalRcvSendHSFeedback(bool immed);
147 : : static void ProcessWalSndrMessage(XLogRecPtr walEnd, TimestampTz sendTime);
148 : : static void WalRcvComputeNextWakeup(WalRcvWakeupReason reason, TimestampTz now);
149 : :
150 : :
151 : : /* Main entry point for walreceiver process */
152 : : void
153 : 0 : WalReceiverMain(const void *startup_data, size_t startup_data_len)
154 : : {
155 : 0 : char conninfo[MAXCONNINFO];
156 : 0 : char *tmp_conninfo;
157 : 0 : char slotname[NAMEDATALEN];
158 : 0 : bool is_temp_slot;
159 : 0 : XLogRecPtr startpoint;
160 : 0 : TimeLineID startpointTLI;
161 : 0 : TimeLineID primaryTLI;
162 : 0 : bool first_stream;
163 : 0 : WalRcvData *walrcv;
164 : 0 : TimestampTz now;
165 : 0 : char *err;
166 : 0 : char *sender_host = NULL;
167 : 0 : int sender_port = 0;
168 : 0 : char *appname;
169 : :
170 [ # # ]: 0 : Assert(startup_data_len == 0);
171 : :
172 : 0 : MyBackendType = B_WAL_RECEIVER;
173 : 0 : 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 : 0 : walrcv = WalRcv;
180 [ # # ]: 0 : 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 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
190 [ # # ]: 0 : Assert(walrcv->pid == 0);
191 [ # # # # : 0 : switch (walrcv->walRcvState)
# ]
192 : : {
193 : : case WALRCV_STOPPING:
194 : : /* If we've already been requested to stop, don't start up. */
195 : 0 : walrcv->walRcvState = WALRCV_STOPPED;
196 : : /* fall through */
197 : :
198 : : case WALRCV_STOPPED:
199 : 0 : SpinLockRelease(&walrcv->mutex);
200 : 0 : ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
201 : 0 : proc_exit(1);
202 : : break;
203 : :
204 : : case WALRCV_STARTING:
205 : : /* The usual case */
206 : : break;
207 : :
208 : : case WALRCV_CONNECTING:
209 : : case WALRCV_WAITING:
210 : : case WALRCV_STREAMING:
211 : 0 : 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 : 0 : }
217 : : /* Advertise our PID so that the startup process can kill us */
218 : 0 : walrcv->pid = MyProcPid;
219 : 0 : walrcv->walRcvState = WALRCV_CONNECTING;
220 : :
221 : : /* Fetch information required to start streaming */
222 : 0 : walrcv->ready_to_display = false;
223 : 0 : strlcpy(conninfo, walrcv->conninfo, MAXCONNINFO);
224 : 0 : strlcpy(slotname, walrcv->slotname, NAMEDATALEN);
225 : 0 : is_temp_slot = walrcv->is_temp_slot;
226 : 0 : startpoint = walrcv->receiveStart;
227 : 0 : 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 [ # # # # ]: 0 : Assert(!is_temp_slot || (slotname[0] == '\0'));
234 : :
235 : : /* Initialise to a sanish value */
236 : 0 : now = GetCurrentTimestamp();
237 : 0 : walrcv->lastMsgSendTime =
238 : 0 : walrcv->lastMsgReceiptTime = walrcv->latestWalEndTime = now;
239 : :
240 : : /* Report our proc number so that others can wake us up */
241 : 0 : walrcv->procno = MyProcNumber;
242 : :
243 : 0 : SpinLockRelease(&walrcv->mutex);
244 : :
245 : 0 : pg_atomic_write_u64(&WalRcv->writtenUpto, 0);
246 : :
247 : : /* Arrange to clean up at walreceiver exit */
248 : 0 : on_shmem_exit(WalRcvDie, PointerGetDatum(&startpointTLI));
249 : :
250 : : /* Properly accept or ignore signals the postmaster might send us */
251 : 0 : pqsignal(SIGHUP, SignalHandlerForConfigReload); /* set flag to read config
252 : : * file */
253 : 0 : pqsignal(SIGINT, SIG_IGN);
254 : 0 : pqsignal(SIGTERM, die); /* request shutdown */
255 : : /* SIGQUIT handler was already set up by InitPostmasterChild */
256 : 0 : pqsignal(SIGALRM, SIG_IGN);
257 : 0 : pqsignal(SIGPIPE, SIG_IGN);
258 : 0 : pqsignal(SIGUSR1, procsignal_sigusr1_handler);
259 : 0 : pqsignal(SIGUSR2, SIG_IGN);
260 : :
261 : : /* Reset some signals that are accepted by postmaster but not here */
262 : 0 : pqsignal(SIGCHLD, SIG_DFL);
263 : :
264 : : /* Load the libpq-specific functions */
265 : 0 : load_file("libpqwalreceiver", false);
266 [ # # ]: 0 : if (WalReceiverFunctions == NULL)
267 [ # # # # ]: 0 : elog(ERROR, "libpqwalreceiver didn't initialize correctly");
268 : :
269 : : /* Unblock signals (they were blocked when the postmaster forked us) */
270 : 0 : sigprocmask(SIG_SETMASK, &UnBlockSig, NULL);
271 : :
272 : : /* Establish the connection to the primary for XLOG streaming */
273 [ # # ]: 0 : appname = cluster_name[0] ? cluster_name : "walreceiver";
274 : 0 : wrconn = walrcv_connect(conninfo, true, false, false, appname, &err);
275 [ # # ]: 0 : if (!wrconn)
276 [ # # # # ]: 0 : ereport(ERROR,
277 : : (errcode(ERRCODE_CONNECTION_FAILURE),
278 : : errmsg("streaming replication receiver \"%s\" could not connect to the primary server: %s",
279 : : appname, err)));
280 : :
281 : : /*
282 : : * Save user-visible connection string. This clobbers the original
283 : : * conninfo, for security. Also save host and port of the sender server
284 : : * this walreceiver is connected to.
285 : : */
286 : 0 : tmp_conninfo = walrcv_get_conninfo(wrconn);
287 : 0 : walrcv_get_senderinfo(wrconn, &sender_host, &sender_port);
288 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
289 : 0 : memset(walrcv->conninfo, 0, MAXCONNINFO);
290 [ # # ]: 0 : if (tmp_conninfo)
291 : 0 : strlcpy(walrcv->conninfo, tmp_conninfo, MAXCONNINFO);
292 : :
293 : 0 : memset(walrcv->sender_host, 0, NI_MAXHOST);
294 [ # # ]: 0 : if (sender_host)
295 : 0 : strlcpy(walrcv->sender_host, sender_host, NI_MAXHOST);
296 : :
297 : 0 : walrcv->sender_port = sender_port;
298 : 0 : walrcv->ready_to_display = true;
299 : 0 : SpinLockRelease(&walrcv->mutex);
300 : :
301 [ # # ]: 0 : if (tmp_conninfo)
302 : 0 : pfree(tmp_conninfo);
303 : :
304 [ # # ]: 0 : if (sender_host)
305 : 0 : pfree(sender_host);
306 : :
307 : 0 : first_stream = true;
308 : 0 : for (;;)
309 : : {
310 : 0 : char *primary_sysid;
311 : 0 : char standby_sysid[32];
312 : 0 : WalRcvStreamOptions options;
313 : :
314 : : /*
315 : : * Check that we're connected to a valid server using the
316 : : * IDENTIFY_SYSTEM replication command.
317 : : */
318 : 0 : primary_sysid = walrcv_identify_system(wrconn, &primaryTLI);
319 : :
320 : 0 : snprintf(standby_sysid, sizeof(standby_sysid), UINT64_FORMAT,
321 : 0 : GetSystemIdentifier());
322 [ # # ]: 0 : if (strcmp(primary_sysid, standby_sysid) != 0)
323 : : {
324 [ # # # # ]: 0 : ereport(ERROR,
325 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
326 : : errmsg("database system identifier differs between the primary and standby"),
327 : : errdetail("The primary's identifier is %s, the standby's identifier is %s.",
328 : : primary_sysid, standby_sysid)));
329 : 0 : }
330 : :
331 : : /*
332 : : * Confirm that the current timeline of the primary is the same or
333 : : * ahead of ours.
334 : : */
335 [ # # ]: 0 : if (primaryTLI < startpointTLI)
336 [ # # # # ]: 0 : ereport(ERROR,
337 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
338 : : errmsg("highest timeline %u of the primary is behind recovery timeline %u",
339 : : primaryTLI, startpointTLI)));
340 : :
341 : : /*
342 : : * Get any missing history files. We do this always, even when we're
343 : : * not interested in that timeline, so that if we're promoted to
344 : : * become the primary later on, we don't select the same timeline that
345 : : * was already used in the current primary. This isn't bullet-proof -
346 : : * you'll need some external software to manage your cluster if you
347 : : * need to ensure that a unique timeline id is chosen in every case,
348 : : * but let's avoid the confusion of timeline id collisions where we
349 : : * can.
350 : : */
351 : 0 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
352 : :
353 : : /*
354 : : * Create temporary replication slot if requested, and update slot
355 : : * name in shared memory. (Note the slot name cannot already be set
356 : : * in this case.)
357 : : */
358 [ # # ]: 0 : if (is_temp_slot)
359 : : {
360 : 0 : snprintf(slotname, sizeof(slotname),
361 : : "pg_walreceiver_%lld",
362 : 0 : (long long int) walrcv_get_backend_pid(wrconn));
363 : :
364 : 0 : walrcv_create_slot(wrconn, slotname, true, false, false, 0, NULL);
365 : :
366 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
367 : 0 : strlcpy(walrcv->slotname, slotname, NAMEDATALEN);
368 : 0 : SpinLockRelease(&walrcv->mutex);
369 : 0 : }
370 : :
371 : : /*
372 : : * Start streaming.
373 : : *
374 : : * We'll try to start at the requested starting point and timeline,
375 : : * even if it's different from the server's latest timeline. In case
376 : : * we've already reached the end of the old timeline, the server will
377 : : * finish the streaming immediately, and we will go back to await
378 : : * orders from the startup process. If recovery_target_timeline is
379 : : * 'latest', the startup process will scan pg_wal and find the new
380 : : * history file, bump recovery target timeline, and ask us to restart
381 : : * on the new timeline.
382 : : */
383 : 0 : options.logical = false;
384 : 0 : options.startpoint = startpoint;
385 [ # # ]: 0 : options.slotname = slotname[0] != '\0' ? slotname : NULL;
386 : 0 : options.proto.physical.startpointTLI = startpointTLI;
387 [ # # ]: 0 : if (walrcv_startstreaming(wrconn, &options))
388 : : {
389 [ # # ]: 0 : if (first_stream)
390 [ # # # # ]: 0 : ereport(LOG,
391 : : errmsg("started streaming WAL from primary at %X/%08X on timeline %u",
392 : : LSN_FORMAT_ARGS(startpoint), startpointTLI));
393 : : else
394 [ # # # # ]: 0 : ereport(LOG,
395 : : errmsg("restarted WAL streaming at %X/%08X on timeline %u",
396 : : LSN_FORMAT_ARGS(startpoint), startpointTLI));
397 : 0 : first_stream = false;
398 : :
399 : : /*
400 : : * Switch to STREAMING after a successful connection if current
401 : : * state is CONNECTING. This switch happens after an initial
402 : : * startup, or after a restart as determined by
403 : : * WalRcvWaitForStartPosition().
404 : : */
405 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
406 [ # # ]: 0 : if (walrcv->walRcvState == WALRCV_CONNECTING)
407 : 0 : walrcv->walRcvState = WALRCV_STREAMING;
408 : 0 : SpinLockRelease(&walrcv->mutex);
409 : :
410 : : /* Initialize LogstreamResult and buffers for processing messages */
411 : 0 : LogstreamResult.Write = LogstreamResult.Flush = GetXLogReplayRecPtr(NULL);
412 : 0 : initStringInfo(&reply_message);
413 : :
414 : : /* Initialize nap wakeup times. */
415 : 0 : now = GetCurrentTimestamp();
416 [ # # ]: 0 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
417 : 0 : WalRcvComputeNextWakeup(i, now);
418 : :
419 : : /* Send initial reply/feedback messages. */
420 : 0 : XLogWalRcvSendReply(true, false);
421 : 0 : XLogWalRcvSendHSFeedback(true);
422 : :
423 : : /* Loop until end-of-streaming or error */
424 : 0 : for (;;)
425 : : {
426 : 0 : char *buf;
427 : 0 : int len;
428 : 0 : bool endofwal = false;
429 : 0 : pgsocket wait_fd = PGINVALID_SOCKET;
430 : 0 : int rc;
431 : 0 : TimestampTz nextWakeup;
432 : 0 : long nap;
433 : :
434 : : /*
435 : : * Exit walreceiver if we're not in recovery. This should not
436 : : * happen, but cross-check the status here.
437 : : */
438 [ # # ]: 0 : if (!RecoveryInProgress())
439 [ # # # # ]: 0 : ereport(FATAL,
440 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
441 : : errmsg("cannot continue WAL streaming, recovery has already ended")));
442 : :
443 : : /* Process any requests or signals received recently */
444 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
445 : :
446 [ # # ]: 0 : if (ConfigReloadPending)
447 : : {
448 : 0 : ConfigReloadPending = false;
449 : 0 : ProcessConfigFile(PGC_SIGHUP);
450 : : /* recompute wakeup times */
451 : 0 : now = GetCurrentTimestamp();
452 [ # # ]: 0 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
453 : 0 : WalRcvComputeNextWakeup(i, now);
454 : 0 : XLogWalRcvSendHSFeedback(true);
455 : 0 : }
456 : :
457 : : /* See if we can read data immediately */
458 : 0 : len = walrcv_receive(wrconn, &buf, &wait_fd);
459 [ # # ]: 0 : if (len != 0)
460 : : {
461 : : /*
462 : : * Process the received data, and any subsequent data we
463 : : * can read without blocking.
464 : : */
465 : 0 : for (;;)
466 : : {
467 [ # # ]: 0 : if (len > 0)
468 : : {
469 : : /*
470 : : * Something was received from primary, so adjust
471 : : * the ping and terminate wakeup times.
472 : : */
473 : 0 : now = GetCurrentTimestamp();
474 : 0 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_TERMINATE,
475 : 0 : now);
476 : 0 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_PING, now);
477 : 0 : XLogWalRcvProcessMsg(buf[0], &buf[1], len - 1,
478 : 0 : startpointTLI);
479 : 0 : }
480 [ # # ]: 0 : else if (len == 0)
481 : 0 : break;
482 [ # # ]: 0 : else if (len < 0)
483 : : {
484 [ # # # # ]: 0 : ereport(LOG,
485 : : (errmsg("replication terminated by primary server"),
486 : : errdetail("End of WAL reached on timeline %u at %X/%08X.",
487 : : startpointTLI,
488 : : LSN_FORMAT_ARGS(LogstreamResult.Write))));
489 : 0 : endofwal = true;
490 : 0 : break;
491 : : }
492 : 0 : len = walrcv_receive(wrconn, &buf, &wait_fd);
493 : : }
494 : :
495 : : /* Let the primary know that we received some data. */
496 : 0 : XLogWalRcvSendReply(false, false);
497 : :
498 : : /*
499 : : * If we've written some records, flush them to disk and
500 : : * let the startup process and primary server know about
501 : : * them.
502 : : */
503 : 0 : XLogWalRcvFlush(false, startpointTLI);
504 : 0 : }
505 : :
506 : : /* Check if we need to exit the streaming loop. */
507 [ # # ]: 0 : if (endofwal)
508 : 0 : break;
509 : :
510 : : /* Find the soonest wakeup time, to limit our nap. */
511 : 0 : nextWakeup = TIMESTAMP_INFINITY;
512 [ # # ]: 0 : for (int i = 0; i < NUM_WALRCV_WAKEUPS; ++i)
513 [ # # ]: 0 : nextWakeup = Min(wakeup[i], nextWakeup);
514 : :
515 : : /* Calculate the nap time, clamping as necessary. */
516 : 0 : now = GetCurrentTimestamp();
517 : 0 : nap = TimestampDifferenceMilliseconds(now, nextWakeup);
518 : :
519 : : /*
520 : : * Ideally we would reuse a WaitEventSet object repeatedly
521 : : * here to avoid the overheads of WaitLatchOrSocket on epoll
522 : : * systems, but we can't be sure that libpq (or any other
523 : : * walreceiver implementation) has the same socket (even if
524 : : * the fd is the same number, it may have been closed and
525 : : * reopened since the last time). In future, if there is a
526 : : * function for removing sockets from WaitEventSet, then we
527 : : * could add and remove just the socket each time, potentially
528 : : * avoiding some system calls.
529 : : */
530 [ # # ]: 0 : Assert(wait_fd != PGINVALID_SOCKET);
531 : 0 : rc = WaitLatchOrSocket(MyLatch,
532 : : WL_EXIT_ON_PM_DEATH | WL_SOCKET_READABLE |
533 : : WL_TIMEOUT | WL_LATCH_SET,
534 : 0 : wait_fd,
535 : 0 : nap,
536 : : WAIT_EVENT_WAL_RECEIVER_MAIN);
537 [ # # ]: 0 : if (rc & WL_LATCH_SET)
538 : : {
539 : 0 : ResetLatch(MyLatch);
540 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
541 : :
542 [ # # ]: 0 : if (walrcv->force_reply)
543 : : {
544 : : /*
545 : : * The recovery process has asked us to send apply
546 : : * feedback now. Make sure the flag is really set to
547 : : * false in shared memory before sending the reply, so
548 : : * we don't miss a new request for a reply.
549 : : */
550 : 0 : walrcv->force_reply = false;
551 : 0 : pg_memory_barrier();
552 : 0 : XLogWalRcvSendReply(true, false);
553 : 0 : }
554 : 0 : }
555 [ # # ]: 0 : if (rc & WL_TIMEOUT)
556 : : {
557 : : /*
558 : : * We didn't receive anything new. If we haven't heard
559 : : * anything from the server for more than
560 : : * wal_receiver_timeout / 2, ping the server. Also, if
561 : : * it's been longer than wal_receiver_status_interval
562 : : * since the last update we sent, send a status update to
563 : : * the primary anyway, to report any progress in applying
564 : : * WAL.
565 : : */
566 : 0 : bool requestReply = false;
567 : :
568 : : /*
569 : : * Report pending statistics to the cumulative stats
570 : : * system. This location is useful for the report as it
571 : : * is not within a tight loop in the WAL receiver, to
572 : : * avoid bloating pgstats with requests, while also making
573 : : * sure that the reports happen each time a status update
574 : : * is sent.
575 : : */
576 : 0 : pgstat_report_wal(false);
577 : :
578 : : /*
579 : : * Check if time since last receive from primary has
580 : : * reached the configured limit.
581 : : */
582 : 0 : now = GetCurrentTimestamp();
583 [ # # ]: 0 : if (now >= wakeup[WALRCV_WAKEUP_TERMINATE])
584 [ # # # # ]: 0 : ereport(ERROR,
585 : : (errcode(ERRCODE_CONNECTION_FAILURE),
586 : : errmsg("terminating walreceiver due to timeout")));
587 : :
588 : : /*
589 : : * If we didn't receive anything new for half of receiver
590 : : * replication timeout, then ping the server.
591 : : */
592 [ # # ]: 0 : if (now >= wakeup[WALRCV_WAKEUP_PING])
593 : : {
594 : 0 : requestReply = true;
595 : 0 : wakeup[WALRCV_WAKEUP_PING] = TIMESTAMP_INFINITY;
596 : 0 : }
597 : :
598 : 0 : XLogWalRcvSendReply(requestReply, requestReply);
599 : 0 : XLogWalRcvSendHSFeedback(false);
600 : 0 : }
601 [ # # # ]: 0 : }
602 : :
603 : : /*
604 : : * The backend finished streaming. Exit streaming COPY-mode from
605 : : * our side, too.
606 : : */
607 : 0 : walrcv_endstreaming(wrconn, &primaryTLI);
608 : :
609 : : /*
610 : : * If the server had switched to a new timeline that we didn't
611 : : * know about when we began streaming, fetch its timeline history
612 : : * file now.
613 : : */
614 : 0 : WalRcvFetchTimeLineHistoryFiles(startpointTLI, primaryTLI);
615 : 0 : }
616 : : else
617 [ # # # # ]: 0 : ereport(LOG,
618 : : (errmsg("primary server contains no more WAL on requested timeline %u",
619 : : startpointTLI)));
620 : :
621 : : /*
622 : : * End of WAL reached on the requested timeline. Close the last
623 : : * segment, and await for new orders from the startup process.
624 : : */
625 [ # # ]: 0 : if (recvFile >= 0)
626 : : {
627 : 0 : char xlogfname[MAXFNAMELEN];
628 : :
629 : 0 : XLogWalRcvFlush(false, startpointTLI);
630 : 0 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
631 [ # # ]: 0 : if (close(recvFile) != 0)
632 [ # # # # ]: 0 : ereport(PANIC,
633 : : (errcode_for_file_access(),
634 : : errmsg("could not close WAL segment %s: %m",
635 : : xlogfname)));
636 : :
637 : : /*
638 : : * Create .done file forcibly to prevent the streamed segment from
639 : : * being archived later.
640 : : */
641 [ # # ]: 0 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
642 : 0 : XLogArchiveForceDone(xlogfname);
643 : : else
644 : 0 : XLogArchiveNotify(xlogfname);
645 : 0 : }
646 : 0 : recvFile = -1;
647 : :
648 [ # # # # ]: 0 : elog(DEBUG1, "walreceiver ended streaming and awaits new instructions");
649 : 0 : WalRcvWaitForStartPosition(&startpoint, &startpointTLI);
650 : 0 : }
651 : : /* not reached */
652 : : }
653 : :
654 : : /*
655 : : * Wait for startup process to set receiveStart and receiveStartTLI.
656 : : */
657 : : static void
658 : 0 : WalRcvWaitForStartPosition(XLogRecPtr *startpoint, TimeLineID *startpointTLI)
659 : : {
660 : 0 : WalRcvData *walrcv = WalRcv;
661 : 0 : int state;
662 : :
663 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
664 : 0 : state = walrcv->walRcvState;
665 [ # # # # ]: 0 : if (state != WALRCV_STREAMING && state != WALRCV_CONNECTING)
666 : : {
667 : 0 : SpinLockRelease(&walrcv->mutex);
668 [ # # ]: 0 : if (state == WALRCV_STOPPING)
669 : 0 : proc_exit(0);
670 : : else
671 [ # # # # ]: 0 : elog(FATAL, "unexpected walreceiver state");
672 : 0 : }
673 : 0 : walrcv->walRcvState = WALRCV_WAITING;
674 : 0 : walrcv->receiveStart = InvalidXLogRecPtr;
675 : 0 : walrcv->receiveStartTLI = 0;
676 : 0 : SpinLockRelease(&walrcv->mutex);
677 : :
678 : 0 : set_ps_display("idle");
679 : :
680 : : /*
681 : : * nudge startup process to notice that we've stopped streaming and are
682 : : * now waiting for instructions.
683 : : */
684 : 0 : WakeupRecovery();
685 : 0 : for (;;)
686 : : {
687 : 0 : ResetLatch(MyLatch);
688 : :
689 [ # # ]: 0 : CHECK_FOR_INTERRUPTS();
690 : :
691 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
692 [ # # # # : 0 : Assert(walrcv->walRcvState == WALRCV_RESTARTING ||
# # ]
693 : : walrcv->walRcvState == WALRCV_WAITING ||
694 : : walrcv->walRcvState == WALRCV_STOPPING);
695 [ # # ]: 0 : if (walrcv->walRcvState == WALRCV_RESTARTING)
696 : : {
697 : : /*
698 : : * No need to handle changes in primary_conninfo or
699 : : * primary_slot_name here. Startup process will signal us to
700 : : * terminate in case those change.
701 : : */
702 : 0 : *startpoint = walrcv->receiveStart;
703 : 0 : *startpointTLI = walrcv->receiveStartTLI;
704 : 0 : walrcv->walRcvState = WALRCV_CONNECTING;
705 : 0 : SpinLockRelease(&walrcv->mutex);
706 : 0 : break;
707 : : }
708 [ # # ]: 0 : if (walrcv->walRcvState == WALRCV_STOPPING)
709 : : {
710 : : /*
711 : : * We should've received SIGTERM if the startup process wants us
712 : : * to die, but might as well check it here too.
713 : : */
714 : 0 : SpinLockRelease(&walrcv->mutex);
715 : 0 : exit(1);
716 : : }
717 : 0 : SpinLockRelease(&walrcv->mutex);
718 : :
719 : 0 : (void) WaitLatch(MyLatch, WL_LATCH_SET | WL_EXIT_ON_PM_DEATH, 0,
720 : : WAIT_EVENT_WAL_RECEIVER_WAIT_START);
721 : : }
722 : :
723 [ # # ]: 0 : if (update_process_title)
724 : : {
725 : 0 : char activitymsg[50];
726 : :
727 : 0 : snprintf(activitymsg, sizeof(activitymsg), "restarting at %X/%08X",
728 : 0 : LSN_FORMAT_ARGS(*startpoint));
729 : 0 : set_ps_display(activitymsg);
730 : 0 : }
731 : 0 : }
732 : :
733 : : /*
734 : : * Fetch any missing timeline history files between 'first' and 'last'
735 : : * (inclusive) from the server.
736 : : */
737 : : static void
738 : 0 : WalRcvFetchTimeLineHistoryFiles(TimeLineID first, TimeLineID last)
739 : : {
740 : 0 : TimeLineID tli;
741 : :
742 [ # # ]: 0 : for (tli = first; tli <= last; tli++)
743 : : {
744 : : /* there's no history file for timeline 1 */
745 [ # # # # ]: 0 : if (tli != 1 && !existsTimeLineHistory(tli))
746 : : {
747 : 0 : char *fname;
748 : 0 : char *content;
749 : 0 : int len;
750 : 0 : char expectedfname[MAXFNAMELEN];
751 : :
752 [ # # # # ]: 0 : ereport(LOG,
753 : : (errmsg("fetching timeline history file for timeline %u from primary server",
754 : : tli)));
755 : :
756 : 0 : walrcv_readtimelinehistoryfile(wrconn, tli, &fname, &content, &len);
757 : :
758 : : /*
759 : : * Check that the filename on the primary matches what we
760 : : * calculated ourselves. This is just a sanity check, it should
761 : : * always match.
762 : : */
763 : 0 : TLHistoryFileName(expectedfname, tli);
764 [ # # ]: 0 : if (strcmp(fname, expectedfname) != 0)
765 [ # # # # ]: 0 : ereport(ERROR,
766 : : (errcode(ERRCODE_PROTOCOL_VIOLATION),
767 : : errmsg_internal("primary reported unexpected file name for timeline history file of timeline %u",
768 : : tli)));
769 : :
770 : : /*
771 : : * Write the file to pg_wal.
772 : : */
773 : 0 : writeTimeLineHistoryFile(tli, content, len);
774 : :
775 : : /*
776 : : * Mark the streamed history file as ready for archiving if
777 : : * archive_mode is always.
778 : : */
779 [ # # ]: 0 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
780 : 0 : XLogArchiveForceDone(fname);
781 : : else
782 : 0 : XLogArchiveNotify(fname);
783 : :
784 : 0 : pfree(fname);
785 : 0 : pfree(content);
786 : 0 : }
787 : 0 : }
788 : 0 : }
789 : :
790 : : /*
791 : : * Mark us as STOPPED in shared memory at exit.
792 : : */
793 : : static void
794 : 0 : WalRcvDie(int code, Datum arg)
795 : : {
796 : 0 : WalRcvData *walrcv = WalRcv;
797 : 0 : TimeLineID *startpointTLI_p = (TimeLineID *) DatumGetPointer(arg);
798 : :
799 [ # # ]: 0 : Assert(*startpointTLI_p != 0);
800 : :
801 : : /* Ensure that all WAL records received are flushed to disk */
802 : 0 : XLogWalRcvFlush(true, *startpointTLI_p);
803 : :
804 : : /* Mark ourselves inactive in shared memory */
805 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
806 [ # # # # : 0 : Assert(walrcv->walRcvState == WALRCV_STREAMING ||
# # # # #
# # # ]
807 : : walrcv->walRcvState == WALRCV_CONNECTING ||
808 : : walrcv->walRcvState == WALRCV_RESTARTING ||
809 : : walrcv->walRcvState == WALRCV_STARTING ||
810 : : walrcv->walRcvState == WALRCV_WAITING ||
811 : : walrcv->walRcvState == WALRCV_STOPPING);
812 [ # # ]: 0 : Assert(walrcv->pid == MyProcPid);
813 : 0 : walrcv->walRcvState = WALRCV_STOPPED;
814 : 0 : walrcv->pid = 0;
815 : 0 : walrcv->procno = INVALID_PROC_NUMBER;
816 : 0 : walrcv->ready_to_display = false;
817 : 0 : SpinLockRelease(&walrcv->mutex);
818 : :
819 : 0 : ConditionVariableBroadcast(&walrcv->walRcvStoppedCV);
820 : :
821 : : /* Terminate the connection gracefully. */
822 [ # # ]: 0 : if (wrconn != NULL)
823 : 0 : walrcv_disconnect(wrconn);
824 : :
825 : : /* Wake up the startup process to notice promptly that we're gone */
826 : 0 : WakeupRecovery();
827 : 0 : }
828 : :
829 : : /*
830 : : * Accept the message from XLOG stream, and process it.
831 : : */
832 : : static void
833 : 0 : XLogWalRcvProcessMsg(unsigned char type, char *buf, Size len, TimeLineID tli)
834 : : {
835 : 0 : int hdrlen;
836 : 0 : XLogRecPtr dataStart;
837 : 0 : XLogRecPtr walEnd;
838 : 0 : TimestampTz sendTime;
839 : 0 : bool replyRequested;
840 : :
841 [ # # # ]: 0 : switch (type)
842 : : {
843 : : case PqReplMsg_WALData:
844 : : {
845 : 0 : StringInfoData incoming_message;
846 : :
847 : 0 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(int64);
848 [ # # ]: 0 : if (len < hdrlen)
849 [ # # # # ]: 0 : ereport(ERROR,
850 : : (errcode(ERRCODE_PROTOCOL_VIOLATION),
851 : : errmsg_internal("invalid WAL message received from primary")));
852 : :
853 : : /* initialize a StringInfo with the given buffer */
854 : 0 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
855 : :
856 : : /* read the fields */
857 : 0 : dataStart = pq_getmsgint64(&incoming_message);
858 : 0 : walEnd = pq_getmsgint64(&incoming_message);
859 : 0 : sendTime = pq_getmsgint64(&incoming_message);
860 : 0 : ProcessWalSndrMessage(walEnd, sendTime);
861 : :
862 : 0 : buf += hdrlen;
863 : 0 : len -= hdrlen;
864 : 0 : XLogWalRcvWrite(buf, len, dataStart, tli);
865 : : break;
866 : 0 : }
867 : : case PqReplMsg_Keepalive:
868 : : {
869 : 0 : StringInfoData incoming_message;
870 : :
871 : 0 : hdrlen = sizeof(int64) + sizeof(int64) + sizeof(char);
872 [ # # ]: 0 : if (len != hdrlen)
873 [ # # # # ]: 0 : ereport(ERROR,
874 : : (errcode(ERRCODE_PROTOCOL_VIOLATION),
875 : : errmsg_internal("invalid keepalive message received from primary")));
876 : :
877 : : /* initialize a StringInfo with the given buffer */
878 : 0 : initReadOnlyStringInfo(&incoming_message, buf, hdrlen);
879 : :
880 : : /* read the fields */
881 : 0 : walEnd = pq_getmsgint64(&incoming_message);
882 : 0 : sendTime = pq_getmsgint64(&incoming_message);
883 : 0 : replyRequested = pq_getmsgbyte(&incoming_message);
884 : :
885 : 0 : ProcessWalSndrMessage(walEnd, sendTime);
886 : :
887 : : /* If the primary requested a reply, send one immediately */
888 [ # # ]: 0 : if (replyRequested)
889 : 0 : XLogWalRcvSendReply(true, false);
890 : : break;
891 : 0 : }
892 : : default:
893 [ # # # # ]: 0 : ereport(ERROR,
894 : : (errcode(ERRCODE_PROTOCOL_VIOLATION),
895 : : errmsg_internal("invalid replication message type %d",
896 : : type)));
897 : 0 : }
898 : 0 : }
899 : :
900 : : /*
901 : : * Write XLOG data to disk.
902 : : */
903 : : static void
904 : 0 : XLogWalRcvWrite(char *buf, Size nbytes, XLogRecPtr recptr, TimeLineID tli)
905 : : {
906 : 0 : int startoff;
907 : 0 : int byteswritten;
908 : 0 : instr_time start;
909 : :
910 [ # # ]: 0 : Assert(tli != 0);
911 : :
912 [ # # ]: 0 : while (nbytes > 0)
913 : : {
914 : 0 : int segbytes;
915 : :
916 : : /* Close the current segment if it's completed */
917 [ # # # # ]: 0 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
918 : 0 : XLogWalRcvClose(recptr, tli);
919 : :
920 [ # # ]: 0 : if (recvFile < 0)
921 : : {
922 : : /* Create/use new log file */
923 : 0 : XLByteToSeg(recptr, recvSegNo, wal_segment_size);
924 : 0 : recvFile = XLogFileInit(recvSegNo, tli);
925 : 0 : recvFileTLI = tli;
926 : 0 : }
927 : :
928 : : /* Calculate the start offset of the received logs */
929 : 0 : startoff = XLogSegmentOffset(recptr, wal_segment_size);
930 : :
931 [ # # ]: 0 : if (startoff + nbytes > wal_segment_size)
932 : 0 : segbytes = wal_segment_size - startoff;
933 : : else
934 : 0 : segbytes = nbytes;
935 : :
936 : : /* OK to write the logs */
937 : 0 : errno = 0;
938 : :
939 : : /*
940 : : * Measure I/O timing to write WAL data, for pg_stat_io.
941 : : */
942 : 0 : start = pgstat_prepare_io_time(track_wal_io_timing);
943 : :
944 : 0 : pgstat_report_wait_start(WAIT_EVENT_WAL_WRITE);
945 : 0 : byteswritten = pg_pwrite(recvFile, buf, segbytes, (pgoff_t) startoff);
946 : 0 : pgstat_report_wait_end();
947 : :
948 : 0 : pgstat_count_io_op_time(IOOBJECT_WAL, IOCONTEXT_NORMAL,
949 : 0 : IOOP_WRITE, start, 1, byteswritten);
950 : :
951 [ # # ]: 0 : if (byteswritten <= 0)
952 : : {
953 : 0 : char xlogfname[MAXFNAMELEN];
954 : 0 : int save_errno;
955 : :
956 : : /* if write didn't set errno, assume no disk space */
957 [ # # ]: 0 : if (errno == 0)
958 : 0 : errno = ENOSPC;
959 : :
960 : 0 : save_errno = errno;
961 : 0 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
962 : 0 : errno = save_errno;
963 [ # # # # ]: 0 : ereport(PANIC,
964 : : (errcode_for_file_access(),
965 : : errmsg("could not write to WAL segment %s "
966 : : "at offset %d, length %d: %m",
967 : : xlogfname, startoff, segbytes)));
968 : 0 : }
969 : :
970 : : /* Update state for write */
971 : 0 : recptr += byteswritten;
972 : :
973 : 0 : nbytes -= byteswritten;
974 : 0 : buf += byteswritten;
975 : :
976 : 0 : LogstreamResult.Write = recptr;
977 : 0 : }
978 : :
979 : : /* Update shared-memory status */
980 : 0 : pg_atomic_write_u64(&WalRcv->writtenUpto, LogstreamResult.Write);
981 : :
982 : : /*
983 : : * If we wrote an LSN that someone was waiting for, notify the waiters.
984 : : */
985 [ # # # # ]: 0 : if (waitLSNState &&
986 : 0 : (LogstreamResult.Write >=
987 : 0 : pg_atomic_read_u64(&waitLSNState->minWaitedLSN[WAIT_LSN_TYPE_STANDBY_WRITE])))
988 : 0 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_WRITE, LogstreamResult.Write);
989 : :
990 : : /*
991 : : * Close the current segment if it's fully written up in the last cycle of
992 : : * the loop, to create its archive notification file soon. Otherwise WAL
993 : : * archiving of the segment will be delayed until any data in the next
994 : : * segment is received and written.
995 : : */
996 [ # # # # ]: 0 : if (recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size))
997 : 0 : XLogWalRcvClose(recptr, tli);
998 : 0 : }
999 : :
1000 : : /*
1001 : : * Flush the log to disk.
1002 : : *
1003 : : * If we're in the midst of dying, it's unwise to do anything that might throw
1004 : : * an error, so we skip sending a reply in that case.
1005 : : */
1006 : : static void
1007 : 0 : XLogWalRcvFlush(bool dying, TimeLineID tli)
1008 : : {
1009 [ # # ]: 0 : Assert(tli != 0);
1010 : :
1011 [ # # ]: 0 : if (LogstreamResult.Flush < LogstreamResult.Write)
1012 : : {
1013 : 0 : WalRcvData *walrcv = WalRcv;
1014 : :
1015 : 0 : issue_xlog_fsync(recvFile, recvSegNo, tli);
1016 : :
1017 : 0 : LogstreamResult.Flush = LogstreamResult.Write;
1018 : :
1019 : : /* Update shared-memory status */
1020 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
1021 [ # # ]: 0 : if (walrcv->flushedUpto < LogstreamResult.Flush)
1022 : : {
1023 : 0 : walrcv->latestChunkStart = walrcv->flushedUpto;
1024 : 0 : walrcv->flushedUpto = LogstreamResult.Flush;
1025 : 0 : walrcv->receivedTLI = tli;
1026 : 0 : }
1027 : 0 : SpinLockRelease(&walrcv->mutex);
1028 : :
1029 : : /*
1030 : : * If we flushed an LSN that someone was waiting for, notify the
1031 : : * waiters.
1032 : : */
1033 [ # # # # ]: 0 : if (waitLSNState &&
1034 : 0 : (LogstreamResult.Flush >=
1035 : 0 : pg_atomic_read_u64(&waitLSNState->minWaitedLSN[WAIT_LSN_TYPE_STANDBY_FLUSH])))
1036 : 0 : WaitLSNWakeup(WAIT_LSN_TYPE_STANDBY_FLUSH, LogstreamResult.Flush);
1037 : :
1038 : : /* Signal the startup process and walsender that new WAL has arrived */
1039 : 0 : WakeupRecovery();
1040 [ # # # # ]: 0 : if (AllowCascadeReplication())
1041 : 0 : WalSndWakeup(true, false);
1042 : :
1043 : : /* Report XLOG streaming progress in PS display */
1044 [ # # ]: 0 : if (update_process_title)
1045 : : {
1046 : 0 : char activitymsg[50];
1047 : :
1048 : 0 : snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%08X",
1049 : 0 : LSN_FORMAT_ARGS(LogstreamResult.Write));
1050 : 0 : set_ps_display(activitymsg);
1051 : 0 : }
1052 : :
1053 : : /* Also let the primary know that we made some progress */
1054 [ # # ]: 0 : if (!dying)
1055 : : {
1056 : 0 : XLogWalRcvSendReply(false, false);
1057 : 0 : XLogWalRcvSendHSFeedback(false);
1058 : 0 : }
1059 : 0 : }
1060 : 0 : }
1061 : :
1062 : : /*
1063 : : * Close the current segment.
1064 : : *
1065 : : * Flush the segment to disk before closing it. Otherwise we have to
1066 : : * reopen and fsync it later.
1067 : : *
1068 : : * Create an archive notification file since the segment is known completed.
1069 : : */
1070 : : static void
1071 : 0 : XLogWalRcvClose(XLogRecPtr recptr, TimeLineID tli)
1072 : : {
1073 : 0 : char xlogfname[MAXFNAMELEN];
1074 : :
1075 [ # # ]: 0 : Assert(recvFile >= 0 && !XLByteInSeg(recptr, recvSegNo, wal_segment_size));
1076 [ # # ]: 0 : Assert(tli != 0);
1077 : :
1078 : : /*
1079 : : * fsync() and close current file before we switch to next one. We would
1080 : : * otherwise have to reopen this file to fsync it later
1081 : : */
1082 : 0 : XLogWalRcvFlush(false, tli);
1083 : :
1084 : 0 : XLogFileName(xlogfname, recvFileTLI, recvSegNo, wal_segment_size);
1085 : :
1086 : : /*
1087 : : * XLOG segment files will be re-read by recovery in startup process soon,
1088 : : * so we don't advise the OS to release cache pages associated with the
1089 : : * file like XLogFileClose() does.
1090 : : */
1091 [ # # ]: 0 : if (close(recvFile) != 0)
1092 [ # # # # ]: 0 : ereport(PANIC,
1093 : : (errcode_for_file_access(),
1094 : : errmsg("could not close WAL segment %s: %m",
1095 : : xlogfname)));
1096 : :
1097 : : /*
1098 : : * Create .done file forcibly to prevent the streamed segment from being
1099 : : * archived later.
1100 : : */
1101 [ # # ]: 0 : if (XLogArchiveMode != ARCHIVE_MODE_ALWAYS)
1102 : 0 : XLogArchiveForceDone(xlogfname);
1103 : : else
1104 : 0 : XLogArchiveNotify(xlogfname);
1105 : :
1106 : 0 : recvFile = -1;
1107 : 0 : }
1108 : :
1109 : : /*
1110 : : * Send reply message to primary, indicating our current WAL locations, oldest
1111 : : * xmin and the current time.
1112 : : *
1113 : : * If 'force' is not set, the message is only sent if enough time has
1114 : : * passed since last status update to reach wal_receiver_status_interval.
1115 : : * If wal_receiver_status_interval is disabled altogether and 'force' is
1116 : : * false, this is a no-op.
1117 : : *
1118 : : * If 'requestReply' is true, requests the server to reply immediately upon
1119 : : * receiving this message. This is used for heartbeats, when approaching
1120 : : * wal_receiver_timeout.
1121 : : */
1122 : : static void
1123 : 0 : XLogWalRcvSendReply(bool force, bool requestReply)
1124 : : {
1125 : : static XLogRecPtr writePtr = 0;
1126 : : static XLogRecPtr flushPtr = 0;
1127 : 0 : XLogRecPtr applyPtr;
1128 : 0 : TimestampTz now;
1129 : :
1130 : : /*
1131 : : * If the user doesn't want status to be reported to the primary, be sure
1132 : : * to exit before doing anything at all.
1133 : : */
1134 [ # # # # ]: 0 : if (!force && wal_receiver_status_interval <= 0)
1135 : 0 : return;
1136 : :
1137 : : /* Get current timestamp. */
1138 : 0 : now = GetCurrentTimestamp();
1139 : :
1140 : : /*
1141 : : * We can compare the write and flush positions to the last message we
1142 : : * sent without taking any lock, but the apply position requires a spin
1143 : : * lock, so we don't check that unless something else has changed or 10
1144 : : * seconds have passed. This means that the apply WAL location will
1145 : : * appear, from the primary's point of view, to lag slightly, but since
1146 : : * this is only for reporting purposes and only on idle systems, that's
1147 : : * probably OK.
1148 : : */
1149 : 0 : if (!force
1150 [ # # ]: 0 : && writePtr == LogstreamResult.Write
1151 [ # # ]: 0 : && flushPtr == LogstreamResult.Flush
1152 [ # # # # ]: 0 : && now < wakeup[WALRCV_WAKEUP_REPLY])
1153 : 0 : return;
1154 : :
1155 : : /* Make sure we wake up when it's time to send another reply. */
1156 : 0 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_REPLY, now);
1157 : :
1158 : : /* Construct a new message */
1159 : 0 : writePtr = LogstreamResult.Write;
1160 : 0 : flushPtr = LogstreamResult.Flush;
1161 : 0 : applyPtr = GetXLogReplayRecPtr(NULL);
1162 : :
1163 : 0 : resetStringInfo(&reply_message);
1164 : 0 : pq_sendbyte(&reply_message, PqReplMsg_StandbyStatusUpdate);
1165 : 0 : pq_sendint64(&reply_message, writePtr);
1166 : 0 : pq_sendint64(&reply_message, flushPtr);
1167 : 0 : pq_sendint64(&reply_message, applyPtr);
1168 : 0 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1169 : 0 : pq_sendbyte(&reply_message, requestReply ? 1 : 0);
1170 : :
1171 : : /* Send it */
1172 [ # # # # ]: 0 : elog(DEBUG2, "sending write %X/%08X flush %X/%08X apply %X/%08X%s",
1173 : : LSN_FORMAT_ARGS(writePtr),
1174 : : LSN_FORMAT_ARGS(flushPtr),
1175 : : LSN_FORMAT_ARGS(applyPtr),
1176 : : requestReply ? " (reply requested)" : "");
1177 : :
1178 : 0 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1179 [ # # ]: 0 : }
1180 : :
1181 : : /*
1182 : : * Send hot standby feedback message to primary, plus the current time,
1183 : : * in case they don't have a watch.
1184 : : *
1185 : : * If the user disables feedback, send one final message to tell sender
1186 : : * to forget about the xmin on this standby. We also send this message
1187 : : * on first connect because a previous connection might have set xmin
1188 : : * on a replication slot. (If we're not using a slot it's harmless to
1189 : : * send a feedback message explicitly setting InvalidTransactionId).
1190 : : */
1191 : : static void
1192 : 0 : XLogWalRcvSendHSFeedback(bool immed)
1193 : : {
1194 : 0 : TimestampTz now;
1195 : 0 : FullTransactionId nextFullXid;
1196 : 0 : TransactionId nextXid;
1197 : 0 : uint32 xmin_epoch,
1198 : : catalog_xmin_epoch;
1199 : 0 : TransactionId xmin,
1200 : : catalog_xmin;
1201 : :
1202 : : /* initially true so we always send at least one feedback message */
1203 : : static bool primary_has_standby_xmin = true;
1204 : :
1205 : : /*
1206 : : * If the user doesn't want status to be reported to the primary, be sure
1207 : : * to exit before doing anything at all.
1208 : : */
1209 [ # # # # ]: 0 : if ((wal_receiver_status_interval <= 0 || !hot_standby_feedback) &&
1210 : 0 : !primary_has_standby_xmin)
1211 : 0 : return;
1212 : :
1213 : : /* Get current timestamp. */
1214 : 0 : now = GetCurrentTimestamp();
1215 : :
1216 : : /* Send feedback at most once per wal_receiver_status_interval. */
1217 [ # # # # ]: 0 : if (!immed && now < wakeup[WALRCV_WAKEUP_HSFEEDBACK])
1218 : 0 : return;
1219 : :
1220 : : /* Make sure we wake up when it's time to send feedback again. */
1221 : 0 : WalRcvComputeNextWakeup(WALRCV_WAKEUP_HSFEEDBACK, now);
1222 : :
1223 : : /*
1224 : : * If Hot Standby is not yet accepting connections there is nothing to
1225 : : * send. Check this after the interval has expired to reduce number of
1226 : : * calls.
1227 : : *
1228 : : * Bailing out here also ensures that we don't send feedback until we've
1229 : : * read our own replication slot state, so we don't tell the primary to
1230 : : * discard needed xmin or catalog_xmin from any slots that may exist on
1231 : : * this replica.
1232 : : */
1233 [ # # ]: 0 : if (!HotStandbyActive())
1234 : 0 : return;
1235 : :
1236 : : /*
1237 : : * Make the expensive call to get the oldest xmin once we are certain
1238 : : * everything else has been checked.
1239 : : */
1240 [ # # ]: 0 : if (hot_standby_feedback)
1241 : : {
1242 : 0 : GetReplicationHorizons(&xmin, &catalog_xmin);
1243 : 0 : }
1244 : : else
1245 : : {
1246 : 0 : xmin = InvalidTransactionId;
1247 : 0 : catalog_xmin = InvalidTransactionId;
1248 : : }
1249 : :
1250 : : /*
1251 : : * Get epoch and adjust if nextXid and oldestXmin are different sides of
1252 : : * the epoch boundary.
1253 : : */
1254 : 0 : nextFullXid = ReadNextFullTransactionId();
1255 : 0 : nextXid = XidFromFullTransactionId(nextFullXid);
1256 : 0 : xmin_epoch = EpochFromFullTransactionId(nextFullXid);
1257 : 0 : catalog_xmin_epoch = xmin_epoch;
1258 [ # # ]: 0 : if (nextXid < xmin)
1259 : 0 : xmin_epoch--;
1260 [ # # ]: 0 : if (nextXid < catalog_xmin)
1261 : 0 : catalog_xmin_epoch--;
1262 : :
1263 [ # # # # ]: 0 : elog(DEBUG2, "sending hot standby feedback xmin %u epoch %u catalog_xmin %u catalog_xmin_epoch %u",
1264 : : xmin, xmin_epoch, catalog_xmin, catalog_xmin_epoch);
1265 : :
1266 : : /* Construct the message and send it. */
1267 : 0 : resetStringInfo(&reply_message);
1268 : 0 : pq_sendbyte(&reply_message, PqReplMsg_HotStandbyFeedback);
1269 : 0 : pq_sendint64(&reply_message, GetCurrentTimestamp());
1270 : 0 : pq_sendint32(&reply_message, xmin);
1271 : 0 : pq_sendint32(&reply_message, xmin_epoch);
1272 : 0 : pq_sendint32(&reply_message, catalog_xmin);
1273 : 0 : pq_sendint32(&reply_message, catalog_xmin_epoch);
1274 : 0 : walrcv_send(wrconn, reply_message.data, reply_message.len);
1275 [ # # # # ]: 0 : if (TransactionIdIsValid(xmin) || TransactionIdIsValid(catalog_xmin))
1276 : 0 : primary_has_standby_xmin = true;
1277 : : else
1278 : 0 : primary_has_standby_xmin = false;
1279 [ # # ]: 0 : }
1280 : :
1281 : : /*
1282 : : * Update shared memory status upon receiving a message from primary.
1283 : : *
1284 : : * 'walEnd' and 'sendTime' are the end-of-WAL and timestamp of the latest
1285 : : * message, reported by primary.
1286 : : */
1287 : : static void
1288 : 0 : ProcessWalSndrMessage(XLogRecPtr walEnd, TimestampTz sendTime)
1289 : : {
1290 : 0 : WalRcvData *walrcv = WalRcv;
1291 : 0 : TimestampTz lastMsgReceiptTime = GetCurrentTimestamp();
1292 : :
1293 : : /* Update shared-memory status */
1294 [ # # ]: 0 : SpinLockAcquire(&walrcv->mutex);
1295 [ # # ]: 0 : if (walrcv->latestWalEnd < walEnd)
1296 : 0 : walrcv->latestWalEndTime = sendTime;
1297 : 0 : walrcv->latestWalEnd = walEnd;
1298 : 0 : walrcv->lastMsgSendTime = sendTime;
1299 : 0 : walrcv->lastMsgReceiptTime = lastMsgReceiptTime;
1300 : 0 : SpinLockRelease(&walrcv->mutex);
1301 : :
1302 [ # # ]: 0 : if (message_level_is_interesting(DEBUG2))
1303 : : {
1304 : 0 : char *sendtime;
1305 : 0 : char *receipttime;
1306 : 0 : int applyDelay;
1307 : :
1308 : : /* Copy because timestamptz_to_str returns a static buffer */
1309 : 0 : sendtime = pstrdup(timestamptz_to_str(sendTime));
1310 : 0 : receipttime = pstrdup(timestamptz_to_str(lastMsgReceiptTime));
1311 : 0 : applyDelay = GetReplicationApplyDelay();
1312 : :
1313 : : /* apply delay is not available */
1314 [ # # ]: 0 : if (applyDelay == -1)
1315 [ # # # # ]: 0 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay (N/A) transfer latency %d ms",
1316 : : sendtime,
1317 : : receipttime,
1318 : : GetReplicationTransferLatency());
1319 : : else
1320 [ # # # # ]: 0 : elog(DEBUG2, "sendtime %s receipttime %s replication apply delay %d ms transfer latency %d ms",
1321 : : sendtime,
1322 : : receipttime,
1323 : : applyDelay,
1324 : : GetReplicationTransferLatency());
1325 : :
1326 : 0 : pfree(sendtime);
1327 : 0 : pfree(receipttime);
1328 : 0 : }
1329 : 0 : }
1330 : :
1331 : : /*
1332 : : * Compute the next wakeup time for a given wakeup reason. Can be called to
1333 : : * initialize a wakeup time, to adjust it for the next wakeup, or to
1334 : : * reinitialize it when GUCs have changed. We ask the caller to pass in the
1335 : : * value of "now" because this frequently avoids multiple calls of
1336 : : * GetCurrentTimestamp(). It had better be a reasonably up-to-date value
1337 : : * though.
1338 : : */
1339 : : static void
1340 : 0 : WalRcvComputeNextWakeup(WalRcvWakeupReason reason, TimestampTz now)
1341 : : {
1342 [ # # # # : 0 : switch (reason)
# ]
1343 : : {
1344 : : case WALRCV_WAKEUP_TERMINATE:
1345 [ # # ]: 0 : if (wal_receiver_timeout <= 0)
1346 : 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1347 : : else
1348 : 0 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout);
1349 : 0 : break;
1350 : : case WALRCV_WAKEUP_PING:
1351 [ # # ]: 0 : if (wal_receiver_timeout <= 0)
1352 : 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1353 : : else
1354 : 0 : wakeup[reason] = TimestampTzPlusMilliseconds(now, wal_receiver_timeout / 2);
1355 : 0 : break;
1356 : : case WALRCV_WAKEUP_HSFEEDBACK:
1357 [ # # # # ]: 0 : if (!hot_standby_feedback || wal_receiver_status_interval <= 0)
1358 : 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1359 : : else
1360 : 0 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1361 : 0 : break;
1362 : : case WALRCV_WAKEUP_REPLY:
1363 [ # # ]: 0 : if (wal_receiver_status_interval <= 0)
1364 : 0 : wakeup[reason] = TIMESTAMP_INFINITY;
1365 : : else
1366 : 0 : wakeup[reason] = TimestampTzPlusSeconds(now, wal_receiver_status_interval);
1367 : 0 : break;
1368 : : /* there's intentionally no default: here */
1369 : : }
1370 : 0 : }
1371 : :
1372 : : /*
1373 : : * Wake up the walreceiver main loop.
1374 : : *
1375 : : * This is called by the startup process whenever interesting xlog records
1376 : : * are applied, so that walreceiver can check if it needs to send an apply
1377 : : * notification back to the primary which may be waiting in a COMMIT with
1378 : : * synchronous_commit = remote_apply.
1379 : : */
1380 : : void
1381 : 0 : WalRcvForceReply(void)
1382 : : {
1383 : 0 : ProcNumber procno;
1384 : :
1385 : 0 : WalRcv->force_reply = true;
1386 : : /* fetching the proc number is probably atomic, but don't rely on it */
1387 [ # # ]: 0 : SpinLockAcquire(&WalRcv->mutex);
1388 : 0 : procno = WalRcv->procno;
1389 : 0 : SpinLockRelease(&WalRcv->mutex);
1390 [ # # ]: 0 : if (procno != INVALID_PROC_NUMBER)
1391 : 0 : SetLatch(&GetPGProcByNumber(procno)->procLatch);
1392 : 0 : }
1393 : :
1394 : : /*
1395 : : * Return a string constant representing the state. This is used
1396 : : * in system functions and views, and should *not* be translated.
1397 : : */
1398 : : static const char *
1399 : 0 : WalRcvGetStateString(WalRcvState state)
1400 : : {
1401 [ # # # # : 0 : switch (state)
# # # # ]
1402 : : {
1403 : : case WALRCV_STOPPED:
1404 : 0 : return "stopped";
1405 : : case WALRCV_STARTING:
1406 : 0 : return "starting";
1407 : : case WALRCV_CONNECTING:
1408 : 0 : return "connecting";
1409 : : case WALRCV_STREAMING:
1410 : 0 : return "streaming";
1411 : : case WALRCV_WAITING:
1412 : 0 : return "waiting";
1413 : : case WALRCV_RESTARTING:
1414 : 0 : return "restarting";
1415 : : case WALRCV_STOPPING:
1416 : 0 : return "stopping";
1417 : : }
1418 : 0 : return "UNKNOWN";
1419 : 0 : }
1420 : :
1421 : : /*
1422 : : * Returns activity of WAL receiver, including pid, state and xlog locations
1423 : : * received from the WAL sender of another server.
1424 : : */
1425 : : Datum
1426 : 0 : pg_stat_get_wal_receiver(PG_FUNCTION_ARGS)
1427 : : {
1428 : 0 : TupleDesc tupdesc;
1429 : 0 : Datum *values;
1430 : 0 : bool *nulls;
1431 : 0 : int pid;
1432 : 0 : bool ready_to_display;
1433 : 0 : WalRcvState state;
1434 : 0 : XLogRecPtr receive_start_lsn;
1435 : 0 : TimeLineID receive_start_tli;
1436 : 0 : XLogRecPtr written_lsn;
1437 : 0 : XLogRecPtr flushed_lsn;
1438 : 0 : TimeLineID received_tli;
1439 : 0 : TimestampTz last_send_time;
1440 : 0 : TimestampTz last_receipt_time;
1441 : 0 : XLogRecPtr latest_end_lsn;
1442 : 0 : TimestampTz latest_end_time;
1443 : 0 : char sender_host[NI_MAXHOST];
1444 : 0 : int sender_port = 0;
1445 : 0 : char slotname[NAMEDATALEN];
1446 : 0 : char conninfo[MAXCONNINFO];
1447 : :
1448 : : /* Take a lock to ensure value consistency */
1449 [ # # ]: 0 : SpinLockAcquire(&WalRcv->mutex);
1450 : 0 : pid = (int) WalRcv->pid;
1451 : 0 : ready_to_display = WalRcv->ready_to_display;
1452 : 0 : state = WalRcv->walRcvState;
1453 : 0 : receive_start_lsn = WalRcv->receiveStart;
1454 : 0 : receive_start_tli = WalRcv->receiveStartTLI;
1455 : 0 : flushed_lsn = WalRcv->flushedUpto;
1456 : 0 : received_tli = WalRcv->receivedTLI;
1457 : 0 : last_send_time = WalRcv->lastMsgSendTime;
1458 : 0 : last_receipt_time = WalRcv->lastMsgReceiptTime;
1459 : 0 : latest_end_lsn = WalRcv->latestWalEnd;
1460 : 0 : latest_end_time = WalRcv->latestWalEndTime;
1461 : 0 : strlcpy(slotname, WalRcv->slotname, sizeof(slotname));
1462 : 0 : strlcpy(sender_host, WalRcv->sender_host, sizeof(sender_host));
1463 : 0 : sender_port = WalRcv->sender_port;
1464 : 0 : strlcpy(conninfo, WalRcv->conninfo, sizeof(conninfo));
1465 : 0 : SpinLockRelease(&WalRcv->mutex);
1466 : :
1467 : : /*
1468 : : * No WAL receiver (or not ready yet), just return a tuple with NULL
1469 : : * values
1470 : : */
1471 [ # # # # ]: 0 : if (pid == 0 || !ready_to_display)
1472 : 0 : PG_RETURN_NULL();
1473 : :
1474 : : /*
1475 : : * Read "writtenUpto" without holding a spinlock. Note that it may not be
1476 : : * consistent with the other shared variables of the WAL receiver
1477 : : * protected by a spinlock, but this should not be used for data integrity
1478 : : * checks.
1479 : : */
1480 : 0 : written_lsn = pg_atomic_read_u64(&WalRcv->writtenUpto);
1481 : :
1482 : : /* determine result type */
1483 [ # # ]: 0 : if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1484 [ # # # # ]: 0 : elog(ERROR, "return type must be a row type");
1485 : :
1486 : 0 : values = palloc0_array(Datum, tupdesc->natts);
1487 : 0 : nulls = palloc0_array(bool, tupdesc->natts);
1488 : :
1489 : : /* Fetch values */
1490 : 0 : values[0] = Int32GetDatum(pid);
1491 : :
1492 [ # # ]: 0 : if (!has_privs_of_role(GetUserId(), ROLE_PG_READ_ALL_STATS))
1493 : : {
1494 : : /*
1495 : : * Only superusers and roles with privileges of pg_read_all_stats can
1496 : : * see details. Other users only get the pid value to know whether it
1497 : : * is a WAL receiver, but no details.
1498 : : */
1499 : 0 : memset(&nulls[1], true, sizeof(bool) * (tupdesc->natts - 1));
1500 : 0 : }
1501 : : else
1502 : : {
1503 : 0 : values[1] = CStringGetTextDatum(WalRcvGetStateString(state));
1504 : :
1505 [ # # ]: 0 : if (!XLogRecPtrIsValid(receive_start_lsn))
1506 : 0 : nulls[2] = true;
1507 : : else
1508 : 0 : values[2] = LSNGetDatum(receive_start_lsn);
1509 : 0 : values[3] = Int32GetDatum(receive_start_tli);
1510 [ # # ]: 0 : if (!XLogRecPtrIsValid(written_lsn))
1511 : 0 : nulls[4] = true;
1512 : : else
1513 : 0 : values[4] = LSNGetDatum(written_lsn);
1514 [ # # ]: 0 : if (!XLogRecPtrIsValid(flushed_lsn))
1515 : 0 : nulls[5] = true;
1516 : : else
1517 : 0 : values[5] = LSNGetDatum(flushed_lsn);
1518 : 0 : values[6] = Int32GetDatum(received_tli);
1519 [ # # ]: 0 : if (last_send_time == 0)
1520 : 0 : nulls[7] = true;
1521 : : else
1522 : 0 : values[7] = TimestampTzGetDatum(last_send_time);
1523 [ # # ]: 0 : if (last_receipt_time == 0)
1524 : 0 : nulls[8] = true;
1525 : : else
1526 : 0 : values[8] = TimestampTzGetDatum(last_receipt_time);
1527 [ # # ]: 0 : if (!XLogRecPtrIsValid(latest_end_lsn))
1528 : 0 : nulls[9] = true;
1529 : : else
1530 : 0 : values[9] = LSNGetDatum(latest_end_lsn);
1531 [ # # ]: 0 : if (latest_end_time == 0)
1532 : 0 : nulls[10] = true;
1533 : : else
1534 : 0 : values[10] = TimestampTzGetDatum(latest_end_time);
1535 [ # # ]: 0 : if (*slotname == '\0')
1536 : 0 : nulls[11] = true;
1537 : : else
1538 : 0 : values[11] = CStringGetTextDatum(slotname);
1539 [ # # ]: 0 : if (*sender_host == '\0')
1540 : 0 : nulls[12] = true;
1541 : : else
1542 : 0 : values[12] = CStringGetTextDatum(sender_host);
1543 [ # # ]: 0 : if (sender_port == 0)
1544 : 0 : nulls[13] = true;
1545 : : else
1546 : 0 : values[13] = Int32GetDatum(sender_port);
1547 [ # # ]: 0 : if (*conninfo == '\0')
1548 : 0 : nulls[14] = true;
1549 : : else
1550 : 0 : values[14] = CStringGetTextDatum(conninfo);
1551 : : }
1552 : :
1553 : : /* Returns the record as Datum */
1554 : 0 : PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
1555 : 0 : }
|
