1 /*-------------------------------------------------------------------------
5 * The WAL sender process (walsender) is new as of Postgres 9.0. It takes
6 * care of sending XLOG from the primary server to a single recipient.
7 * (Note that there can be more than one walsender process concurrently.)
8 * It is started by the postmaster when the walreceiver of a standby server
9 * connects to the primary server and requests XLOG streaming replication.
10 * It attempts to keep reading XLOG records from the disk and sending them
11 * to the standby server, as long as the connection is alive (i.e., like
12 * any backend, there is a one-to-one relationship between a connection
13 * and a walsender process).
15 * Normal termination is by SIGTERM, which instructs the walsender to
16 * close the connection and exit(0) at next convenient moment. Emergency
17 * termination is by SIGQUIT; like any backend, the walsender will simply
18 * abort and exit on SIGQUIT. A close of the connection and a FATAL error
19 * are treated as not a crash but approximately normal termination;
20 * the walsender will exit quickly without sending any more XLOG records.
22 * If the server is shut down, postmaster sends us SIGUSR2 after all
23 * regular backends have exited and the shutdown checkpoint has been written.
24 * This instruct walsender to send any outstanding WAL, including the
25 * shutdown checkpoint record, and then exit.
28 * Portions Copyright (c) 2010-2011, PostgreSQL Global Development Group
31 * src/backend/replication/walsender.c
33 *-------------------------------------------------------------------------
41 #include "access/xlog_internal.h"
42 #include "access/transam.h"
43 #include "catalog/pg_type.h"
44 #include "libpq/libpq.h"
45 #include "libpq/pqformat.h"
46 #include "libpq/pqsignal.h"
47 #include "miscadmin.h"
48 #include "replication/basebackup.h"
49 #include "replication/replnodes.h"
50 #include "replication/walprotocol.h"
51 #include "replication/walsender.h"
52 #include "storage/fd.h"
53 #include "storage/ipc.h"
54 #include "storage/pmsignal.h"
55 #include "storage/proc.h"
56 #include "storage/procarray.h"
57 #include "tcop/tcopprot.h"
58 #include "utils/builtins.h"
59 #include "utils/guc.h"
60 #include "utils/memutils.h"
61 #include "utils/ps_status.h"
62 #include "utils/resowner.h"
65 /* Array of WalSnds in shared memory */
66 WalSndCtlData *WalSndCtl = NULL;
68 /* My slot in the shared memory array */
69 static WalSnd *MyWalSnd = NULL;
72 bool am_walsender = false; /* Am I a walsender process ? */
74 /* User-settable parameters for walsender */
75 int max_wal_senders = 0; /* the maximum number of concurrent walsenders */
76 int WalSndDelay = 200; /* max sleep time between some actions */
79 * These variables are used similarly to openLogFile/Id/Seg/Off,
80 * but for walsender to read the XLOG.
82 static int sendFile = -1;
83 static uint32 sendId = 0;
84 static uint32 sendSeg = 0;
85 static uint32 sendOff = 0;
88 * How far have we sent WAL already? This is also advertised in
89 * MyWalSnd->sentPtr. (Actually, this is the next WAL location to send.)
91 static XLogRecPtr sentPtr = {0, 0};
94 * Buffer for processing reply messages.
96 static StringInfoData reply_message;
98 /* Flags set by signal handlers for later service in main loop */
99 static volatile sig_atomic_t got_SIGHUP = false;
100 volatile sig_atomic_t walsender_shutdown_requested = false;
101 volatile sig_atomic_t walsender_ready_to_stop = false;
103 /* Signal handlers */
104 static void WalSndSigHupHandler(SIGNAL_ARGS);
105 static void WalSndShutdownHandler(SIGNAL_ARGS);
106 static void WalSndQuickDieHandler(SIGNAL_ARGS);
107 static void WalSndXLogSendHandler(SIGNAL_ARGS);
108 static void WalSndLastCycleHandler(SIGNAL_ARGS);
110 /* Prototypes for private functions */
111 static bool HandleReplicationCommand(const char *cmd_string);
112 static int WalSndLoop(void);
113 static void InitWalSnd(void);
114 static void WalSndHandshake(void);
115 static void WalSndKill(int code, Datum arg);
116 static bool XLogSend(char *msgbuf, bool *caughtup);
117 static void IdentifySystem(void);
118 static void StartReplication(StartReplicationCmd * cmd);
119 static void ProcessStandbyMessage(void);
120 static void ProcessStandbyReplyMessage(void);
121 static void ProcessStandbyHSFeedbackMessage(void);
122 static void ProcessRepliesIfAny(void);
125 /* Main entry point for walsender process */
129 MemoryContext walsnd_context;
131 if (RecoveryInProgress())
133 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
134 errmsg("recovery is still in progress, can't accept WAL streaming connections")));
136 /* Create a per-walsender data structure in shared memory */
140 * Create a memory context that we will do all our work in. We do this so
141 * that we can reset the context during error recovery and thereby avoid
142 * possible memory leaks. Formerly this code just ran in
143 * TopMemoryContext, but resetting that would be a really bad idea.
145 * XXX: we don't actually attempt error recovery in walsender, we just
146 * close the connection and exit.
148 walsnd_context = AllocSetContextCreate(TopMemoryContext,
150 ALLOCSET_DEFAULT_MINSIZE,
151 ALLOCSET_DEFAULT_INITSIZE,
152 ALLOCSET_DEFAULT_MAXSIZE);
153 MemoryContextSwitchTo(walsnd_context);
155 /* Set up resource owner */
156 CurrentResourceOwner = ResourceOwnerCreate(NULL, "walsender top-level resource owner");
158 /* Unblock signals (they were blocked when the postmaster forked us) */
159 PG_SETMASK(&UnBlockSig);
161 /* Tell the standby that walsender is ready for receiving commands */
162 ReadyForQuery(DestRemote);
164 /* Handle handshake messages before streaming */
167 /* Initialize shared memory status */
169 /* use volatile pointer to prevent code rearrangement */
170 volatile WalSnd *walsnd = MyWalSnd;
172 SpinLockAcquire(&walsnd->mutex);
173 walsnd->sentPtr = sentPtr;
174 SpinLockRelease(&walsnd->mutex);
177 /* Main loop of walsender */
182 * Execute commands from walreceiver, until we enter streaming mode.
185 WalSndHandshake(void)
187 StringInfoData input_message;
188 bool replication_started = false;
190 initStringInfo(&input_message);
192 while (!replication_started)
196 WalSndSetState(WALSNDSTATE_STARTUP);
197 set_ps_display("idle", false);
199 /* Wait for a command to arrive */
200 firstchar = pq_getbyte();
203 * Emergency bailout if postmaster has died. This is to avoid the
204 * necessity for manual cleanup of all postmaster children.
206 if (!PostmasterIsAlive(true))
210 * Check for any other interesting events that happened while we
216 ProcessConfigFile(PGC_SIGHUP);
219 if (firstchar != EOF)
222 * Read the message contents. This is expected to be done without
223 * blocking because we've been able to get message type code.
225 if (pq_getmessage(&input_message, 0))
226 firstchar = EOF; /* suitable message already logged */
229 /* Handle the very limited subset of commands expected in this phase */
232 case 'Q': /* Query message */
234 const char *query_string;
236 query_string = pq_getmsgstring(&input_message);
237 pq_getmsgend(&input_message);
239 if (HandleReplicationCommand(query_string))
240 replication_started = true;
245 /* standby is closing the connection */
249 /* standby disconnected unexpectedly */
251 (errcode(ERRCODE_PROTOCOL_VIOLATION),
252 errmsg("unexpected EOF on standby connection")));
257 (errcode(ERRCODE_PROTOCOL_VIOLATION),
258 errmsg("invalid standby handshake message type %d", firstchar)));
272 char xpos[MAXFNAMELEN];
276 * Reply with a result set with one row, three columns. First col is system
277 * ID, second is timeline ID, and third is current xlog location.
280 snprintf(sysid, sizeof(sysid), UINT64_FORMAT,
281 GetSystemIdentifier());
282 snprintf(tli, sizeof(tli), "%u", ThisTimeLineID);
284 logptr = GetInsertRecPtr();
286 snprintf(xpos, sizeof(xpos), "%X/%X",
287 logptr.xlogid, logptr.xrecoff);
289 /* Send a RowDescription message */
290 pq_beginmessage(&buf, 'T');
291 pq_sendint(&buf, 3, 2); /* 3 fields */
294 pq_sendstring(&buf, "systemid"); /* col name */
295 pq_sendint(&buf, 0, 4); /* table oid */
296 pq_sendint(&buf, 0, 2); /* attnum */
297 pq_sendint(&buf, TEXTOID, 4); /* type oid */
298 pq_sendint(&buf, -1, 2); /* typlen */
299 pq_sendint(&buf, 0, 4); /* typmod */
300 pq_sendint(&buf, 0, 2); /* format code */
303 pq_sendstring(&buf, "timeline"); /* col name */
304 pq_sendint(&buf, 0, 4); /* table oid */
305 pq_sendint(&buf, 0, 2); /* attnum */
306 pq_sendint(&buf, INT4OID, 4); /* type oid */
307 pq_sendint(&buf, 4, 2); /* typlen */
308 pq_sendint(&buf, 0, 4); /* typmod */
309 pq_sendint(&buf, 0, 2); /* format code */
312 pq_sendstring(&buf, "xlogpos");
313 pq_sendint(&buf, 0, 4);
314 pq_sendint(&buf, 0, 2);
315 pq_sendint(&buf, TEXTOID, 4);
316 pq_sendint(&buf, -1, 2);
317 pq_sendint(&buf, 0, 4);
318 pq_sendint(&buf, 0, 2);
321 /* Send a DataRow message */
322 pq_beginmessage(&buf, 'D');
323 pq_sendint(&buf, 3, 2); /* # of columns */
324 pq_sendint(&buf, strlen(sysid), 4); /* col1 len */
325 pq_sendbytes(&buf, (char *) &sysid, strlen(sysid));
326 pq_sendint(&buf, strlen(tli), 4); /* col2 len */
327 pq_sendbytes(&buf, (char *) tli, strlen(tli));
328 pq_sendint(&buf, strlen(xpos), 4); /* col3 len */
329 pq_sendbytes(&buf, (char *) xpos, strlen(xpos));
333 /* Send CommandComplete and ReadyForQuery messages */
334 EndCommand("SELECT", DestRemote);
335 ReadyForQuery(DestRemote);
336 /* ReadyForQuery did pq_flush for us */
343 StartReplication(StartReplicationCmd * cmd)
348 * Let postmaster know that we're streaming. Once we've declared us as
349 * a WAL sender process, postmaster will let us outlive the bgwriter and
350 * kill us last in the shutdown sequence, so we get a chance to stream
351 * all remaining WAL at shutdown, including the shutdown checkpoint.
352 * Note that there's no going back, and we mustn't write any WAL records
355 MarkPostmasterChildWalSender();
358 * Check that we're logging enough information in the WAL for
361 * NOTE: This only checks the current value of wal_level. Even if the
362 * current setting is not 'minimal', there can be old WAL in the pg_xlog
363 * directory that was created with 'minimal'. So this is not bulletproof,
364 * the purpose is just to give a user-friendly error message that hints
365 * how to configure the system correctly.
367 if (wal_level == WAL_LEVEL_MINIMAL)
369 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
370 errmsg("standby connections not allowed because wal_level=minimal")));
372 /* Send a CopyBothResponse message, and start streaming */
373 pq_beginmessage(&buf, 'W');
374 pq_sendbyte(&buf, 0);
375 pq_sendint(&buf, 0, 2);
380 * Initialize position to the received one, then the xlog records begin to
381 * be shipped from that position
383 sentPtr = cmd->startpoint;
387 * Execute an incoming replication command.
390 HandleReplicationCommand(const char *cmd_string)
392 bool replication_started = false;
395 MemoryContext cmd_context;
396 MemoryContext old_context;
398 elog(DEBUG1, "received replication command: %s", cmd_string);
400 cmd_context = AllocSetContextCreate(CurrentMemoryContext,
401 "Replication command context",
402 ALLOCSET_DEFAULT_MINSIZE,
403 ALLOCSET_DEFAULT_INITSIZE,
404 ALLOCSET_DEFAULT_MAXSIZE);
405 old_context = MemoryContextSwitchTo(cmd_context);
407 replication_scanner_init(cmd_string);
408 parse_rc = replication_yyparse();
411 (errcode(ERRCODE_SYNTAX_ERROR),
412 (errmsg_internal("replication command parser returned %d",
415 cmd_node = replication_parse_result;
417 switch (cmd_node->type)
419 case T_IdentifySystemCmd:
423 case T_StartReplicationCmd:
424 StartReplication((StartReplicationCmd *) cmd_node);
426 /* break out of the loop */
427 replication_started = true;
430 case T_BaseBackupCmd:
431 SendBaseBackup((BaseBackupCmd *) cmd_node);
433 /* Send CommandComplete and ReadyForQuery messages */
434 EndCommand("SELECT", DestRemote);
435 ReadyForQuery(DestRemote);
436 /* ReadyForQuery did pq_flush for us */
441 (errcode(ERRCODE_PROTOCOL_VIOLATION),
442 errmsg("invalid standby query string: %s", cmd_string)));
446 MemoryContextSwitchTo(old_context);
447 MemoryContextDelete(cmd_context);
449 return replication_started;
453 * Check if the remote end has closed the connection.
456 ProcessRepliesIfAny(void)
458 unsigned char firstchar;
463 r = pq_getbyte_if_available(&firstchar);
466 /* unexpected error or EOF */
468 (errcode(ERRCODE_PROTOCOL_VIOLATION),
469 errmsg("unexpected EOF on standby connection")));
474 /* no data available without blocking */
478 /* Handle the very limited subset of commands expected in this phase */
482 * 'd' means a standby reply wrapped in a CopyData packet.
485 ProcessStandbyMessage();
489 * 'X' means that the standby is closing down the socket.
496 (errcode(ERRCODE_PROTOCOL_VIOLATION),
497 errmsg("invalid standby closing message type %d",
504 * Process a status update message received from standby.
507 ProcessStandbyMessage(void)
511 resetStringInfo(&reply_message);
514 * Read the message contents.
516 if (pq_getmessage(&reply_message, 0))
519 (errcode(ERRCODE_PROTOCOL_VIOLATION),
520 errmsg("unexpected EOF on standby connection")));
525 * Check message type from the first byte. At the moment, there is only
528 msgtype = pq_getmsgbyte(&reply_message);
533 ProcessStandbyReplyMessage();
537 ProcessStandbyHSFeedbackMessage();
542 (errcode(ERRCODE_PROTOCOL_VIOLATION),
543 errmsg("unexpected message type %c", msgtype)));
549 * Regular reply from standby advising of WAL positions on standby server.
552 ProcessStandbyReplyMessage(void)
554 StandbyReplyMessage reply;
556 pq_copymsgbytes(&reply_message, (char *) &reply, sizeof(StandbyReplyMessage));
558 elog(DEBUG2, "write %X/%X flush %X/%X apply %X/%X",
559 reply.write.xlogid, reply.write.xrecoff,
560 reply.flush.xlogid, reply.flush.xrecoff,
561 reply.apply.xlogid, reply.apply.xrecoff);
564 * Update shared state for this WalSender process
565 * based on reply data from standby.
568 /* use volatile pointer to prevent code rearrangement */
569 volatile WalSnd *walsnd = MyWalSnd;
571 SpinLockAcquire(&walsnd->mutex);
572 walsnd->write = reply.write;
573 walsnd->flush = reply.flush;
574 walsnd->apply = reply.apply;
575 SpinLockRelease(&walsnd->mutex);
580 * Hot Standby feedback
583 ProcessStandbyHSFeedbackMessage(void)
585 StandbyHSFeedbackMessage reply;
586 TransactionId newxmin = InvalidTransactionId;
588 pq_copymsgbytes(&reply_message, (char *) &reply, sizeof(StandbyHSFeedbackMessage));
590 elog(DEBUG2, "hot standby feedback xmin %u epoch %u",
595 * Update the WalSender's proc xmin to allow it to be visible
596 * to snapshots. This will hold back the removal of dead rows
597 * and thereby prevent the generation of cleanup conflicts
598 * on the standby server.
600 if (TransactionIdIsValid(reply.xmin))
602 TransactionId nextXid;
604 bool epochOK = false;
606 GetNextXidAndEpoch(&nextXid, &nextEpoch);
609 * Epoch of oldestXmin should be same as standby or
610 * if the counter has wrapped, then one less than reply.
612 if (reply.xmin <= nextXid)
614 if (reply.epoch == nextEpoch)
619 if (nextEpoch > 0 && reply.epoch == nextEpoch - 1)
624 * Feedback from standby must not go backwards, nor should it go
625 * forwards further than our most recent xid.
627 if (epochOK && TransactionIdPrecedesOrEquals(reply.xmin, nextXid))
629 if (!TransactionIdIsValid(MyProc->xmin))
631 TransactionId oldestXmin = GetOldestXmin(true, true);
632 if (TransactionIdPrecedes(oldestXmin, reply.xmin))
633 newxmin = reply.xmin;
635 newxmin = oldestXmin;
639 if (TransactionIdPrecedes(MyProc->xmin, reply.xmin))
640 newxmin = reply.xmin;
642 newxmin = MyProc->xmin; /* stay the same */
648 * Grab the ProcArrayLock to set xmin, or invalidate for bad reply
650 if (MyProc->xmin != newxmin)
652 LWLockAcquire(ProcArrayLock, LW_SHARED);
653 MyProc->xmin = newxmin;
654 LWLockRelease(ProcArrayLock);
658 /* Main loop of walsender process */
662 char *output_message;
663 bool caughtup = false;
666 * Allocate buffer that will be used for each output message. We do this
667 * just once to reduce palloc overhead. The buffer must be made large
668 * enough for maximum-sized messages.
670 output_message = palloc(1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE);
673 * Allocate buffer that will be used for processing reply messages. As
674 * above, do this just once to reduce palloc overhead.
676 initStringInfo(&reply_message);
678 /* Loop forever, unless we get an error */
682 * Emergency bailout if postmaster has died. This is to avoid the
683 * necessity for manual cleanup of all postmaster children.
685 if (!PostmasterIsAlive(true))
688 /* Process any requests or signals received recently */
692 ProcessConfigFile(PGC_SIGHUP);
696 * When SIGUSR2 arrives, we send all outstanding logs up to the
697 * shutdown checkpoint record (i.e., the latest record) and exit.
699 if (walsender_ready_to_stop)
701 if (!XLogSend(output_message, &caughtup))
703 ProcessRepliesIfAny();
705 walsender_shutdown_requested = true;
708 /* Normal exit from the walsender is here */
709 if (walsender_shutdown_requested)
711 /* Inform the standby that XLOG streaming was done */
712 pq_puttextmessage('C', "COPY 0");
719 * If we had sent all accumulated WAL in last round, nap for the
720 * configured time before retrying.
725 * Even if we wrote all the WAL that was available when we started
726 * sending, more might have arrived while we were sending this
727 * batch. We had the latch set while sending, so we have not
728 * received any signals from that time. Let's arm the latch
729 * again, and after that check that we're still up-to-date.
731 ResetLatch(&MyWalSnd->latch);
733 if (!XLogSend(output_message, &caughtup))
735 if (caughtup && !got_SIGHUP && !walsender_ready_to_stop && !walsender_shutdown_requested)
738 * XXX: We don't really need the periodic wakeups anymore,
739 * WaitLatchOrSocket should reliably wake up as soon as
740 * something interesting happens.
744 WaitLatchOrSocket(&MyWalSnd->latch, MyProcPort->sock,
745 WalSndDelay * 1000L);
750 /* Attempt to send the log once every loop */
751 if (!XLogSend(output_message, &caughtup))
755 /* Update our state to indicate if we're behind or not */
756 WalSndSetState(caughtup ? WALSNDSTATE_STREAMING : WALSNDSTATE_CATCHUP);
757 ProcessRepliesIfAny();
761 * Get here on send failure. Clean up and exit.
763 * Reset whereToSendOutput to prevent ereport from attempting to send any
764 * more messages to the standby.
766 if (whereToSendOutput == DestRemote)
767 whereToSendOutput = DestNone;
770 return 1; /* keep the compiler quiet */
773 /* Initialize a per-walsender data structure for this walsender process */
780 * WalSndCtl should be set up already (we inherit this by fork() or
781 * EXEC_BACKEND mechanism from the postmaster).
783 Assert(WalSndCtl != NULL);
784 Assert(MyWalSnd == NULL);
787 * Find a free walsender slot and reserve it. If this fails, we must be
788 * out of WalSnd structures.
790 for (i = 0; i < max_wal_senders; i++)
792 /* use volatile pointer to prevent code rearrangement */
793 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
795 SpinLockAcquire(&walsnd->mutex);
797 if (walsnd->pid != 0)
799 SpinLockRelease(&walsnd->mutex);
805 * Found a free slot. Reserve it for us.
807 walsnd->pid = MyProcPid;
808 MemSet(&walsnd->sentPtr, 0, sizeof(XLogRecPtr));
809 walsnd->state = WALSNDSTATE_STARTUP;
810 SpinLockRelease(&walsnd->mutex);
811 /* don't need the lock anymore */
812 OwnLatch((Latch *) &walsnd->latch);
813 MyWalSnd = (WalSnd *) walsnd;
818 if (MyWalSnd == NULL)
820 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
821 errmsg("number of requested standby connections "
822 "exceeds max_wal_senders (currently %d)",
825 /* Arrange to clean up at walsender exit */
826 on_shmem_exit(WalSndKill, 0);
829 /* Destroy the per-walsender data structure for this walsender process */
831 WalSndKill(int code, Datum arg)
833 Assert(MyWalSnd != NULL);
836 * Mark WalSnd struct no longer in use. Assume that no lock is required
840 DisownLatch(&MyWalSnd->latch);
842 /* WalSnd struct isn't mine anymore */
847 * Read 'nbytes' bytes from WAL into 'buf', starting at location 'recptr'
849 * XXX probably this should be improved to suck data directly from the
850 * WAL buffers when possible.
852 * Will open, and keep open, one WAL segment stored in the global file
853 * descriptor sendFile. This means if XLogRead is used once, there will
854 * always be one descriptor left open until the process ends, but never
858 XLogRead(char *buf, XLogRecPtr recptr, Size nbytes)
860 XLogRecPtr startRecPtr = recptr;
861 char path[MAXPGPATH];
862 uint32 lastRemovedLog;
863 uint32 lastRemovedSeg;
873 startoff = recptr.xrecoff % XLogSegSize;
875 if (sendFile < 0 || !XLByteInSeg(recptr, sendId, sendSeg))
877 /* Switch to another logfile segment */
881 XLByteToSeg(recptr, sendId, sendSeg);
882 XLogFilePath(path, ThisTimeLineID, sendId, sendSeg);
884 sendFile = BasicOpenFile(path, O_RDONLY | PG_BINARY, 0);
888 * If the file is not found, assume it's because the standby
889 * asked for a too old WAL segment that has already been
890 * removed or recycled.
894 char filename[MAXFNAMELEN];
896 XLogFileName(filename, ThisTimeLineID, sendId, sendSeg);
898 (errcode_for_file_access(),
899 errmsg("requested WAL segment %s has already been removed",
904 (errcode_for_file_access(),
905 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
906 path, sendId, sendSeg)));
911 /* Need to seek in the file? */
912 if (sendOff != startoff)
914 if (lseek(sendFile, (off_t) startoff, SEEK_SET) < 0)
916 (errcode_for_file_access(),
917 errmsg("could not seek in log file %u, segment %u to offset %u: %m",
918 sendId, sendSeg, startoff)));
922 /* How many bytes are within this segment? */
923 if (nbytes > (XLogSegSize - startoff))
924 segbytes = XLogSegSize - startoff;
928 readbytes = read(sendFile, buf, segbytes);
931 (errcode_for_file_access(),
932 errmsg("could not read from log file %u, segment %u, offset %u, "
934 sendId, sendSeg, sendOff, (unsigned long) segbytes)));
936 /* Update state for read */
937 XLByteAdvance(recptr, readbytes);
939 sendOff += readbytes;
945 * After reading into the buffer, check that what we read was valid. We do
946 * this after reading, because even though the segment was present when we
947 * opened it, it might get recycled or removed while we read it. The
948 * read() succeeds in that case, but the data we tried to read might
949 * already have been overwritten with new WAL records.
951 XLogGetLastRemoved(&lastRemovedLog, &lastRemovedSeg);
952 XLByteToSeg(startRecPtr, log, seg);
953 if (log < lastRemovedLog ||
954 (log == lastRemovedLog && seg <= lastRemovedSeg))
956 char filename[MAXFNAMELEN];
958 XLogFileName(filename, ThisTimeLineID, log, seg);
960 (errcode_for_file_access(),
961 errmsg("requested WAL segment %s has already been removed",
967 * Read up to MAX_SEND_SIZE bytes of WAL that's been flushed to disk,
968 * but not yet sent to the client, and send it.
970 * msgbuf is a work area in which the output message is constructed. It's
971 * passed in just so we can avoid re-palloc'ing the buffer on each cycle.
972 * It must be of size 1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE.
974 * If there is no unsent WAL remaining, *caughtup is set to true, otherwise
975 * *caughtup is set to false.
977 * Returns true if OK, false if trouble.
980 XLogSend(char *msgbuf, bool *caughtup)
982 XLogRecPtr SendRqstPtr;
986 WalDataMessageHeader msghdr;
989 * Attempt to send all data that's already been written out and fsync'd to
990 * disk. We cannot go further than what's been written out given the
991 * current implementation of XLogRead(). And in any case it's unsafe to
992 * send WAL that is not securely down to disk on the master: if the master
993 * subsequently crashes and restarts, slaves must not have applied any WAL
994 * that gets lost on the master.
996 SendRqstPtr = GetFlushRecPtr();
998 /* Quick exit if nothing to do */
999 if (XLByteLE(SendRqstPtr, sentPtr))
1006 * Figure out how much to send in one message. If there's no more than
1007 * MAX_SEND_SIZE bytes to send, send everything. Otherwise send
1008 * MAX_SEND_SIZE bytes, but round back to logfile or page boundary.
1010 * The rounding is not only for performance reasons. Walreceiver relies on
1011 * the fact that we never split a WAL record across two messages. Since a
1012 * long WAL record is split at page boundary into continuation records,
1013 * page boundary is always a safe cut-off point. We also assume that
1014 * SendRqstPtr never points to the middle of a WAL record.
1017 if (startptr.xrecoff >= XLogFileSize)
1020 * crossing a logid boundary, skip the non-existent last log segment
1021 * in previous logical log file.
1023 startptr.xlogid += 1;
1024 startptr.xrecoff = 0;
1028 XLByteAdvance(endptr, MAX_SEND_SIZE);
1029 if (endptr.xlogid != startptr.xlogid)
1031 /* Don't cross a logfile boundary within one message */
1032 Assert(endptr.xlogid == startptr.xlogid + 1);
1033 endptr.xlogid = startptr.xlogid;
1034 endptr.xrecoff = XLogFileSize;
1037 /* if we went beyond SendRqstPtr, back off */
1038 if (XLByteLE(SendRqstPtr, endptr))
1040 endptr = SendRqstPtr;
1045 /* round down to page boundary. */
1046 endptr.xrecoff -= (endptr.xrecoff % XLOG_BLCKSZ);
1050 nbytes = endptr.xrecoff - startptr.xrecoff;
1051 Assert(nbytes <= MAX_SEND_SIZE);
1054 * OK to read and send the slice.
1059 * Read the log directly into the output buffer to avoid extra memcpy
1062 XLogRead(msgbuf + 1 + sizeof(WalDataMessageHeader), startptr, nbytes);
1065 * We fill the message header last so that the send timestamp is taken as
1068 msghdr.dataStart = startptr;
1069 msghdr.walEnd = SendRqstPtr;
1070 msghdr.sendTime = GetCurrentTimestamp();
1072 memcpy(msgbuf + 1, &msghdr, sizeof(WalDataMessageHeader));
1074 pq_putmessage('d', msgbuf, 1 + sizeof(WalDataMessageHeader) + nbytes);
1076 /* Flush pending output to the client */
1082 /* Update shared memory status */
1084 /* use volatile pointer to prevent code rearrangement */
1085 volatile WalSnd *walsnd = MyWalSnd;
1087 SpinLockAcquire(&walsnd->mutex);
1088 walsnd->sentPtr = sentPtr;
1089 SpinLockRelease(&walsnd->mutex);
1092 /* Report progress of XLOG streaming in PS display */
1093 if (update_process_title)
1095 char activitymsg[50];
1097 snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%X",
1098 sentPtr.xlogid, sentPtr.xrecoff);
1099 set_ps_display(activitymsg, false);
1105 /* SIGHUP: set flag to re-read config file at next convenient time */
1107 WalSndSigHupHandler(SIGNAL_ARGS)
1111 SetLatch(&MyWalSnd->latch);
1114 /* SIGTERM: set flag to shut down */
1116 WalSndShutdownHandler(SIGNAL_ARGS)
1118 walsender_shutdown_requested = true;
1120 SetLatch(&MyWalSnd->latch);
1124 * WalSndQuickDieHandler() occurs when signalled SIGQUIT by the postmaster.
1126 * Some backend has bought the farm,
1127 * so we need to stop what we're doing and exit.
1130 WalSndQuickDieHandler(SIGNAL_ARGS)
1132 PG_SETMASK(&BlockSig);
1135 * We DO NOT want to run proc_exit() callbacks -- we're here because
1136 * shared memory may be corrupted, so we don't want to try to clean up our
1137 * transaction. Just nail the windows shut and get out of town. Now that
1138 * there's an atexit callback to prevent third-party code from breaking
1139 * things by calling exit() directly, we have to reset the callbacks
1140 * explicitly to make this work as intended.
1145 * Note we do exit(2) not exit(0). This is to force the postmaster into a
1146 * system reset cycle if some idiot DBA sends a manual SIGQUIT to a random
1147 * backend. This is necessary precisely because we don't clean up our
1148 * shared memory state. (The "dead man switch" mechanism in pmsignal.c
1149 * should ensure the postmaster sees this as a crash, too, but no harm in
1150 * being doubly sure.)
1155 /* SIGUSR1: set flag to send WAL records */
1157 WalSndXLogSendHandler(SIGNAL_ARGS)
1159 latch_sigusr1_handler();
1162 /* SIGUSR2: set flag to do a last cycle and shut down afterwards */
1164 WalSndLastCycleHandler(SIGNAL_ARGS)
1166 walsender_ready_to_stop = true;
1168 SetLatch(&MyWalSnd->latch);
1171 /* Set up signal handlers */
1175 /* Set up signal handlers */
1176 pqsignal(SIGHUP, WalSndSigHupHandler); /* set flag to read config
1178 pqsignal(SIGINT, SIG_IGN); /* not used */
1179 pqsignal(SIGTERM, WalSndShutdownHandler); /* request shutdown */
1180 pqsignal(SIGQUIT, WalSndQuickDieHandler); /* hard crash time */
1181 pqsignal(SIGALRM, SIG_IGN);
1182 pqsignal(SIGPIPE, SIG_IGN);
1183 pqsignal(SIGUSR1, WalSndXLogSendHandler); /* request WAL sending */
1184 pqsignal(SIGUSR2, WalSndLastCycleHandler); /* request a last cycle and
1187 /* Reset some signals that are accepted by postmaster but not here */
1188 pqsignal(SIGCHLD, SIG_DFL);
1189 pqsignal(SIGTTIN, SIG_DFL);
1190 pqsignal(SIGTTOU, SIG_DFL);
1191 pqsignal(SIGCONT, SIG_DFL);
1192 pqsignal(SIGWINCH, SIG_DFL);
1195 /* Report shared-memory space needed by WalSndShmemInit */
1197 WalSndShmemSize(void)
1201 size = offsetof(WalSndCtlData, walsnds);
1202 size = add_size(size, mul_size(max_wal_senders, sizeof(WalSnd)));
1207 /* Allocate and initialize walsender-related shared memory */
1209 WalSndShmemInit(void)
1214 WalSndCtl = (WalSndCtlData *)
1215 ShmemInitStruct("Wal Sender Ctl", WalSndShmemSize(), &found);
1219 /* First time through, so initialize */
1220 MemSet(WalSndCtl, 0, WalSndShmemSize());
1222 for (i = 0; i < max_wal_senders; i++)
1224 WalSnd *walsnd = &WalSndCtl->walsnds[i];
1226 SpinLockInit(&walsnd->mutex);
1227 InitSharedLatch(&walsnd->latch);
1232 /* Wake up all walsenders */
1238 for (i = 0; i < max_wal_senders; i++)
1239 SetLatch(&WalSndCtl->walsnds[i].latch);
1242 /* Set state for current walsender (only called in walsender) */
1244 WalSndSetState(WalSndState state)
1246 /* use volatile pointer to prevent code rearrangement */
1247 volatile WalSnd *walsnd = MyWalSnd;
1249 Assert(am_walsender);
1251 if (walsnd->state == state)
1254 SpinLockAcquire(&walsnd->mutex);
1255 walsnd->state = state;
1256 SpinLockRelease(&walsnd->mutex);
1260 * Return a string constant representing the state. This is used
1261 * in system views, and should *not* be translated.
1264 WalSndGetStateString(WalSndState state)
1268 case WALSNDSTATE_STARTUP:
1270 case WALSNDSTATE_BACKUP:
1272 case WALSNDSTATE_CATCHUP:
1274 case WALSNDSTATE_STREAMING:
1282 * Returns activity of walsenders, including pids and xlog locations sent to
1286 pg_stat_get_wal_senders(PG_FUNCTION_ARGS)
1288 #define PG_STAT_GET_WAL_SENDERS_COLS 6
1289 ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1291 Tuplestorestate *tupstore;
1292 MemoryContext per_query_ctx;
1293 MemoryContext oldcontext;
1296 /* check to see if caller supports us returning a tuplestore */
1297 if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
1299 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1300 errmsg("set-valued function called in context that cannot accept a set")));
1301 if (!(rsinfo->allowedModes & SFRM_Materialize))
1303 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1304 errmsg("materialize mode required, but it is not " \
1305 "allowed in this context")));
1307 /* Build a tuple descriptor for our result type */
1308 if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1309 elog(ERROR, "return type must be a row type");
1311 per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
1312 oldcontext = MemoryContextSwitchTo(per_query_ctx);
1314 tupstore = tuplestore_begin_heap(true, false, work_mem);
1315 rsinfo->returnMode = SFRM_Materialize;
1316 rsinfo->setResult = tupstore;
1317 rsinfo->setDesc = tupdesc;
1319 MemoryContextSwitchTo(oldcontext);
1321 for (i = 0; i < max_wal_senders; i++)
1323 /* use volatile pointer to prevent code rearrangement */
1324 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1325 char location[MAXFNAMELEN];
1331 Datum values[PG_STAT_GET_WAL_SENDERS_COLS];
1332 bool nulls[PG_STAT_GET_WAL_SENDERS_COLS];
1334 if (walsnd->pid == 0)
1337 SpinLockAcquire(&walsnd->mutex);
1338 sentPtr = walsnd->sentPtr;
1339 state = walsnd->state;
1340 write = walsnd->write;
1341 flush = walsnd->flush;
1342 apply = walsnd->apply;
1343 SpinLockRelease(&walsnd->mutex);
1345 memset(nulls, 0, sizeof(nulls));
1346 values[0] = Int32GetDatum(walsnd->pid);
1351 * Only superusers can see details. Other users only get
1352 * the pid value to know it's a walsender, but no details.
1362 values[1] = CStringGetTextDatum(WalSndGetStateString(state));
1364 snprintf(location, sizeof(location), "%X/%X",
1365 sentPtr.xlogid, sentPtr.xrecoff);
1366 values[2] = CStringGetTextDatum(location);
1368 if (write.xlogid == 0 && write.xrecoff == 0)
1370 snprintf(location, sizeof(location), "%X/%X",
1371 write.xlogid, write.xrecoff);
1372 values[3] = CStringGetTextDatum(location);
1374 if (flush.xlogid == 0 && flush.xrecoff == 0)
1376 snprintf(location, sizeof(location), "%X/%X",
1377 flush.xlogid, flush.xrecoff);
1378 values[4] = CStringGetTextDatum(location);
1380 if (apply.xlogid == 0 && apply.xrecoff == 0)
1382 snprintf(location, sizeof(location), "%X/%X",
1383 apply.xlogid, apply.xrecoff);
1384 values[5] = CStringGetTextDatum(location);
1387 tuplestore_putvalues(tupstore, tupdesc, values, nulls);
1390 /* clean up and return the tuplestore */
1391 tuplestore_donestoring(tupstore);
1397 * This isn't currently used for anything. Monitoring tools might be
1398 * interested in the future, and we'll need something like this in the
1399 * future for synchronous replication.
1403 * Returns the oldest Send position among walsenders. Or InvalidXLogRecPtr
1407 GetOldestWALSendPointer(void)
1409 XLogRecPtr oldest = {0, 0};
1413 for (i = 0; i < max_wal_senders; i++)
1415 /* use volatile pointer to prevent code rearrangement */
1416 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1419 if (walsnd->pid == 0)
1422 SpinLockAcquire(&walsnd->mutex);
1423 recptr = walsnd->sentPtr;
1424 SpinLockRelease(&walsnd->mutex);
1426 if (recptr.xlogid == 0 && recptr.xrecoff == 0)
1429 if (!found || XLByteLT(recptr, oldest))