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 "tcop/tcopprot.h"
57 #include "utils/builtins.h"
58 #include "utils/guc.h"
59 #include "utils/memutils.h"
60 #include "utils/ps_status.h"
61 #include "utils/resowner.h"
64 /* Array of WalSnds in shared memory */
65 WalSndCtlData *WalSndCtl = NULL;
67 /* My slot in the shared memory array */
68 static WalSnd *MyWalSnd = NULL;
71 bool am_walsender = false; /* Am I a walsender process ? */
73 /* User-settable parameters for walsender */
74 int max_wal_senders = 0; /* the maximum number of concurrent walsenders */
75 int WalSndDelay = 200; /* max sleep time between some actions */
78 * These variables are used similarly to openLogFile/Id/Seg/Off,
79 * but for walsender to read the XLOG.
81 static int sendFile = -1;
82 static uint32 sendId = 0;
83 static uint32 sendSeg = 0;
84 static uint32 sendOff = 0;
87 * How far have we sent WAL already? This is also advertised in
88 * MyWalSnd->sentPtr. (Actually, this is the next WAL location to send.)
90 static XLogRecPtr sentPtr = {0, 0};
93 * Buffer for processing reply messages.
95 static StringInfoData reply_message;
97 /* Flags set by signal handlers for later service in main loop */
98 static volatile sig_atomic_t got_SIGHUP = false;
99 volatile sig_atomic_t walsender_shutdown_requested = false;
100 volatile sig_atomic_t walsender_ready_to_stop = false;
102 /* Signal handlers */
103 static void WalSndSigHupHandler(SIGNAL_ARGS);
104 static void WalSndShutdownHandler(SIGNAL_ARGS);
105 static void WalSndQuickDieHandler(SIGNAL_ARGS);
106 static void WalSndXLogSendHandler(SIGNAL_ARGS);
107 static void WalSndLastCycleHandler(SIGNAL_ARGS);
109 /* Prototypes for private functions */
110 static bool HandleReplicationCommand(const char *cmd_string);
111 static int WalSndLoop(void);
112 static void InitWalSnd(void);
113 static void WalSndHandshake(void);
114 static void WalSndKill(int code, Datum arg);
115 static bool XLogSend(char *msgbuf, bool *caughtup);
116 static void IdentifySystem(void);
117 static void StartReplication(StartReplicationCmd * cmd);
118 static void ProcessStandbyReplyMessage(void);
119 static void ProcessRepliesIfAny(void);
122 /* Main entry point for walsender process */
126 MemoryContext walsnd_context;
128 if (RecoveryInProgress())
130 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
131 errmsg("recovery is still in progress, can't accept WAL streaming connections")));
133 /* Create a per-walsender data structure in shared memory */
137 * Create a memory context that we will do all our work in. We do this so
138 * that we can reset the context during error recovery and thereby avoid
139 * possible memory leaks. Formerly this code just ran in
140 * TopMemoryContext, but resetting that would be a really bad idea.
142 * XXX: we don't actually attempt error recovery in walsender, we just
143 * close the connection and exit.
145 walsnd_context = AllocSetContextCreate(TopMemoryContext,
147 ALLOCSET_DEFAULT_MINSIZE,
148 ALLOCSET_DEFAULT_INITSIZE,
149 ALLOCSET_DEFAULT_MAXSIZE);
150 MemoryContextSwitchTo(walsnd_context);
152 /* Set up resource owner */
153 CurrentResourceOwner = ResourceOwnerCreate(NULL, "walsender top-level resource owner");
155 /* Unblock signals (they were blocked when the postmaster forked us) */
156 PG_SETMASK(&UnBlockSig);
158 /* Tell the standby that walsender is ready for receiving commands */
159 ReadyForQuery(DestRemote);
161 /* Handle handshake messages before streaming */
164 /* Initialize shared memory status */
166 /* use volatile pointer to prevent code rearrangement */
167 volatile WalSnd *walsnd = MyWalSnd;
169 SpinLockAcquire(&walsnd->mutex);
170 walsnd->sentPtr = sentPtr;
171 SpinLockRelease(&walsnd->mutex);
174 /* Main loop of walsender */
179 * Execute commands from walreceiver, until we enter streaming mode.
182 WalSndHandshake(void)
184 StringInfoData input_message;
185 bool replication_started = false;
187 initStringInfo(&input_message);
189 while (!replication_started)
193 WalSndSetState(WALSNDSTATE_STARTUP);
194 set_ps_display("idle", false);
196 /* Wait for a command to arrive */
197 firstchar = pq_getbyte();
200 * Emergency bailout if postmaster has died. This is to avoid the
201 * necessity for manual cleanup of all postmaster children.
203 if (!PostmasterIsAlive(true))
207 * Check for any other interesting events that happened while we
213 ProcessConfigFile(PGC_SIGHUP);
216 if (firstchar != EOF)
219 * Read the message contents. This is expected to be done without
220 * blocking because we've been able to get message type code.
222 if (pq_getmessage(&input_message, 0))
223 firstchar = EOF; /* suitable message already logged */
226 /* Handle the very limited subset of commands expected in this phase */
229 case 'Q': /* Query message */
231 const char *query_string;
233 query_string = pq_getmsgstring(&input_message);
234 pq_getmsgend(&input_message);
236 if (HandleReplicationCommand(query_string))
237 replication_started = true;
242 /* standby is closing the connection */
246 /* standby disconnected unexpectedly */
248 (errcode(ERRCODE_PROTOCOL_VIOLATION),
249 errmsg("unexpected EOF on standby connection")));
254 (errcode(ERRCODE_PROTOCOL_VIOLATION),
255 errmsg("invalid standby handshake message type %d", firstchar)));
269 char xpos[MAXFNAMELEN];
273 * Reply with a result set with one row, three columns. First col is system
274 * ID, second is timeline ID, and third is current xlog location.
277 snprintf(sysid, sizeof(sysid), UINT64_FORMAT,
278 GetSystemIdentifier());
279 snprintf(tli, sizeof(tli), "%u", ThisTimeLineID);
281 logptr = GetInsertRecPtr();
283 snprintf(xpos, sizeof(xpos), "%X/%X",
284 logptr.xlogid, logptr.xrecoff);
286 /* Send a RowDescription message */
287 pq_beginmessage(&buf, 'T');
288 pq_sendint(&buf, 3, 2); /* 3 fields */
291 pq_sendstring(&buf, "systemid"); /* col name */
292 pq_sendint(&buf, 0, 4); /* table oid */
293 pq_sendint(&buf, 0, 2); /* attnum */
294 pq_sendint(&buf, TEXTOID, 4); /* type oid */
295 pq_sendint(&buf, -1, 2); /* typlen */
296 pq_sendint(&buf, 0, 4); /* typmod */
297 pq_sendint(&buf, 0, 2); /* format code */
300 pq_sendstring(&buf, "timeline"); /* col name */
301 pq_sendint(&buf, 0, 4); /* table oid */
302 pq_sendint(&buf, 0, 2); /* attnum */
303 pq_sendint(&buf, INT4OID, 4); /* type oid */
304 pq_sendint(&buf, 4, 2); /* typlen */
305 pq_sendint(&buf, 0, 4); /* typmod */
306 pq_sendint(&buf, 0, 2); /* format code */
309 pq_sendstring(&buf, "xlogpos");
310 pq_sendint(&buf, 0, 4);
311 pq_sendint(&buf, 0, 2);
312 pq_sendint(&buf, TEXTOID, 4);
313 pq_sendint(&buf, -1, 2);
314 pq_sendint(&buf, 0, 4);
315 pq_sendint(&buf, 0, 2);
318 /* Send a DataRow message */
319 pq_beginmessage(&buf, 'D');
320 pq_sendint(&buf, 3, 2); /* # of columns */
321 pq_sendint(&buf, strlen(sysid), 4); /* col1 len */
322 pq_sendbytes(&buf, (char *) &sysid, strlen(sysid));
323 pq_sendint(&buf, strlen(tli), 4); /* col2 len */
324 pq_sendbytes(&buf, (char *) tli, strlen(tli));
325 pq_sendint(&buf, strlen(xpos), 4); /* col3 len */
326 pq_sendbytes(&buf, (char *) xpos, strlen(xpos));
330 /* Send CommandComplete and ReadyForQuery messages */
331 EndCommand("SELECT", DestRemote);
332 ReadyForQuery(DestRemote);
333 /* ReadyForQuery did pq_flush for us */
340 StartReplication(StartReplicationCmd * cmd)
345 * Let postmaster know that we're streaming. Once we've declared us as
346 * a WAL sender process, postmaster will let us outlive the bgwriter and
347 * kill us last in the shutdown sequence, so we get a chance to stream
348 * all remaining WAL at shutdown, including the shutdown checkpoint.
349 * Note that there's no going back, and we mustn't write any WAL records
352 MarkPostmasterChildWalSender();
355 * Check that we're logging enough information in the WAL for
358 * NOTE: This only checks the current value of wal_level. Even if the
359 * current setting is not 'minimal', there can be old WAL in the pg_xlog
360 * directory that was created with 'minimal'. So this is not bulletproof,
361 * the purpose is just to give a user-friendly error message that hints
362 * how to configure the system correctly.
364 if (wal_level == WAL_LEVEL_MINIMAL)
366 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
367 errmsg("standby connections not allowed because wal_level=minimal")));
369 /* Send a CopyBothResponse message, and start streaming */
370 pq_beginmessage(&buf, 'W');
371 pq_sendbyte(&buf, 0);
372 pq_sendint(&buf, 0, 2);
377 * Initialize position to the received one, then the xlog records begin to
378 * be shipped from that position
380 sentPtr = cmd->startpoint;
384 * Execute an incoming replication command.
387 HandleReplicationCommand(const char *cmd_string)
389 bool replication_started = false;
392 MemoryContext cmd_context;
393 MemoryContext old_context;
395 elog(DEBUG1, "received replication command: %s", cmd_string);
397 cmd_context = AllocSetContextCreate(CurrentMemoryContext,
398 "Replication command context",
399 ALLOCSET_DEFAULT_MINSIZE,
400 ALLOCSET_DEFAULT_INITSIZE,
401 ALLOCSET_DEFAULT_MAXSIZE);
402 old_context = MemoryContextSwitchTo(cmd_context);
404 replication_scanner_init(cmd_string);
405 parse_rc = replication_yyparse();
408 (errcode(ERRCODE_SYNTAX_ERROR),
409 (errmsg_internal("replication command parser returned %d",
412 cmd_node = replication_parse_result;
414 switch (cmd_node->type)
416 case T_IdentifySystemCmd:
420 case T_StartReplicationCmd:
421 StartReplication((StartReplicationCmd *) cmd_node);
423 /* break out of the loop */
424 replication_started = true;
427 case T_BaseBackupCmd:
428 SendBaseBackup((BaseBackupCmd *) cmd_node);
430 /* Send CommandComplete and ReadyForQuery messages */
431 EndCommand("SELECT", DestRemote);
432 ReadyForQuery(DestRemote);
433 /* ReadyForQuery did pq_flush for us */
438 (errcode(ERRCODE_PROTOCOL_VIOLATION),
439 errmsg("invalid standby query string: %s", cmd_string)));
443 MemoryContextSwitchTo(old_context);
444 MemoryContextDelete(cmd_context);
446 return replication_started;
450 * Check if the remote end has closed the connection.
453 ProcessRepliesIfAny(void)
455 unsigned char firstchar;
458 r = pq_getbyte_if_available(&firstchar);
461 /* unexpected error or EOF */
463 (errcode(ERRCODE_PROTOCOL_VIOLATION),
464 errmsg("unexpected EOF on standby connection")));
469 /* no data available without blocking */
473 /* Handle the very limited subset of commands expected in this phase */
477 * 'd' means a standby reply wrapped in a CopyData packet.
480 ProcessStandbyReplyMessage();
484 * 'X' means that the standby is closing down the socket.
491 (errcode(ERRCODE_PROTOCOL_VIOLATION),
492 errmsg("invalid standby closing message type %d",
498 * Process a status update message received from standby.
501 ProcessStandbyReplyMessage(void)
503 StandbyReplyMessage reply;
506 resetStringInfo(&reply_message);
509 * Read the message contents.
511 if (pq_getmessage(&reply_message, 0))
514 (errcode(ERRCODE_PROTOCOL_VIOLATION),
515 errmsg("unexpected EOF on standby connection")));
520 * Check message type from the first byte. At the moment, there is only
523 msgtype = pq_getmsgbyte(&reply_message);
527 (errcode(ERRCODE_PROTOCOL_VIOLATION),
528 errmsg("unexpected message type %c", msgtype)));
532 pq_copymsgbytes(&reply_message, (char *) &reply, sizeof(StandbyReplyMessage));
534 elog(DEBUG2, "write %X/%X flush %X/%X apply %X/%X ",
535 reply.write.xlogid, reply.write.xrecoff,
536 reply.flush.xlogid, reply.flush.xrecoff,
537 reply.apply.xlogid, reply.apply.xrecoff);
540 * Update shared state for this WalSender process
541 * based on reply data from standby.
544 /* use volatile pointer to prevent code rearrangement */
545 volatile WalSnd *walsnd = MyWalSnd;
547 SpinLockAcquire(&walsnd->mutex);
548 walsnd->write = reply.write;
549 walsnd->flush = reply.flush;
550 walsnd->apply = reply.apply;
551 SpinLockRelease(&walsnd->mutex);
555 /* Main loop of walsender process */
559 char *output_message;
560 bool caughtup = false;
563 * Allocate buffer that will be used for each output message. We do this
564 * just once to reduce palloc overhead. The buffer must be made large
565 * enough for maximum-sized messages.
567 output_message = palloc(1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE);
570 * Allocate buffer that will be used for processing reply messages. As
571 * above, do this just once to reduce palloc overhead.
573 initStringInfo(&reply_message);
575 /* Loop forever, unless we get an error */
579 * Emergency bailout if postmaster has died. This is to avoid the
580 * necessity for manual cleanup of all postmaster children.
582 if (!PostmasterIsAlive(true))
585 /* Process any requests or signals received recently */
589 ProcessConfigFile(PGC_SIGHUP);
593 * When SIGUSR2 arrives, we send all outstanding logs up to the
594 * shutdown checkpoint record (i.e., the latest record) and exit.
596 if (walsender_ready_to_stop)
598 if (!XLogSend(output_message, &caughtup))
600 ProcessRepliesIfAny();
602 walsender_shutdown_requested = true;
605 /* Normal exit from the walsender is here */
606 if (walsender_shutdown_requested)
608 /* Inform the standby that XLOG streaming was done */
609 pq_puttextmessage('C', "COPY 0");
616 * If we had sent all accumulated WAL in last round, nap for the
617 * configured time before retrying.
622 * Even if we wrote all the WAL that was available when we started
623 * sending, more might have arrived while we were sending this
624 * batch. We had the latch set while sending, so we have not
625 * received any signals from that time. Let's arm the latch
626 * again, and after that check that we're still up-to-date.
628 ResetLatch(&MyWalSnd->latch);
630 if (!XLogSend(output_message, &caughtup))
632 if (caughtup && !got_SIGHUP && !walsender_ready_to_stop && !walsender_shutdown_requested)
635 * XXX: We don't really need the periodic wakeups anymore,
636 * WaitLatchOrSocket should reliably wake up as soon as
637 * something interesting happens.
641 WaitLatchOrSocket(&MyWalSnd->latch, MyProcPort->sock,
642 WalSndDelay * 1000L);
647 /* Attempt to send the log once every loop */
648 if (!XLogSend(output_message, &caughtup))
652 /* Update our state to indicate if we're behind or not */
653 WalSndSetState(caughtup ? WALSNDSTATE_STREAMING : WALSNDSTATE_CATCHUP);
654 ProcessRepliesIfAny();
658 * Get here on send failure. Clean up and exit.
660 * Reset whereToSendOutput to prevent ereport from attempting to send any
661 * more messages to the standby.
663 if (whereToSendOutput == DestRemote)
664 whereToSendOutput = DestNone;
667 return 1; /* keep the compiler quiet */
670 /* Initialize a per-walsender data structure for this walsender process */
677 * WalSndCtl should be set up already (we inherit this by fork() or
678 * EXEC_BACKEND mechanism from the postmaster).
680 Assert(WalSndCtl != NULL);
681 Assert(MyWalSnd == NULL);
684 * Find a free walsender slot and reserve it. If this fails, we must be
685 * out of WalSnd structures.
687 for (i = 0; i < max_wal_senders; i++)
689 /* use volatile pointer to prevent code rearrangement */
690 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
692 SpinLockAcquire(&walsnd->mutex);
694 if (walsnd->pid != 0)
696 SpinLockRelease(&walsnd->mutex);
702 * Found a free slot. Reserve it for us.
704 walsnd->pid = MyProcPid;
705 MemSet(&walsnd->sentPtr, 0, sizeof(XLogRecPtr));
706 walsnd->state = WALSNDSTATE_STARTUP;
707 SpinLockRelease(&walsnd->mutex);
708 /* don't need the lock anymore */
709 OwnLatch((Latch *) &walsnd->latch);
710 MyWalSnd = (WalSnd *) walsnd;
715 if (MyWalSnd == NULL)
717 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
718 errmsg("number of requested standby connections "
719 "exceeds max_wal_senders (currently %d)",
722 /* Arrange to clean up at walsender exit */
723 on_shmem_exit(WalSndKill, 0);
726 /* Destroy the per-walsender data structure for this walsender process */
728 WalSndKill(int code, Datum arg)
730 Assert(MyWalSnd != NULL);
733 * Mark WalSnd struct no longer in use. Assume that no lock is required
737 DisownLatch(&MyWalSnd->latch);
739 /* WalSnd struct isn't mine anymore */
744 * Read 'nbytes' bytes from WAL into 'buf', starting at location 'recptr'
746 * XXX probably this should be improved to suck data directly from the
747 * WAL buffers when possible.
749 * Will open, and keep open, one WAL segment stored in the global file
750 * descriptor sendFile. This means if XLogRead is used once, there will
751 * always be one descriptor left open until the process ends, but never
755 XLogRead(char *buf, XLogRecPtr recptr, Size nbytes)
757 XLogRecPtr startRecPtr = recptr;
758 char path[MAXPGPATH];
759 uint32 lastRemovedLog;
760 uint32 lastRemovedSeg;
770 startoff = recptr.xrecoff % XLogSegSize;
772 if (sendFile < 0 || !XLByteInSeg(recptr, sendId, sendSeg))
774 /* Switch to another logfile segment */
778 XLByteToSeg(recptr, sendId, sendSeg);
779 XLogFilePath(path, ThisTimeLineID, sendId, sendSeg);
781 sendFile = BasicOpenFile(path, O_RDONLY | PG_BINARY, 0);
785 * If the file is not found, assume it's because the standby
786 * asked for a too old WAL segment that has already been
787 * removed or recycled.
791 char filename[MAXFNAMELEN];
793 XLogFileName(filename, ThisTimeLineID, sendId, sendSeg);
795 (errcode_for_file_access(),
796 errmsg("requested WAL segment %s has already been removed",
801 (errcode_for_file_access(),
802 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
803 path, sendId, sendSeg)));
808 /* Need to seek in the file? */
809 if (sendOff != startoff)
811 if (lseek(sendFile, (off_t) startoff, SEEK_SET) < 0)
813 (errcode_for_file_access(),
814 errmsg("could not seek in log file %u, segment %u to offset %u: %m",
815 sendId, sendSeg, startoff)));
819 /* How many bytes are within this segment? */
820 if (nbytes > (XLogSegSize - startoff))
821 segbytes = XLogSegSize - startoff;
825 readbytes = read(sendFile, buf, segbytes);
828 (errcode_for_file_access(),
829 errmsg("could not read from log file %u, segment %u, offset %u, "
831 sendId, sendSeg, sendOff, (unsigned long) segbytes)));
833 /* Update state for read */
834 XLByteAdvance(recptr, readbytes);
836 sendOff += readbytes;
842 * After reading into the buffer, check that what we read was valid. We do
843 * this after reading, because even though the segment was present when we
844 * opened it, it might get recycled or removed while we read it. The
845 * read() succeeds in that case, but the data we tried to read might
846 * already have been overwritten with new WAL records.
848 XLogGetLastRemoved(&lastRemovedLog, &lastRemovedSeg);
849 XLByteToSeg(startRecPtr, log, seg);
850 if (log < lastRemovedLog ||
851 (log == lastRemovedLog && seg <= lastRemovedSeg))
853 char filename[MAXFNAMELEN];
855 XLogFileName(filename, ThisTimeLineID, log, seg);
857 (errcode_for_file_access(),
858 errmsg("requested WAL segment %s has already been removed",
864 * Read up to MAX_SEND_SIZE bytes of WAL that's been flushed to disk,
865 * but not yet sent to the client, and send it.
867 * msgbuf is a work area in which the output message is constructed. It's
868 * passed in just so we can avoid re-palloc'ing the buffer on each cycle.
869 * It must be of size 1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE.
871 * If there is no unsent WAL remaining, *caughtup is set to true, otherwise
872 * *caughtup is set to false.
874 * Returns true if OK, false if trouble.
877 XLogSend(char *msgbuf, bool *caughtup)
879 XLogRecPtr SendRqstPtr;
883 WalDataMessageHeader msghdr;
886 * Attempt to send all data that's already been written out and fsync'd to
887 * disk. We cannot go further than what's been written out given the
888 * current implementation of XLogRead(). And in any case it's unsafe to
889 * send WAL that is not securely down to disk on the master: if the master
890 * subsequently crashes and restarts, slaves must not have applied any WAL
891 * that gets lost on the master.
893 SendRqstPtr = GetFlushRecPtr();
895 /* Quick exit if nothing to do */
896 if (XLByteLE(SendRqstPtr, sentPtr))
903 * Figure out how much to send in one message. If there's no more than
904 * MAX_SEND_SIZE bytes to send, send everything. Otherwise send
905 * MAX_SEND_SIZE bytes, but round back to logfile or page boundary.
907 * The rounding is not only for performance reasons. Walreceiver relies on
908 * the fact that we never split a WAL record across two messages. Since a
909 * long WAL record is split at page boundary into continuation records,
910 * page boundary is always a safe cut-off point. We also assume that
911 * SendRqstPtr never points to the middle of a WAL record.
914 if (startptr.xrecoff >= XLogFileSize)
917 * crossing a logid boundary, skip the non-existent last log segment
918 * in previous logical log file.
920 startptr.xlogid += 1;
921 startptr.xrecoff = 0;
925 XLByteAdvance(endptr, MAX_SEND_SIZE);
926 if (endptr.xlogid != startptr.xlogid)
928 /* Don't cross a logfile boundary within one message */
929 Assert(endptr.xlogid == startptr.xlogid + 1);
930 endptr.xlogid = startptr.xlogid;
931 endptr.xrecoff = XLogFileSize;
934 /* if we went beyond SendRqstPtr, back off */
935 if (XLByteLE(SendRqstPtr, endptr))
937 endptr = SendRqstPtr;
942 /* round down to page boundary. */
943 endptr.xrecoff -= (endptr.xrecoff % XLOG_BLCKSZ);
947 nbytes = endptr.xrecoff - startptr.xrecoff;
948 Assert(nbytes <= MAX_SEND_SIZE);
951 * OK to read and send the slice.
956 * Read the log directly into the output buffer to avoid extra memcpy
959 XLogRead(msgbuf + 1 + sizeof(WalDataMessageHeader), startptr, nbytes);
962 * We fill the message header last so that the send timestamp is taken as
965 msghdr.dataStart = startptr;
966 msghdr.walEnd = SendRqstPtr;
967 msghdr.sendTime = GetCurrentTimestamp();
969 memcpy(msgbuf + 1, &msghdr, sizeof(WalDataMessageHeader));
971 pq_putmessage('d', msgbuf, 1 + sizeof(WalDataMessageHeader) + nbytes);
973 /* Flush pending output to the client */
979 /* Update shared memory status */
981 /* use volatile pointer to prevent code rearrangement */
982 volatile WalSnd *walsnd = MyWalSnd;
984 SpinLockAcquire(&walsnd->mutex);
985 walsnd->sentPtr = sentPtr;
986 SpinLockRelease(&walsnd->mutex);
989 /* Report progress of XLOG streaming in PS display */
990 if (update_process_title)
992 char activitymsg[50];
994 snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%X",
995 sentPtr.xlogid, sentPtr.xrecoff);
996 set_ps_display(activitymsg, false);
1002 /* SIGHUP: set flag to re-read config file at next convenient time */
1004 WalSndSigHupHandler(SIGNAL_ARGS)
1008 SetLatch(&MyWalSnd->latch);
1011 /* SIGTERM: set flag to shut down */
1013 WalSndShutdownHandler(SIGNAL_ARGS)
1015 walsender_shutdown_requested = true;
1017 SetLatch(&MyWalSnd->latch);
1021 * WalSndQuickDieHandler() occurs when signalled SIGQUIT by the postmaster.
1023 * Some backend has bought the farm,
1024 * so we need to stop what we're doing and exit.
1027 WalSndQuickDieHandler(SIGNAL_ARGS)
1029 PG_SETMASK(&BlockSig);
1032 * We DO NOT want to run proc_exit() callbacks -- we're here because
1033 * shared memory may be corrupted, so we don't want to try to clean up our
1034 * transaction. Just nail the windows shut and get out of town. Now that
1035 * there's an atexit callback to prevent third-party code from breaking
1036 * things by calling exit() directly, we have to reset the callbacks
1037 * explicitly to make this work as intended.
1042 * Note we do exit(2) not exit(0). This is to force the postmaster into a
1043 * system reset cycle if some idiot DBA sends a manual SIGQUIT to a random
1044 * backend. This is necessary precisely because we don't clean up our
1045 * shared memory state. (The "dead man switch" mechanism in pmsignal.c
1046 * should ensure the postmaster sees this as a crash, too, but no harm in
1047 * being doubly sure.)
1052 /* SIGUSR1: set flag to send WAL records */
1054 WalSndXLogSendHandler(SIGNAL_ARGS)
1056 latch_sigusr1_handler();
1059 /* SIGUSR2: set flag to do a last cycle and shut down afterwards */
1061 WalSndLastCycleHandler(SIGNAL_ARGS)
1063 walsender_ready_to_stop = true;
1065 SetLatch(&MyWalSnd->latch);
1068 /* Set up signal handlers */
1072 /* Set up signal handlers */
1073 pqsignal(SIGHUP, WalSndSigHupHandler); /* set flag to read config
1075 pqsignal(SIGINT, SIG_IGN); /* not used */
1076 pqsignal(SIGTERM, WalSndShutdownHandler); /* request shutdown */
1077 pqsignal(SIGQUIT, WalSndQuickDieHandler); /* hard crash time */
1078 pqsignal(SIGALRM, SIG_IGN);
1079 pqsignal(SIGPIPE, SIG_IGN);
1080 pqsignal(SIGUSR1, WalSndXLogSendHandler); /* request WAL sending */
1081 pqsignal(SIGUSR2, WalSndLastCycleHandler); /* request a last cycle and
1084 /* Reset some signals that are accepted by postmaster but not here */
1085 pqsignal(SIGCHLD, SIG_DFL);
1086 pqsignal(SIGTTIN, SIG_DFL);
1087 pqsignal(SIGTTOU, SIG_DFL);
1088 pqsignal(SIGCONT, SIG_DFL);
1089 pqsignal(SIGWINCH, SIG_DFL);
1092 /* Report shared-memory space needed by WalSndShmemInit */
1094 WalSndShmemSize(void)
1098 size = offsetof(WalSndCtlData, walsnds);
1099 size = add_size(size, mul_size(max_wal_senders, sizeof(WalSnd)));
1104 /* Allocate and initialize walsender-related shared memory */
1106 WalSndShmemInit(void)
1111 WalSndCtl = (WalSndCtlData *)
1112 ShmemInitStruct("Wal Sender Ctl", WalSndShmemSize(), &found);
1116 /* First time through, so initialize */
1117 MemSet(WalSndCtl, 0, WalSndShmemSize());
1119 for (i = 0; i < max_wal_senders; i++)
1121 WalSnd *walsnd = &WalSndCtl->walsnds[i];
1123 SpinLockInit(&walsnd->mutex);
1124 InitSharedLatch(&walsnd->latch);
1129 /* Wake up all walsenders */
1135 for (i = 0; i < max_wal_senders; i++)
1136 SetLatch(&WalSndCtl->walsnds[i].latch);
1139 /* Set state for current walsender (only called in walsender) */
1141 WalSndSetState(WalSndState state)
1143 /* use volatile pointer to prevent code rearrangement */
1144 volatile WalSnd *walsnd = MyWalSnd;
1146 Assert(am_walsender);
1148 if (walsnd->state == state)
1151 SpinLockAcquire(&walsnd->mutex);
1152 walsnd->state = state;
1153 SpinLockRelease(&walsnd->mutex);
1157 * Return a string constant representing the state. This is used
1158 * in system views, and should *not* be translated.
1161 WalSndGetStateString(WalSndState state)
1165 case WALSNDSTATE_STARTUP:
1167 case WALSNDSTATE_BACKUP:
1169 case WALSNDSTATE_CATCHUP:
1171 case WALSNDSTATE_STREAMING:
1179 * Returns activity of walsenders, including pids and xlog locations sent to
1183 pg_stat_get_wal_senders(PG_FUNCTION_ARGS)
1185 #define PG_STAT_GET_WAL_SENDERS_COLS 6
1186 ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1188 Tuplestorestate *tupstore;
1189 MemoryContext per_query_ctx;
1190 MemoryContext oldcontext;
1193 /* check to see if caller supports us returning a tuplestore */
1194 if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
1196 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1197 errmsg("set-valued function called in context that cannot accept a set")));
1198 if (!(rsinfo->allowedModes & SFRM_Materialize))
1200 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1201 errmsg("materialize mode required, but it is not " \
1202 "allowed in this context")));
1204 /* Build a tuple descriptor for our result type */
1205 if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1206 elog(ERROR, "return type must be a row type");
1208 per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
1209 oldcontext = MemoryContextSwitchTo(per_query_ctx);
1211 tupstore = tuplestore_begin_heap(true, false, work_mem);
1212 rsinfo->returnMode = SFRM_Materialize;
1213 rsinfo->setResult = tupstore;
1214 rsinfo->setDesc = tupdesc;
1216 MemoryContextSwitchTo(oldcontext);
1218 for (i = 0; i < max_wal_senders; i++)
1220 /* use volatile pointer to prevent code rearrangement */
1221 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1222 char location[MAXFNAMELEN];
1228 Datum values[PG_STAT_GET_WAL_SENDERS_COLS];
1229 bool nulls[PG_STAT_GET_WAL_SENDERS_COLS];
1231 if (walsnd->pid == 0)
1234 SpinLockAcquire(&walsnd->mutex);
1235 sentPtr = walsnd->sentPtr;
1236 state = walsnd->state;
1237 write = walsnd->write;
1238 flush = walsnd->flush;
1239 apply = walsnd->apply;
1240 SpinLockRelease(&walsnd->mutex);
1242 memset(nulls, 0, sizeof(nulls));
1243 values[0] = Int32GetDatum(walsnd->pid);
1248 * Only superusers can see details. Other users only get
1249 * the pid value to know it's a walsender, but no details.
1259 values[1] = CStringGetTextDatum(WalSndGetStateString(state));
1261 snprintf(location, sizeof(location), "%X/%X",
1262 sentPtr.xlogid, sentPtr.xrecoff);
1263 values[2] = CStringGetTextDatum(location);
1265 if (write.xlogid == 0 && write.xrecoff == 0)
1267 snprintf(location, sizeof(location), "%X/%X",
1268 write.xlogid, write.xrecoff);
1269 values[3] = CStringGetTextDatum(location);
1271 if (flush.xlogid == 0 && flush.xrecoff == 0)
1273 snprintf(location, sizeof(location), "%X/%X",
1274 flush.xlogid, flush.xrecoff);
1275 values[4] = CStringGetTextDatum(location);
1277 if (apply.xlogid == 0 && apply.xrecoff == 0)
1279 snprintf(location, sizeof(location), "%X/%X",
1280 apply.xlogid, apply.xrecoff);
1281 values[5] = CStringGetTextDatum(location);
1284 tuplestore_putvalues(tupstore, tupdesc, values, nulls);
1287 /* clean up and return the tuplestore */
1288 tuplestore_donestoring(tupstore);
1294 * This isn't currently used for anything. Monitoring tools might be
1295 * interested in the future, and we'll need something like this in the
1296 * future for synchronous replication.
1300 * Returns the oldest Send position among walsenders. Or InvalidXLogRecPtr
1304 GetOldestWALSendPointer(void)
1306 XLogRecPtr oldest = {0, 0};
1310 for (i = 0; i < max_wal_senders; i++)
1312 /* use volatile pointer to prevent code rearrangement */
1313 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1316 if (walsnd->pid == 0)
1319 SpinLockAcquire(&walsnd->mutex);
1320 recptr = walsnd->sentPtr;
1321 SpinLockRelease(&walsnd->mutex);
1323 if (recptr.xlogid == 0 && recptr.xrecoff == 0)
1326 if (!found || XLByteLT(recptr, oldest))