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 "catalog/pg_type.h"
43 #include "libpq/libpq.h"
44 #include "libpq/pqformat.h"
45 #include "libpq/pqsignal.h"
46 #include "miscadmin.h"
47 #include "replication/basebackup.h"
48 #include "replication/replnodes.h"
49 #include "replication/walprotocol.h"
50 #include "replication/walsender.h"
51 #include "storage/fd.h"
52 #include "storage/ipc.h"
53 #include "storage/pmsignal.h"
54 #include "tcop/tcopprot.h"
55 #include "utils/builtins.h"
56 #include "utils/guc.h"
57 #include "utils/memutils.h"
58 #include "utils/ps_status.h"
59 #include "utils/resowner.h"
62 /* Array of WalSnds in shared memory */
63 WalSndCtlData *WalSndCtl = NULL;
65 /* My slot in the shared memory array */
66 static WalSnd *MyWalSnd = NULL;
69 bool am_walsender = false; /* Am I a walsender process ? */
71 /* User-settable parameters for walsender */
72 int max_wal_senders = 0; /* the maximum number of concurrent walsenders */
73 int WalSndDelay = 200; /* max sleep time between some actions */
76 * These variables are used similarly to openLogFile/Id/Seg/Off,
77 * but for walsender to read the XLOG.
79 static int sendFile = -1;
80 static uint32 sendId = 0;
81 static uint32 sendSeg = 0;
82 static uint32 sendOff = 0;
85 * How far have we sent WAL already? This is also advertised in
86 * MyWalSnd->sentPtr. (Actually, this is the next WAL location to send.)
88 static XLogRecPtr sentPtr = {0, 0};
90 /* Flags set by signal handlers for later service in main loop */
91 static volatile sig_atomic_t got_SIGHUP = false;
92 volatile sig_atomic_t walsender_shutdown_requested = false;
93 volatile sig_atomic_t walsender_ready_to_stop = false;
96 static void WalSndSigHupHandler(SIGNAL_ARGS);
97 static void WalSndShutdownHandler(SIGNAL_ARGS);
98 static void WalSndQuickDieHandler(SIGNAL_ARGS);
99 static void WalSndXLogSendHandler(SIGNAL_ARGS);
100 static void WalSndLastCycleHandler(SIGNAL_ARGS);
102 /* Prototypes for private functions */
103 static bool HandleReplicationCommand(const char *cmd_string);
104 static int WalSndLoop(void);
105 static void InitWalSnd(void);
106 static void WalSndHandshake(void);
107 static void WalSndKill(int code, Datum arg);
108 static void XLogRead(char *buf, XLogRecPtr recptr, Size nbytes);
109 static bool XLogSend(char *msgbuf, bool *caughtup);
110 static void CheckClosedConnection(void);
111 static void IdentifySystem(void);
112 static void StartReplication(StartReplicationCmd * cmd);
115 /* Main entry point for walsender process */
119 MemoryContext walsnd_context;
121 if (RecoveryInProgress())
123 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
124 errmsg("recovery is still in progress, can't accept WAL streaming connections")));
126 /* Create a per-walsender data structure in shared memory */
130 * Create a memory context that we will do all our work in. We do this so
131 * that we can reset the context during error recovery and thereby avoid
132 * possible memory leaks. Formerly this code just ran in
133 * TopMemoryContext, but resetting that would be a really bad idea.
135 * XXX: we don't actually attempt error recovery in walsender, we just
136 * close the connection and exit.
138 walsnd_context = AllocSetContextCreate(TopMemoryContext,
140 ALLOCSET_DEFAULT_MINSIZE,
141 ALLOCSET_DEFAULT_INITSIZE,
142 ALLOCSET_DEFAULT_MAXSIZE);
143 MemoryContextSwitchTo(walsnd_context);
145 /* Set up resource owner */
146 CurrentResourceOwner = ResourceOwnerCreate(NULL, "walsender top-level resource owner");
148 /* Unblock signals (they were blocked when the postmaster forked us) */
149 PG_SETMASK(&UnBlockSig);
151 /* Tell the standby that walsender is ready for receiving commands */
152 ReadyForQuery(DestRemote);
154 /* Handle handshake messages before streaming */
157 /* Initialize shared memory status */
159 /* use volatile pointer to prevent code rearrangement */
160 volatile WalSnd *walsnd = MyWalSnd;
162 SpinLockAcquire(&walsnd->mutex);
163 walsnd->sentPtr = sentPtr;
164 SpinLockRelease(&walsnd->mutex);
167 /* Main loop of walsender */
172 * Execute commands from walreceiver, until we enter streaming mode.
175 WalSndHandshake(void)
177 StringInfoData input_message;
178 bool replication_started = false;
180 initStringInfo(&input_message);
182 while (!replication_started)
186 WalSndSetState(WALSNDSTATE_STARTUP);
187 set_ps_display("idle", false);
189 /* Wait for a command to arrive */
190 firstchar = pq_getbyte();
193 * Emergency bailout if postmaster has died. This is to avoid the
194 * necessity for manual cleanup of all postmaster children.
196 if (!PostmasterIsAlive(true))
200 * Check for any other interesting events that happened while we
206 ProcessConfigFile(PGC_SIGHUP);
209 if (firstchar != EOF)
212 * Read the message contents. This is expected to be done without
213 * blocking because we've been able to get message type code.
215 if (pq_getmessage(&input_message, 0))
216 firstchar = EOF; /* suitable message already logged */
219 /* Handle the very limited subset of commands expected in this phase */
222 case 'Q': /* Query message */
224 const char *query_string;
226 query_string = pq_getmsgstring(&input_message);
227 pq_getmsgend(&input_message);
229 if (HandleReplicationCommand(query_string))
230 replication_started = true;
235 /* standby is closing the connection */
239 /* standby disconnected unexpectedly */
241 (errcode(ERRCODE_PROTOCOL_VIOLATION),
242 errmsg("unexpected EOF on standby connection")));
247 (errcode(ERRCODE_PROTOCOL_VIOLATION),
248 errmsg("invalid standby handshake message type %d", firstchar)));
264 * Reply with a result set with one row, two columns. First col is system
265 * ID, and second is timeline ID
268 snprintf(sysid, sizeof(sysid), UINT64_FORMAT,
269 GetSystemIdentifier());
270 snprintf(tli, sizeof(tli), "%u", ThisTimeLineID);
272 /* Send a RowDescription message */
273 pq_beginmessage(&buf, 'T');
274 pq_sendint(&buf, 2, 2); /* 2 fields */
277 pq_sendstring(&buf, "systemid"); /* col name */
278 pq_sendint(&buf, 0, 4); /* table oid */
279 pq_sendint(&buf, 0, 2); /* attnum */
280 pq_sendint(&buf, TEXTOID, 4); /* type oid */
281 pq_sendint(&buf, -1, 2); /* typlen */
282 pq_sendint(&buf, 0, 4); /* typmod */
283 pq_sendint(&buf, 0, 2); /* format code */
286 pq_sendstring(&buf, "timeline"); /* col name */
287 pq_sendint(&buf, 0, 4); /* table oid */
288 pq_sendint(&buf, 0, 2); /* attnum */
289 pq_sendint(&buf, INT4OID, 4); /* type oid */
290 pq_sendint(&buf, 4, 2); /* typlen */
291 pq_sendint(&buf, 0, 4); /* typmod */
292 pq_sendint(&buf, 0, 2); /* format code */
295 /* Send a DataRow message */
296 pq_beginmessage(&buf, 'D');
297 pq_sendint(&buf, 2, 2); /* # of columns */
298 pq_sendint(&buf, strlen(sysid), 4); /* col1 len */
299 pq_sendbytes(&buf, (char *) &sysid, strlen(sysid));
300 pq_sendint(&buf, strlen(tli), 4); /* col2 len */
301 pq_sendbytes(&buf, (char *) tli, strlen(tli));
304 /* Send CommandComplete and ReadyForQuery messages */
305 EndCommand("SELECT", DestRemote);
306 ReadyForQuery(DestRemote);
307 /* ReadyForQuery did pq_flush for us */
314 StartReplication(StartReplicationCmd * cmd)
319 * Let postmaster know that we're streaming. Once we've declared us as
320 * a WAL sender process, postmaster will let us outlive the bgwriter and
321 * kill us last in the shutdown sequence, so we get a chance to stream
322 * all remaining WAL at shutdown, including the shutdown checkpoint.
323 * Note that there's no going back, and we mustn't write any WAL records
326 MarkPostmasterChildWalSender();
329 * Check that we're logging enough information in the WAL for
332 * NOTE: This only checks the current value of wal_level. Even if the
333 * current setting is not 'minimal', there can be old WAL in the pg_xlog
334 * directory that was created with 'minimal'. So this is not bulletproof,
335 * the purpose is just to give a user-friendly error message that hints
336 * how to configure the system correctly.
338 if (wal_level == WAL_LEVEL_MINIMAL)
340 (errcode(ERRCODE_CANNOT_CONNECT_NOW),
341 errmsg("standby connections not allowed because wal_level=minimal")));
343 /* Send a CopyBothResponse message, and start streaming */
344 pq_beginmessage(&buf, 'W');
345 pq_sendbyte(&buf, 0);
346 pq_sendint(&buf, 0, 2);
351 * Initialize position to the received one, then the xlog records begin to
352 * be shipped from that position
354 sentPtr = cmd->startpoint;
358 * Execute an incoming replication command.
361 HandleReplicationCommand(const char *cmd_string)
363 bool replication_started = false;
366 MemoryContext cmd_context;
367 MemoryContext old_context;
369 elog(DEBUG1, "received replication command: %s", cmd_string);
371 cmd_context = AllocSetContextCreate(CurrentMemoryContext,
372 "Replication command context",
373 ALLOCSET_DEFAULT_MINSIZE,
374 ALLOCSET_DEFAULT_INITSIZE,
375 ALLOCSET_DEFAULT_MAXSIZE);
376 old_context = MemoryContextSwitchTo(cmd_context);
378 replication_scanner_init(cmd_string);
379 parse_rc = replication_yyparse();
382 (errcode(ERRCODE_SYNTAX_ERROR),
383 (errmsg_internal("replication command parser returned %d",
386 cmd_node = replication_parse_result;
388 switch (cmd_node->type)
390 case T_IdentifySystemCmd:
394 case T_StartReplicationCmd:
395 StartReplication((StartReplicationCmd *) cmd_node);
397 /* break out of the loop */
398 replication_started = true;
401 case T_BaseBackupCmd:
403 BaseBackupCmd *cmd = (BaseBackupCmd *) cmd_node;
405 SendBaseBackup(cmd->label, cmd->progress);
407 /* Send CommandComplete and ReadyForQuery messages */
408 EndCommand("SELECT", DestRemote);
409 ReadyForQuery(DestRemote);
410 /* ReadyForQuery did pq_flush for us */
416 (errcode(ERRCODE_PROTOCOL_VIOLATION),
417 errmsg("invalid standby query string: %s", cmd_string)));
421 MemoryContextSwitchTo(old_context);
422 MemoryContextDelete(cmd_context);
424 return replication_started;
428 * Check if the remote end has closed the connection.
431 CheckClosedConnection(void)
433 unsigned char firstchar;
436 r = pq_getbyte_if_available(&firstchar);
439 /* unexpected error or EOF */
441 (errcode(ERRCODE_PROTOCOL_VIOLATION),
442 errmsg("unexpected EOF on standby connection")));
447 /* no data available without blocking */
451 /* Handle the very limited subset of commands expected in this phase */
455 * 'X' means that the standby is closing down the socket.
462 (errcode(ERRCODE_PROTOCOL_VIOLATION),
463 errmsg("invalid standby closing message type %d",
468 /* Main loop of walsender process */
472 char *output_message;
473 bool caughtup = false;
476 * Allocate buffer that will be used for each output message. We do this
477 * just once to reduce palloc overhead. The buffer must be made large
478 * enough for maximum-sized messages.
480 output_message = palloc(1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE);
482 /* Loop forever, unless we get an error */
486 * Emergency bailout if postmaster has died. This is to avoid the
487 * necessity for manual cleanup of all postmaster children.
489 if (!PostmasterIsAlive(true))
492 /* Process any requests or signals received recently */
496 ProcessConfigFile(PGC_SIGHUP);
500 * When SIGUSR2 arrives, we send all outstanding logs up to the
501 * shutdown checkpoint record (i.e., the latest record) and exit.
503 if (walsender_ready_to_stop)
505 if (!XLogSend(output_message, &caughtup))
508 walsender_shutdown_requested = true;
511 /* Normal exit from the walsender is here */
512 if (walsender_shutdown_requested)
514 /* Inform the standby that XLOG streaming was done */
515 pq_puttextmessage('C', "COPY 0");
522 * If we had sent all accumulated WAL in last round, nap for the
523 * configured time before retrying.
528 * Even if we wrote all the WAL that was available when we started
529 * sending, more might have arrived while we were sending this
530 * batch. We had the latch set while sending, so we have not
531 * received any signals from that time. Let's arm the latch
532 * again, and after that check that we're still up-to-date.
534 ResetLatch(&MyWalSnd->latch);
536 if (!XLogSend(output_message, &caughtup))
538 if (caughtup && !got_SIGHUP && !walsender_ready_to_stop && !walsender_shutdown_requested)
541 * XXX: We don't really need the periodic wakeups anymore,
542 * WaitLatchOrSocket should reliably wake up as soon as
543 * something interesting happens.
547 WaitLatchOrSocket(&MyWalSnd->latch, MyProcPort->sock,
548 WalSndDelay * 1000L);
551 /* Check if the connection was closed */
552 CheckClosedConnection();
556 /* Attempt to send the log once every loop */
557 if (!XLogSend(output_message, &caughtup))
561 /* Update our state to indicate if we're behind or not */
562 WalSndSetState(caughtup ? WALSNDSTATE_STREAMING : WALSNDSTATE_CATCHUP);
566 * Get here on send failure. Clean up and exit.
568 * Reset whereToSendOutput to prevent ereport from attempting to send any
569 * more messages to the standby.
571 if (whereToSendOutput == DestRemote)
572 whereToSendOutput = DestNone;
575 return 1; /* keep the compiler quiet */
578 /* Initialize a per-walsender data structure for this walsender process */
585 * WalSndCtl should be set up already (we inherit this by fork() or
586 * EXEC_BACKEND mechanism from the postmaster).
588 Assert(WalSndCtl != NULL);
589 Assert(MyWalSnd == NULL);
592 * Find a free walsender slot and reserve it. If this fails, we must be
593 * out of WalSnd structures.
595 for (i = 0; i < max_wal_senders; i++)
597 /* use volatile pointer to prevent code rearrangement */
598 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
600 SpinLockAcquire(&walsnd->mutex);
602 if (walsnd->pid != 0)
604 SpinLockRelease(&walsnd->mutex);
610 * Found a free slot. Reserve it for us.
612 walsnd->pid = MyProcPid;
613 MemSet(&walsnd->sentPtr, 0, sizeof(XLogRecPtr));
614 walsnd->state = WALSNDSTATE_STARTUP;
615 SpinLockRelease(&walsnd->mutex);
616 /* don't need the lock anymore */
617 OwnLatch((Latch *) &walsnd->latch);
618 MyWalSnd = (WalSnd *) walsnd;
623 if (MyWalSnd == NULL)
625 (errcode(ERRCODE_TOO_MANY_CONNECTIONS),
626 errmsg("number of requested standby connections "
627 "exceeds max_wal_senders (currently %d)",
630 /* Arrange to clean up at walsender exit */
631 on_shmem_exit(WalSndKill, 0);
634 /* Destroy the per-walsender data structure for this walsender process */
636 WalSndKill(int code, Datum arg)
638 Assert(MyWalSnd != NULL);
641 * Mark WalSnd struct no longer in use. Assume that no lock is required
645 DisownLatch(&MyWalSnd->latch);
647 /* WalSnd struct isn't mine anymore */
652 * Read 'nbytes' bytes from WAL into 'buf', starting at location 'recptr'
654 * XXX probably this should be improved to suck data directly from the
655 * WAL buffers when possible.
658 XLogRead(char *buf, XLogRecPtr recptr, Size nbytes)
660 XLogRecPtr startRecPtr = recptr;
661 char path[MAXPGPATH];
662 uint32 lastRemovedLog;
663 uint32 lastRemovedSeg;
673 startoff = recptr.xrecoff % XLogSegSize;
675 if (sendFile < 0 || !XLByteInSeg(recptr, sendId, sendSeg))
677 /* Switch to another logfile segment */
681 XLByteToSeg(recptr, sendId, sendSeg);
682 XLogFilePath(path, ThisTimeLineID, sendId, sendSeg);
684 sendFile = BasicOpenFile(path, O_RDONLY | PG_BINARY, 0);
688 * If the file is not found, assume it's because the standby
689 * asked for a too old WAL segment that has already been
690 * removed or recycled.
694 char filename[MAXFNAMELEN];
696 XLogFileName(filename, ThisTimeLineID, sendId, sendSeg);
698 (errcode_for_file_access(),
699 errmsg("requested WAL segment %s has already been removed",
704 (errcode_for_file_access(),
705 errmsg("could not open file \"%s\" (log file %u, segment %u): %m",
706 path, sendId, sendSeg)));
711 /* Need to seek in the file? */
712 if (sendOff != startoff)
714 if (lseek(sendFile, (off_t) startoff, SEEK_SET) < 0)
716 (errcode_for_file_access(),
717 errmsg("could not seek in log file %u, segment %u to offset %u: %m",
718 sendId, sendSeg, startoff)));
722 /* How many bytes are within this segment? */
723 if (nbytes > (XLogSegSize - startoff))
724 segbytes = XLogSegSize - startoff;
728 readbytes = read(sendFile, buf, segbytes);
731 (errcode_for_file_access(),
732 errmsg("could not read from log file %u, segment %u, offset %u, "
734 sendId, sendSeg, sendOff, (unsigned long) segbytes)));
736 /* Update state for read */
737 XLByteAdvance(recptr, readbytes);
739 sendOff += readbytes;
745 * After reading into the buffer, check that what we read was valid. We do
746 * this after reading, because even though the segment was present when we
747 * opened it, it might get recycled or removed while we read it. The
748 * read() succeeds in that case, but the data we tried to read might
749 * already have been overwritten with new WAL records.
751 XLogGetLastRemoved(&lastRemovedLog, &lastRemovedSeg);
752 XLByteToSeg(startRecPtr, log, seg);
753 if (log < lastRemovedLog ||
754 (log == lastRemovedLog && seg <= lastRemovedSeg))
756 char filename[MAXFNAMELEN];
758 XLogFileName(filename, ThisTimeLineID, log, seg);
760 (errcode_for_file_access(),
761 errmsg("requested WAL segment %s has already been removed",
767 * Read up to MAX_SEND_SIZE bytes of WAL that's been flushed to disk,
768 * but not yet sent to the client, and send it.
770 * msgbuf is a work area in which the output message is constructed. It's
771 * passed in just so we can avoid re-palloc'ing the buffer on each cycle.
772 * It must be of size 1 + sizeof(WalDataMessageHeader) + MAX_SEND_SIZE.
774 * If there is no unsent WAL remaining, *caughtup is set to true, otherwise
775 * *caughtup is set to false.
777 * Returns true if OK, false if trouble.
780 XLogSend(char *msgbuf, bool *caughtup)
782 XLogRecPtr SendRqstPtr;
786 WalDataMessageHeader msghdr;
789 * Attempt to send all data that's already been written out and fsync'd to
790 * disk. We cannot go further than what's been written out given the
791 * current implementation of XLogRead(). And in any case it's unsafe to
792 * send WAL that is not securely down to disk on the master: if the master
793 * subsequently crashes and restarts, slaves must not have applied any WAL
794 * that gets lost on the master.
796 SendRqstPtr = GetFlushRecPtr();
798 /* Quick exit if nothing to do */
799 if (XLByteLE(SendRqstPtr, sentPtr))
806 * Figure out how much to send in one message. If there's no more than
807 * MAX_SEND_SIZE bytes to send, send everything. Otherwise send
808 * MAX_SEND_SIZE bytes, but round back to logfile or page boundary.
810 * The rounding is not only for performance reasons. Walreceiver relies on
811 * the fact that we never split a WAL record across two messages. Since a
812 * long WAL record is split at page boundary into continuation records,
813 * page boundary is always a safe cut-off point. We also assume that
814 * SendRqstPtr never points to the middle of a WAL record.
817 if (startptr.xrecoff >= XLogFileSize)
820 * crossing a logid boundary, skip the non-existent last log segment
821 * in previous logical log file.
823 startptr.xlogid += 1;
824 startptr.xrecoff = 0;
828 XLByteAdvance(endptr, MAX_SEND_SIZE);
829 if (endptr.xlogid != startptr.xlogid)
831 /* Don't cross a logfile boundary within one message */
832 Assert(endptr.xlogid == startptr.xlogid + 1);
833 endptr.xlogid = startptr.xlogid;
834 endptr.xrecoff = XLogFileSize;
837 /* if we went beyond SendRqstPtr, back off */
838 if (XLByteLE(SendRqstPtr, endptr))
840 endptr = SendRqstPtr;
845 /* round down to page boundary. */
846 endptr.xrecoff -= (endptr.xrecoff % XLOG_BLCKSZ);
850 nbytes = endptr.xrecoff - startptr.xrecoff;
851 Assert(nbytes <= MAX_SEND_SIZE);
854 * OK to read and send the slice.
859 * Read the log directly into the output buffer to avoid extra memcpy
862 XLogRead(msgbuf + 1 + sizeof(WalDataMessageHeader), startptr, nbytes);
865 * We fill the message header last so that the send timestamp is taken as
868 msghdr.dataStart = startptr;
869 msghdr.walEnd = SendRqstPtr;
870 msghdr.sendTime = GetCurrentTimestamp();
872 memcpy(msgbuf + 1, &msghdr, sizeof(WalDataMessageHeader));
874 pq_putmessage('d', msgbuf, 1 + sizeof(WalDataMessageHeader) + nbytes);
876 /* Flush pending output to the client */
882 /* Update shared memory status */
884 /* use volatile pointer to prevent code rearrangement */
885 volatile WalSnd *walsnd = MyWalSnd;
887 SpinLockAcquire(&walsnd->mutex);
888 walsnd->sentPtr = sentPtr;
889 SpinLockRelease(&walsnd->mutex);
892 /* Report progress of XLOG streaming in PS display */
893 if (update_process_title)
895 char activitymsg[50];
897 snprintf(activitymsg, sizeof(activitymsg), "streaming %X/%X",
898 sentPtr.xlogid, sentPtr.xrecoff);
899 set_ps_display(activitymsg, false);
905 /* SIGHUP: set flag to re-read config file at next convenient time */
907 WalSndSigHupHandler(SIGNAL_ARGS)
911 SetLatch(&MyWalSnd->latch);
914 /* SIGTERM: set flag to shut down */
916 WalSndShutdownHandler(SIGNAL_ARGS)
918 walsender_shutdown_requested = true;
920 SetLatch(&MyWalSnd->latch);
924 * WalSndQuickDieHandler() occurs when signalled SIGQUIT by the postmaster.
926 * Some backend has bought the farm,
927 * so we need to stop what we're doing and exit.
930 WalSndQuickDieHandler(SIGNAL_ARGS)
932 PG_SETMASK(&BlockSig);
935 * We DO NOT want to run proc_exit() callbacks -- we're here because
936 * shared memory may be corrupted, so we don't want to try to clean up our
937 * transaction. Just nail the windows shut and get out of town. Now that
938 * there's an atexit callback to prevent third-party code from breaking
939 * things by calling exit() directly, we have to reset the callbacks
940 * explicitly to make this work as intended.
945 * Note we do exit(2) not exit(0). This is to force the postmaster into a
946 * system reset cycle if some idiot DBA sends a manual SIGQUIT to a random
947 * backend. This is necessary precisely because we don't clean up our
948 * shared memory state. (The "dead man switch" mechanism in pmsignal.c
949 * should ensure the postmaster sees this as a crash, too, but no harm in
950 * being doubly sure.)
955 /* SIGUSR1: set flag to send WAL records */
957 WalSndXLogSendHandler(SIGNAL_ARGS)
959 latch_sigusr1_handler();
962 /* SIGUSR2: set flag to do a last cycle and shut down afterwards */
964 WalSndLastCycleHandler(SIGNAL_ARGS)
966 walsender_ready_to_stop = true;
968 SetLatch(&MyWalSnd->latch);
971 /* Set up signal handlers */
975 /* Set up signal handlers */
976 pqsignal(SIGHUP, WalSndSigHupHandler); /* set flag to read config
978 pqsignal(SIGINT, SIG_IGN); /* not used */
979 pqsignal(SIGTERM, WalSndShutdownHandler); /* request shutdown */
980 pqsignal(SIGQUIT, WalSndQuickDieHandler); /* hard crash time */
981 pqsignal(SIGALRM, SIG_IGN);
982 pqsignal(SIGPIPE, SIG_IGN);
983 pqsignal(SIGUSR1, WalSndXLogSendHandler); /* request WAL sending */
984 pqsignal(SIGUSR2, WalSndLastCycleHandler); /* request a last cycle and
987 /* Reset some signals that are accepted by postmaster but not here */
988 pqsignal(SIGCHLD, SIG_DFL);
989 pqsignal(SIGTTIN, SIG_DFL);
990 pqsignal(SIGTTOU, SIG_DFL);
991 pqsignal(SIGCONT, SIG_DFL);
992 pqsignal(SIGWINCH, SIG_DFL);
995 /* Report shared-memory space needed by WalSndShmemInit */
997 WalSndShmemSize(void)
1001 size = offsetof(WalSndCtlData, walsnds);
1002 size = add_size(size, mul_size(max_wal_senders, sizeof(WalSnd)));
1007 /* Allocate and initialize walsender-related shared memory */
1009 WalSndShmemInit(void)
1014 WalSndCtl = (WalSndCtlData *)
1015 ShmemInitStruct("Wal Sender Ctl", WalSndShmemSize(), &found);
1019 /* First time through, so initialize */
1020 MemSet(WalSndCtl, 0, WalSndShmemSize());
1022 for (i = 0; i < max_wal_senders; i++)
1024 WalSnd *walsnd = &WalSndCtl->walsnds[i];
1026 SpinLockInit(&walsnd->mutex);
1027 InitSharedLatch(&walsnd->latch);
1032 /* Wake up all walsenders */
1038 for (i = 0; i < max_wal_senders; i++)
1039 SetLatch(&WalSndCtl->walsnds[i].latch);
1042 /* Set state for current walsender (only called in walsender) */
1044 WalSndSetState(WalSndState state)
1046 /* use volatile pointer to prevent code rearrangement */
1047 volatile WalSnd *walsnd = MyWalSnd;
1049 Assert(am_walsender);
1051 if (walsnd->state == state)
1054 SpinLockAcquire(&walsnd->mutex);
1055 walsnd->state = state;
1056 SpinLockRelease(&walsnd->mutex);
1060 * Return a string constant representing the state. This is used
1061 * in system views, and should *not* be translated.
1064 WalSndGetStateString(WalSndState state)
1068 case WALSNDSTATE_STARTUP:
1070 case WALSNDSTATE_BACKUP:
1072 case WALSNDSTATE_CATCHUP:
1074 case WALSNDSTATE_STREAMING:
1082 * Returns activity of walsenders, including pids and xlog locations sent to
1086 pg_stat_get_wal_senders(PG_FUNCTION_ARGS)
1088 #define PG_STAT_GET_WAL_SENDERS_COLS 3
1089 ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
1091 Tuplestorestate *tupstore;
1092 MemoryContext per_query_ctx;
1093 MemoryContext oldcontext;
1096 /* check to see if caller supports us returning a tuplestore */
1097 if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
1099 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1100 errmsg("set-valued function called in context that cannot accept a set")));
1101 if (!(rsinfo->allowedModes & SFRM_Materialize))
1103 (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1104 errmsg("materialize mode required, but it is not " \
1105 "allowed in this context")));
1107 /* Build a tuple descriptor for our result type */
1108 if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
1109 elog(ERROR, "return type must be a row type");
1111 per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
1112 oldcontext = MemoryContextSwitchTo(per_query_ctx);
1114 tupstore = tuplestore_begin_heap(true, false, work_mem);
1115 rsinfo->returnMode = SFRM_Materialize;
1116 rsinfo->setResult = tupstore;
1117 rsinfo->setDesc = tupdesc;
1119 MemoryContextSwitchTo(oldcontext);
1121 for (i = 0; i < max_wal_senders; i++)
1123 /* use volatile pointer to prevent code rearrangement */
1124 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1125 char sent_location[MAXFNAMELEN];
1128 Datum values[PG_STAT_GET_WAL_SENDERS_COLS];
1129 bool nulls[PG_STAT_GET_WAL_SENDERS_COLS];
1131 if (walsnd->pid == 0)
1134 SpinLockAcquire(&walsnd->mutex);
1135 sentPtr = walsnd->sentPtr;
1136 state = walsnd->state;
1137 SpinLockRelease(&walsnd->mutex);
1139 snprintf(sent_location, sizeof(sent_location), "%X/%X",
1140 sentPtr.xlogid, sentPtr.xrecoff);
1142 memset(nulls, 0, sizeof(nulls));
1143 values[0] = Int32GetDatum(walsnd->pid);
1144 values[1] = CStringGetTextDatum(WalSndGetStateString(state));
1145 values[2] = CStringGetTextDatum(sent_location);
1147 tuplestore_putvalues(tupstore, tupdesc, values, nulls);
1150 /* clean up and return the tuplestore */
1151 tuplestore_donestoring(tupstore);
1157 * This isn't currently used for anything. Monitoring tools might be
1158 * interested in the future, and we'll need something like this in the
1159 * future for synchronous replication.
1163 * Returns the oldest Send position among walsenders. Or InvalidXLogRecPtr
1167 GetOldestWALSendPointer(void)
1169 XLogRecPtr oldest = {0, 0};
1173 for (i = 0; i < max_wal_senders; i++)
1175 /* use volatile pointer to prevent code rearrangement */
1176 volatile WalSnd *walsnd = &WalSndCtl->walsnds[i];
1179 if (walsnd->pid == 0)
1182 SpinLockAcquire(&walsnd->mutex);
1183 recptr = walsnd->sentPtr;
1184 SpinLockRelease(&walsnd->mutex);
1186 if (recptr.xlogid == 0 && recptr.xrecoff == 0)
1189 if (!found || XLByteLT(recptr, oldest))