From 1cc3c542b53033ec3b0ad68f2a709f565dd87a3f Mon Sep 17 00:00:00 2001 From: Peter Eisentraut Date: Mon, 1 May 2017 15:09:06 -0400 Subject: [PATCH] Prevent panic during shutdown checkpoint When the checkpointer writes the shutdown checkpoint, it checks afterwards whether any WAL has been written since it started and throws a PANIC if so. At that point, only walsenders are still active, so one might think this could not happen, but walsenders can also generate WAL, for instance in BASE_BACKUP and certain variants of CREATE_REPLICATION_SLOT. So they can trigger this panic if such a command is run while the shutdown checkpoint is being written. To fix this, divide the walsender shutdown into two phases. First, the postmaster sends a SIGUSR2 signal to all walsenders. The walsenders then put themselves into the "stopping" state. In this state, they reject any new commands. (For simplicity, we reject all new commands, so that in the future we do not have to track meticulously which commands might generate WAL.) The checkpointer waits for all walsenders to reach this state before proceeding with the shutdown checkpoint. After the shutdown checkpoint is done, the postmaster sends SIGINT (previously unused) to the walsenders. This triggers the existing shutdown behavior of sending out the shutdown checkpoint record and then terminating. Author: Michael Paquier Reported-by: Fujii Masao --- src/backend/access/transam/xlog.c | 6 ++ src/backend/postmaster/postmaster.c | 7 +- src/backend/replication/walsender.c | 143 ++++++++++++++++++++++++---- src/include/replication/walsender.h | 1 + src/include/replication/walsender_private.h | 3 +- 5 files changed, 136 insertions(+), 24 deletions(-) diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index 3cc74bfe2e..8ae6f092a3 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7949,6 +7949,12 @@ ShutdownXLOG(int code, Datum arg) ereport(IsPostmasterEnvironment ? LOG : NOTICE, (errmsg("shutting down"))); + /* + * Wait for WAL senders to be in stopping state. This prevents commands + * from writing new WAL. + */ + WalSndWaitStopping(); + if (RecoveryInProgress()) CreateRestartPoint(CHECKPOINT_IS_SHUTDOWN | CHECKPOINT_IMMEDIATE); else diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c index 80d3c52241..259a441941 100644 --- a/src/backend/postmaster/postmaster.c +++ b/src/backend/postmaster/postmaster.c @@ -2852,7 +2852,7 @@ reaper(SIGNAL_ARGS) * Waken walsenders for the last time. No regular backends * should be around anymore. */ - SignalChildren(SIGUSR2); + SignalChildren(SIGINT); pmState = PM_SHUTDOWN_2; @@ -3590,7 +3590,9 @@ PostmasterStateMachine(void) /* * If we get here, we are proceeding with normal shutdown. All * the regular children are gone, and it's time to tell the - * checkpointer to do a shutdown checkpoint. + * checkpointer to do a shutdown checkpoint. All WAL senders + * are told to switch to a stopping state so that the shutdown + * checkpoint can go ahead. */ Assert(Shutdown > NoShutdown); /* Start the checkpointer if not running */ @@ -3599,6 +3601,7 @@ PostmasterStateMachine(void) /* And tell it to shut down */ if (CheckpointerPID != 0) { + SignalSomeChildren(SIGUSR2, BACKEND_TYPE_WALSND); signal_child(CheckpointerPID, SIGUSR2); pmState = PM_SHUTDOWN; } diff --git a/src/backend/replication/walsender.c b/src/backend/replication/walsender.c index 31e9c8ad6d..de76d41030 100644 --- a/src/backend/replication/walsender.c +++ b/src/backend/replication/walsender.c @@ -24,11 +24,14 @@ * are treated as not a crash but approximately normal termination; * the walsender will exit quickly without sending any more XLOG records. * - * If the server is shut down, postmaster sends us SIGUSR2 after all - * regular backends have exited and the shutdown checkpoint has been written. - * This instructs walsender to send any outstanding WAL, including the - * shutdown checkpoint record, wait for it to be replicated to the standby, - * and then exit. + * If the server is shut down, postmaster sends us SIGUSR2 after all regular + * backends have exited. This causes the walsender to switch to the "stopping" + * state. In this state, the walsender will reject any replication command + * that may generate WAL activity. The checkpointer begins the shutdown + * checkpoint once all walsenders are confirmed as stopping. When the shutdown + * checkpoint finishes, the postmaster sends us SIGINT. This instructs + * walsender to send any outstanding WAL, including the shutdown checkpoint + * record, wait for it to be replicated to the standby, and then exit. * * * Portions Copyright (c) 2010-2015, PostgreSQL Global Development Group @@ -170,13 +173,14 @@ static bool WalSndCaughtUp = false; /* Flags set by signal handlers for later service in main loop */ static volatile sig_atomic_t got_SIGHUP = false; -static volatile sig_atomic_t walsender_ready_to_stop = false; +static volatile sig_atomic_t got_SIGINT = false; +static volatile sig_atomic_t got_SIGUSR2 = false; /* - * This is set while we are streaming. When not set, SIGUSR2 signal will be + * This is set while we are streaming. When not set, SIGINT signal will be * handled like SIGTERM. When set, the main loop is responsible for checking - * walsender_ready_to_stop and terminating when it's set (after streaming any - * remaining WAL). + * got_SIGINT and terminating when it's set (after streaming any remaining + * WAL). */ static volatile sig_atomic_t replication_active = false; @@ -186,6 +190,7 @@ static XLogRecPtr logical_startptr = InvalidXLogRecPtr; /* Signal handlers */ static void WalSndSigHupHandler(SIGNAL_ARGS); static void WalSndXLogSendHandler(SIGNAL_ARGS); +static void WalSndSwitchStopping(SIGNAL_ARGS); static void WalSndLastCycleHandler(SIGNAL_ARGS); /* Prototypes for private functions */ @@ -263,11 +268,14 @@ WalSndErrorCleanup() ReplicationSlotRelease(); replication_active = false; - if (walsender_ready_to_stop) + if (got_SIGINT) proc_exit(0); /* Revert back to startup state */ WalSndSetState(WALSNDSTATE_STARTUP); + + if (got_SIGUSR2) + WalSndSetState(WALSNDSTATE_STOPPING); } /* @@ -671,7 +679,7 @@ StartReplication(StartReplicationCmd *cmd) WalSndLoop(XLogSendPhysical); replication_active = false; - if (walsender_ready_to_stop) + if (got_SIGINT) proc_exit(0); WalSndSetState(WALSNDSTATE_STARTUP); @@ -962,7 +970,7 @@ StartLogicalReplication(StartReplicationCmd *cmd) { ereport(LOG, (errmsg("terminating walsender process after promotion"))); - walsender_ready_to_stop = true; + got_SIGINT = true; } WalSndSetState(WALSNDSTATE_CATCHUP); @@ -1017,7 +1025,7 @@ StartLogicalReplication(StartReplicationCmd *cmd) ReplicationSlotRelease(); replication_active = false; - if (walsender_ready_to_stop) + if (got_SIGINT) proc_exit(0); WalSndSetState(WALSNDSTATE_STARTUP); @@ -1204,6 +1212,14 @@ WalSndWaitForWal(XLogRecPtr loc) RecentFlushPtr = GetXLogReplayRecPtr(NULL); /* + * If postmaster asked us to switch to the stopping state, do so. + * Shutdown is in progress and this will allow the checkpointer to + * move on with the shutdown checkpoint. + */ + if (got_SIGUSR2) + WalSndSetState(WALSNDSTATE_STOPPING); + + /* * If postmaster asked us to stop, don't wait here anymore. This will * cause the xlogreader to return without reading a full record, which * is the fastest way to reach the mainloop which then can quit. @@ -1212,7 +1228,7 @@ WalSndWaitForWal(XLogRecPtr loc) * RecentFlushPtr, so we can send all remaining data before shutting * down. */ - if (walsender_ready_to_stop) + if (got_SIGINT) break; /* @@ -1283,6 +1299,22 @@ exec_replication_command(const char *cmd_string) MemoryContext old_context; /* + * If WAL sender has been told that shutdown is getting close, switch its + * status accordingly to handle the next replication commands correctly. + */ + if (got_SIGUSR2) + WalSndSetState(WALSNDSTATE_STOPPING); + + /* + * Throw error if in stopping mode. We need prevent commands that could + * generate WAL while the shutdown checkpoint is being written. To be + * safe, we just prohibit all new commands. + */ + if (MyWalSnd->state == WALSNDSTATE_STOPPING) + ereport(ERROR, + (errmsg("cannot execute new commands while WAL sender is in stopping mode"))); + + /* * Log replication command if log_replication_commands is enabled. Even * when it's disabled, log the command with DEBUG1 level for backward * compatibility. @@ -1876,13 +1908,20 @@ WalSndLoop(WalSndSendDataCallback send_data) } /* - * When SIGUSR2 arrives, we send any outstanding logs up to the + * At the reception of SIGUSR2, switch the WAL sender to the stopping + * state. + */ + if (got_SIGUSR2) + WalSndSetState(WALSNDSTATE_STOPPING); + + /* + * When SIGINT arrives, we send any outstanding logs up to the * shutdown checkpoint record (i.e., the latest record), wait for * them to be replicated to the standby, and exit. This may be a * normal termination at shutdown, or a promotion, the walsender * is not sure which. */ - if (walsender_ready_to_stop) + if (got_SIGINT) WalSndDone(send_data); } @@ -2590,7 +2629,23 @@ WalSndXLogSendHandler(SIGNAL_ARGS) errno = save_errno; } -/* SIGUSR2: set flag to do a last cycle and shut down afterwards */ +/* SIGUSR2: set flag to switch to stopping state */ +static void +WalSndSwitchStopping(SIGNAL_ARGS) +{ + int save_errno = errno; + + got_SIGUSR2 = true; + SetLatch(MyLatch); + + errno = save_errno; +} + +/* + * SIGINT: set flag to do a last cycle and shut down afterwards. The WAL + * sender should already have been switched to WALSNDSTATE_STOPPING at + * this point. + */ static void WalSndLastCycleHandler(SIGNAL_ARGS) { @@ -2605,7 +2660,7 @@ WalSndLastCycleHandler(SIGNAL_ARGS) if (!replication_active) kill(MyProcPid, SIGTERM); - walsender_ready_to_stop = true; + got_SIGINT = true; SetLatch(MyLatch); errno = save_errno; @@ -2618,14 +2673,14 @@ WalSndSignals(void) /* Set up signal handlers */ pqsignal(SIGHUP, WalSndSigHupHandler); /* set flag to read config * file */ - pqsignal(SIGINT, SIG_IGN); /* not used */ + pqsignal(SIGINT, WalSndLastCycleHandler); /* request a last cycle and + * shutdown */ pqsignal(SIGTERM, die); /* request shutdown */ pqsignal(SIGQUIT, quickdie); /* hard crash time */ InitializeTimeouts(); /* establishes SIGALRM handler */ pqsignal(SIGPIPE, SIG_IGN); pqsignal(SIGUSR1, WalSndXLogSendHandler); /* request WAL sending */ - pqsignal(SIGUSR2, WalSndLastCycleHandler); /* request a last cycle and - * shutdown */ + pqsignal(SIGUSR2, WalSndSwitchStopping); /* switch to stopping state */ /* Reset some signals that are accepted by postmaster but not here */ pqsignal(SIGCHLD, SIG_DFL); @@ -2703,6 +2758,50 @@ WalSndWakeup(void) } } +/* + * Wait that all the WAL senders have reached the stopping state. This is + * used by the checkpointer to control when shutdown checkpoints can + * safely begin. + */ +void +WalSndWaitStopping(void) +{ + for (;;) + { + int i; + bool all_stopped = true; + + for (i = 0; i < max_wal_senders; i++) + { + WalSndState state; + WalSnd *walsnd = &WalSndCtl->walsnds[i]; + + SpinLockAcquire(&walsnd->mutex); + + if (walsnd->pid == 0) + { + SpinLockRelease(&walsnd->mutex); + continue; + } + + state = walsnd->state; + SpinLockRelease(&walsnd->mutex); + + if (state != WALSNDSTATE_STOPPING) + { + all_stopped = false; + break; + } + } + + /* safe to leave if confirmation is done for all WAL senders */ + if (all_stopped) + return; + + pg_usleep(10000L); /* wait for 10 msec */ + } +} + /* Set state for current walsender (only called in walsender) */ void WalSndSetState(WalSndState state) @@ -2737,6 +2836,8 @@ WalSndGetStateString(WalSndState state) return "catchup"; case WALSNDSTATE_STREAMING: return "streaming"; + case WALSNDSTATE_STOPPING: + return "stopping"; } return "UNKNOWN"; } diff --git a/src/include/replication/walsender.h b/src/include/replication/walsender.h index cb3f6bd21f..71857bebf0 100644 --- a/src/include/replication/walsender.h +++ b/src/include/replication/walsender.h @@ -34,6 +34,7 @@ extern void WalSndSignals(void); extern Size WalSndShmemSize(void); extern void WalSndShmemInit(void); extern void WalSndWakeup(void); +extern void WalSndWaitStopping(void); extern void WalSndRqstFileReload(void); extern Datum pg_stat_get_wal_senders(PG_FUNCTION_ARGS); diff --git a/src/include/replication/walsender_private.h b/src/include/replication/walsender_private.h index 6dae480285..ab541da916 100644 --- a/src/include/replication/walsender_private.h +++ b/src/include/replication/walsender_private.h @@ -24,7 +24,8 @@ typedef enum WalSndState WALSNDSTATE_STARTUP = 0, WALSNDSTATE_BACKUP, WALSNDSTATE_CATCHUP, - WALSNDSTATE_STREAMING + WALSNDSTATE_STREAMING, + WALSNDSTATE_STOPPING } WalSndState; /* -- 2.13.0