From 6ccc387d4825ed265a8f82629d27ed7f340a27e7 Mon Sep 17 00:00:00 2001 From: Nisha Moond Date: Mon, 3 Feb 2025 15:20:40 +0530 Subject: [PATCH v70 1/2] Introduce inactive_timeout based replication slot invalidation Till now, postgres has the ability to invalidate inactive replication slots based on the amount of WAL (set via max_slot_wal_keep_size GUC) that will be needed for the slots in case they become active. However, choosing a default value for this GUC is a bit tricky. Because the amount of WAL a database generates, and the allocated storage per instance will vary greatly in production, making it difficult to pin down a one-size-fits-all value. It is often easy for users to set a timeout of say 1 or 2 or n days, after which all the inactive slots get invalidated. This commit introduces a GUC named idle_replication_slot_timeout. When set, postgres invalidates slots (during non-shutdown checkpoints) that are idle for longer than this amount of time. Note that the idle timeout invalidation mechanism is not applicable for slots that do not reserve WAL or for slots on the standby server that are being synced from the primary server (i.e., standby slots having 'synced' field 'true'). Synced slots are always considered to be inactive because they don't perform logical decoding to produce changes. --- doc/src/sgml/config.sgml | 42 +++++ doc/src/sgml/logical-replication.sgml | 5 + doc/src/sgml/system-views.sgml | 7 + src/backend/replication/slot.c | 168 ++++++++++++++++-- src/backend/utils/adt/timestamp.c | 18 ++ src/backend/utils/misc/guc_tables.c | 12 ++ src/backend/utils/misc/postgresql.conf.sample | 1 + src/bin/pg_basebackup/pg_createsubscriber.c | 4 + src/bin/pg_upgrade/server.c | 7 + src/include/replication/slot.h | 3 + src/include/utils/guc_hooks.h | 2 + src/include/utils/timestamp.h | 3 + 12 files changed, 257 insertions(+), 15 deletions(-) diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml index a782f10998..11fb18b458 100644 --- a/doc/src/sgml/config.sgml +++ b/doc/src/sgml/config.sgml @@ -4423,6 +4423,48 @@ restore_command = 'copy "C:\\server\\archivedir\\%f" "%p"' # Windows + + idle_replication_slot_timeout (integer) + + idle_replication_slot_timeout configuration parameter + + + + + Invalidate replication slots that have remained idle longer than this + duration. If this value is specified without units, it is taken as + minutes. A value of zero (the default) disables the idle timeout + invalidation mechanism. This parameter can only be set in the + postgresql.conf file or on the server command + line. + + + + Slot invalidation due to idle timeout occurs during checkpoint. + Because checkpoints happen at checkpoint_timeout + intervals, there can be some lag between when the + idle_replication_slot_timeout was exceeded and when + the slot invalidation is triggered at the next checkpoint. + To avoid such lags, users can force a checkpoint to promptly invalidate + inactive slots. The duration of slot inactivity is calculated using the + slot's pg_replication_slots.inactive_since + value. + + + + Note that the idle timeout invalidation mechanism is not applicable + for slots that do not reserve WAL or for slots on the standby server + that are being synced from the primary server (i.e., standby slots + having pg_replication_slots.synced + value true). Synced slots are always considered to + be inactive because they don't perform logical decoding to produce + changes. Slots that appear idle due to a disrupted connection between + the publisher and subscriber are also excluded, as they are managed by + wal_sender_timeout. + + + + wal_sender_timeout (integer) diff --git a/doc/src/sgml/logical-replication.sgml b/doc/src/sgml/logical-replication.sgml index 613abcd28b..3d18e507bb 100644 --- a/doc/src/sgml/logical-replication.sgml +++ b/doc/src/sgml/logical-replication.sgml @@ -2390,6 +2390,11 @@ CONTEXT: processing remote data for replication origin "pg_16395" during "INSER plus some reserve for table synchronization. + + Logical replication slots are also affected by + idle_replication_slot_timeout. + + max_wal_senders should be set to at least the same as diff --git a/doc/src/sgml/system-views.sgml b/doc/src/sgml/system-views.sgml index be81c2b51d..f58b9406e4 100644 --- a/doc/src/sgml/system-views.sgml +++ b/doc/src/sgml/system-views.sgml @@ -2621,6 +2621,13 @@ SELECT * FROM pg_locks pl LEFT JOIN pg_prepared_xacts ppx perform logical decoding. It is set only for logical slots. + + + idle_timeout means that the slot has remained + idle longer than the configured + duration. + + diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index fe5acd8b1f..29749ce917 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -107,10 +107,11 @@ const char *const SlotInvalidationCauses[] = { [RS_INVAL_WAL_REMOVED] = "wal_removed", [RS_INVAL_HORIZON] = "rows_removed", [RS_INVAL_WAL_LEVEL] = "wal_level_insufficient", + [RS_INVAL_IDLE_TIMEOUT] = "idle_timeout", }; /* Maximum number of invalidation causes */ -#define RS_INVAL_MAX_CAUSES RS_INVAL_WAL_LEVEL +#define RS_INVAL_MAX_CAUSES RS_INVAL_IDLE_TIMEOUT StaticAssertDecl(lengthof(SlotInvalidationCauses) == (RS_INVAL_MAX_CAUSES + 1), "array length mismatch"); @@ -141,6 +142,12 @@ ReplicationSlot *MyReplicationSlot = NULL; int max_replication_slots = 10; /* the maximum number of replication * slots */ +/* + * Invalidate replication slots that have remained idle longer than this + * duration; '0' disables it. + */ +int idle_replication_slot_timeout_mins = 0; + /* * This GUC lists streaming replication standby server slot names that * logical WAL sender processes will wait for. @@ -1512,12 +1519,18 @@ ReportSlotInvalidation(ReplicationSlotInvalidationCause cause, NameData slotname, XLogRecPtr restart_lsn, XLogRecPtr oldestLSN, - TransactionId snapshotConflictHorizon) + TransactionId snapshotConflictHorizon, + TimestampTz inactive_since) { + int minutes = 0; + int secs = 0; + long elapsed_secs = 0; + TimestampTz now = 0; StringInfoData err_detail; - bool hint = false; + StringInfoData err_hint; initStringInfo(&err_detail); + initStringInfo(&err_hint); switch (cause) { @@ -1525,13 +1538,15 @@ ReportSlotInvalidation(ReplicationSlotInvalidationCause cause, { unsigned long long ex = oldestLSN - restart_lsn; - hint = true; appendStringInfo(&err_detail, ngettext("The slot's restart_lsn %X/%X exceeds the limit by %llu byte.", "The slot's restart_lsn %X/%X exceeds the limit by %llu bytes.", ex), LSN_FORMAT_ARGS(restart_lsn), ex); + /* translator: %s is a GUC variable name */ + appendStringInfo(&err_hint, _("You might need to increase \"%s\"."), + "max_slot_wal_keep_size"); break; } case RS_INVAL_HORIZON: @@ -1542,6 +1557,24 @@ ReportSlotInvalidation(ReplicationSlotInvalidationCause cause, case RS_INVAL_WAL_LEVEL: appendStringInfoString(&err_detail, _("Logical decoding on standby requires \"wal_level\" >= \"logical\" on the primary server.")); break; + + case RS_INVAL_IDLE_TIMEOUT: + Assert(inactive_since > 0); + + /* Calculate the idle time duration of the slot */ + now = GetCurrentTimestamp(); + elapsed_secs = (now - inactive_since) / USECS_PER_SEC; + minutes = elapsed_secs / SECS_PER_MINUTE; + secs = elapsed_secs % SECS_PER_MINUTE; + + /* translator: %s is a GUC variable name */ + appendStringInfo(&err_detail, _("The slot's idle time of %d minutes and %d seconds exceeds the configured \"%s\" duration."), + minutes, secs, "idle_replication_slot_timeout"); + /* translator: %s is a GUC variable name */ + appendStringInfo(&err_hint, _("You might need to increase \"%s\"."), + "idle_replication_slot_timeout"); + break; + case RS_INVAL_NONE: pg_unreachable(); } @@ -1553,9 +1586,31 @@ ReportSlotInvalidation(ReplicationSlotInvalidationCause cause, errmsg("invalidating obsolete replication slot \"%s\"", NameStr(slotname)), errdetail_internal("%s", err_detail.data), - hint ? errhint("You might need to increase \"%s\".", "max_slot_wal_keep_size") : 0); + err_hint.len ? errhint("%s", err_hint.data) : 0); pfree(err_detail.data); + pfree(err_hint.data); +} + +/* + * Can we invalidate an idle replication slot? + * + * Idle timeout invalidation is allowed only when: + * + * 1. Idle timeout is set + * 2. Slot has WAL reserved + * 3. Slot is inactive + * 4. The slot is not being synced from the primary while the server + * is in recovery. As synced slots are always considered to be inactive + * because they don't perform logical decoding to produce changes. + */ +static inline bool +CanInvalidateIdleSlot(ReplicationSlot *s) +{ + return (idle_replication_slot_timeout_mins != 0 && + !XLogRecPtrIsInvalid(s->data.restart_lsn) && + s->inactive_since > 0 && + !(RecoveryInProgress() && s->data.synced)); } /* @@ -1585,6 +1640,7 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, TransactionId initial_catalog_effective_xmin = InvalidTransactionId; XLogRecPtr initial_restart_lsn = InvalidXLogRecPtr; ReplicationSlotInvalidationCause invalidation_cause_prev PG_USED_FOR_ASSERTS_ONLY = RS_INVAL_NONE; + TimestampTz inactive_since = 0; for (;;) { @@ -1592,6 +1648,7 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, NameData slotname; int active_pid = 0; ReplicationSlotInvalidationCause invalidation_cause = RS_INVAL_NONE; + TimestampTz now = 0; Assert(LWLockHeldByMeInMode(ReplicationSlotControlLock, LW_SHARED)); @@ -1602,6 +1659,15 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, break; } + if (cause == RS_INVAL_IDLE_TIMEOUT) + { + /* + * Assign the current time here to reduce system call overhead + * while holding the spinlock in subsequent code. + */ + now = GetCurrentTimestamp(); + } + /* * Check if the slot needs to be invalidated. If it needs to be * invalidated, and is not currently acquired, acquire it and mark it @@ -1655,6 +1721,21 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, if (SlotIsLogical(s)) invalidation_cause = cause; break; + case RS_INVAL_IDLE_TIMEOUT: + Assert(now > 0); + + /* + * Check if the slot needs to be invalidated due to + * idle_replication_slot_timeout GUC. + */ + if (CanInvalidateIdleSlot(s) && + TimestampDifferenceExceedsSeconds(s->inactive_since, now, + idle_replication_slot_timeout_mins * SECS_PER_MINUTE)) + { + invalidation_cause = cause; + inactive_since = s->inactive_since; + } + break; case RS_INVAL_NONE: pg_unreachable(); } @@ -1705,9 +1786,10 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, /* * The logical replication slots shouldn't be invalidated as GUC - * max_slot_wal_keep_size is set to -1 during the binary upgrade. See - * check_old_cluster_for_valid_slots() where we ensure that no - * invalidated before the upgrade. + * max_slot_wal_keep_size is set to -1 and + * idle_replication_slot_timeout is set to 0 during the binary + * upgrade. See check_old_cluster_for_valid_slots() where we ensure + * that no invalidated before the upgrade. */ Assert(!(*invalidated && SlotIsLogical(s) && IsBinaryUpgrade)); @@ -1739,7 +1821,8 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, { ReportSlotInvalidation(invalidation_cause, true, active_pid, slotname, restart_lsn, - oldestLSN, snapshotConflictHorizon); + oldestLSN, snapshotConflictHorizon, + inactive_since); if (MyBackendType == B_STARTUP) (void) SendProcSignal(active_pid, @@ -1785,7 +1868,8 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, ReportSlotInvalidation(invalidation_cause, false, active_pid, slotname, restart_lsn, - oldestLSN, snapshotConflictHorizon); + oldestLSN, snapshotConflictHorizon, + inactive_since); /* done with this slot for now */ break; @@ -1800,14 +1884,16 @@ InvalidatePossiblyObsoleteSlot(ReplicationSlotInvalidationCause cause, /* * Invalidate slots that require resources about to be removed. * - * Returns true when any slot have got invalidated. + * Returns true if there are any invalidated slots. * - * Whether a slot needs to be invalidated depends on the cause. A slot is - * removed if it: + * Whether a slot needs to be invalidated depends on the invalidation cause. + * A slot is invalidated if it: * - RS_INVAL_WAL_REMOVED: requires a LSN older than the given segment * - RS_INVAL_HORIZON: requires a snapshot <= the given horizon in the given * db; dboid may be InvalidOid for shared relations - * - RS_INVAL_WAL_LEVEL: is logical + * - RS_INVAL_WAL_LEVEL: is logical and wal_level is insufficient + * - RS_INVAL_IDLE_TIMEOUT: has been idle longer than the configured + * "idle_replication_slot_timeout" duration. * * NB - this runs as part of checkpoint, so avoid raising errors if possible. */ @@ -1860,7 +1946,8 @@ restart: } /* - * Flush all replication slots to disk. + * Flush all replication slots to disk. Also, invalidate obsolete slots during + * non-shutdown checkpoint. * * It is convenient to flush dirty replication slots at the time of checkpoint. * Additionally, in case of a shutdown checkpoint, we also identify the slots @@ -1918,6 +2005,38 @@ CheckPointReplicationSlots(bool is_shutdown) SaveSlotToPath(s, path, LOG); } LWLockRelease(ReplicationSlotAllocationLock); + + if (!is_shutdown) + { + elog(DEBUG1, "performing replication slot invalidation checks"); + + /* + * NB: We will make another pass over replication slots for + * invalidation checks to keep the code simple. Testing shows that + * there is no noticeable overhead (when compared with wal_removed + * invalidation) even if we were to do idle_timeout invalidation of + * thousands of replication slots here. If it is ever proven that this + * assumption is wrong, we will have to perform the invalidation + * checks in the above for loop with the following changes: + * + * - Acquire ControlLock lock once before the loop. + * + * - Call InvalidatePossiblyObsoleteSlot for each slot. + * + * - Handle the cases in which ControlLock gets released just like + * InvalidateObsoleteReplicationSlots does. + * + * - Avoid saving slot info to disk two times for each invalidated + * slot. + * + * XXX: Should we move idle_timeout invalidation check closer to + * wal_removed in CreateCheckPoint and CreateRestartPoint? + */ + InvalidateObsoleteReplicationSlots(RS_INVAL_IDLE_TIMEOUT, + 0, + InvalidOid, + InvalidTransactionId); + } } /* @@ -2802,3 +2921,22 @@ WaitForStandbyConfirmation(XLogRecPtr wait_for_lsn) ConditionVariableCancelSleep(); } + +/* + * GUC check_hook for idle_replication_slot_timeout + * + * The value of idle_replication_slot_timeout must be set to 0 during + * a binary upgrade. See start_postmaster() in pg_upgrade for more details. + */ +bool +check_idle_replication_slot_timeout(int *newval, void **extra, GucSource source) +{ + if (IsBinaryUpgrade && *newval != 0) + { + GUC_check_errdetail("The value of \"%s\" must be set to 0 during binary upgrade mode.", + "idle_replication_slot_timeout"); + return false; + } + + return true; +} diff --git a/src/backend/utils/adt/timestamp.c b/src/backend/utils/adt/timestamp.c index ba9bae0506..9682f9dbdc 100644 --- a/src/backend/utils/adt/timestamp.c +++ b/src/backend/utils/adt/timestamp.c @@ -1786,6 +1786,24 @@ TimestampDifferenceExceeds(TimestampTz start_time, return (diff >= msec * INT64CONST(1000)); } +/* + * Check if the difference between two timestamps is >= a given + * threshold (expressed in seconds). + */ +bool +TimestampDifferenceExceedsSeconds(TimestampTz start_time, + TimestampTz stop_time, + int threshold_sec) +{ + long secs; + int usecs; + + /* Calculate the difference in seconds */ + TimestampDifference(start_time, stop_time, &secs, &usecs); + + return (secs >= threshold_sec); +} + /* * Convert a time_t to TimestampTz. * diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c index 71448bb4fd..070d8e3c56 100644 --- a/src/backend/utils/misc/guc_tables.c +++ b/src/backend/utils/misc/guc_tables.c @@ -3048,6 +3048,18 @@ struct config_int ConfigureNamesInt[] = NULL, NULL, NULL }, + { + {"idle_replication_slot_timeout", PGC_SIGHUP, REPLICATION_SENDING, + gettext_noop("Sets the duration a replication slot can remain idle before " + "it is invalidated."), + NULL, + GUC_UNIT_MIN + }, + &idle_replication_slot_timeout_mins, + 0, 0, INT_MAX / SECS_PER_MINUTE, + check_idle_replication_slot_timeout, NULL, NULL + }, + { {"commit_delay", PGC_SUSET, WAL_SETTINGS, gettext_noop("Sets the delay in microseconds between transaction commit and " diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample index 079efa1baa..c064a07973 100644 --- a/src/backend/utils/misc/postgresql.conf.sample +++ b/src/backend/utils/misc/postgresql.conf.sample @@ -326,6 +326,7 @@ # (change requires restart) #wal_keep_size = 0 # in megabytes; 0 disables #max_slot_wal_keep_size = -1 # in megabytes; -1 disables +#idle_replication_slot_timeout = 0 # in minutes; 0 disables #wal_sender_timeout = 60s # in milliseconds; 0 disables #track_commit_timestamp = off # collect timestamp of transaction commit # (change requires restart) diff --git a/src/bin/pg_basebackup/pg_createsubscriber.c b/src/bin/pg_basebackup/pg_createsubscriber.c index faf18ccf13..f2ef8d5ccc 100644 --- a/src/bin/pg_basebackup/pg_createsubscriber.c +++ b/src/bin/pg_basebackup/pg_createsubscriber.c @@ -1438,6 +1438,10 @@ start_standby_server(const struct CreateSubscriberOptions *opt, bool restricted_ appendPQExpBuffer(pg_ctl_cmd, "\"%s\" start -D ", pg_ctl_path); appendShellString(pg_ctl_cmd, subscriber_dir); appendPQExpBuffer(pg_ctl_cmd, " -s -o \"-c sync_replication_slots=off\""); + + /* Prevent unintended slot invalidation */ + appendPQExpBuffer(pg_ctl_cmd, " -o \"-c idle_replication_slot_timeout=0\""); + if (restricted_access) { appendPQExpBuffer(pg_ctl_cmd, " -o \"-p %s\"", opt->sub_port); diff --git a/src/bin/pg_upgrade/server.c b/src/bin/pg_upgrade/server.c index de6971cde6..873e5b5117 100644 --- a/src/bin/pg_upgrade/server.c +++ b/src/bin/pg_upgrade/server.c @@ -252,6 +252,13 @@ start_postmaster(ClusterInfo *cluster, bool report_and_exit_on_error) if (GET_MAJOR_VERSION(cluster->major_version) >= 1700) appendPQExpBufferStr(&pgoptions, " -c max_slot_wal_keep_size=-1"); + /* + * Use idle_replication_slot_timeout=0 to prevent slot invalidation due to + * idle_timeout by checkpointer process during upgrade. + */ + if (GET_MAJOR_VERSION(cluster->major_version) >= 1800) + appendPQExpBufferStr(&pgoptions, " -c idle_replication_slot_timeout=0"); + /* * Use -b to disable autovacuum and logical replication launcher * (effective in PG17 or later for the latter). diff --git a/src/include/replication/slot.h b/src/include/replication/slot.h index 000c36d30d..b69ddc1fb1 100644 --- a/src/include/replication/slot.h +++ b/src/include/replication/slot.h @@ -56,6 +56,8 @@ typedef enum ReplicationSlotInvalidationCause RS_INVAL_HORIZON, /* wal_level insufficient for slot */ RS_INVAL_WAL_LEVEL, + /* idle slot timeout has occurred */ + RS_INVAL_IDLE_TIMEOUT, } ReplicationSlotInvalidationCause; extern PGDLLIMPORT const char *const SlotInvalidationCauses[]; @@ -254,6 +256,7 @@ extern PGDLLIMPORT ReplicationSlot *MyReplicationSlot; /* GUCs */ extern PGDLLIMPORT int max_replication_slots; extern PGDLLIMPORT char *synchronized_standby_slots; +extern PGDLLIMPORT int idle_replication_slot_timeout_mins; /* shmem initialization functions */ extern Size ReplicationSlotsShmemSize(void); diff --git a/src/include/utils/guc_hooks.h b/src/include/utils/guc_hooks.h index 87999218d6..951451a976 100644 --- a/src/include/utils/guc_hooks.h +++ b/src/include/utils/guc_hooks.h @@ -174,5 +174,7 @@ extern void assign_wal_sync_method(int new_wal_sync_method, void *extra); extern bool check_synchronized_standby_slots(char **newval, void **extra, GucSource source); extern void assign_synchronized_standby_slots(const char *newval, void *extra); +extern bool check_idle_replication_slot_timeout(int *newval, void **extra, + GucSource source); #endif /* GUC_HOOKS_H */ diff --git a/src/include/utils/timestamp.h b/src/include/utils/timestamp.h index d26f023fb8..9963bddc0e 100644 --- a/src/include/utils/timestamp.h +++ b/src/include/utils/timestamp.h @@ -107,6 +107,9 @@ extern long TimestampDifferenceMilliseconds(TimestampTz start_time, extern bool TimestampDifferenceExceeds(TimestampTz start_time, TimestampTz stop_time, int msec); +extern bool TimestampDifferenceExceedsSeconds(TimestampTz start_time, + TimestampTz stop_time, + int threshold_sec); extern TimestampTz time_t_to_timestamptz(pg_time_t tm); extern pg_time_t timestamptz_to_time_t(TimestampTz t); -- 2.34.1