Allow time delayed standbys and recovery

Set min_recovery_apply_delay to force a delay in recovery apply for commit and
restore point WAL records. Other records are replayed immediately. Delay is
measured between WAL record time and local standby time.

Robert Haas, Fabrízio de Royes Mello and Simon Riggs
Detailed review by Mitsumasa Kondo

Author: simonat2ndQuadrant

doc/src/sgml/recovery-config.sgml

@@ -142,6 +142,56 @@ restore_command = 'copy "C:\\server\\archivedir\\%f" "%p"'  # Windows
       </listitem>
      </varlistentry>
 
+     <varlistentry id="min-recovery-apply-delay" xreflabel="min_recovery_apply_delay">
+      <term><varname>min_recovery_apply_delay</varname> (<type>integer</type>)</term>
+      <indexterm>
+        <primary><varname>min_recovery_apply_delay</> recovery parameter</primary>
+      </indexterm>
+      <listitem>
+       <para>
+        By default, a standby server keeps restoring WAL records from the
+        primary as soon as possible. It may be useful to have a time-delayed
+        copy of the data, offering various options to correct data loss errors.
+        This paramater allows you to delay recovery by a fixed period of time,
+        specified in milliseconds if no unit is specified.  For example, if
+        you set this parameter to <literal>5min</literal>, the standby will
+        replay each transaction commit only when the system time on the standby
+        is at least five minutes past the commit time reported by the master.
+       </para>
+       <para>
+        It is possible that the replication delay between servers exceeds the
+        value of this parameter, in which case no delay is added.
+        Note that the delay is calculated between the WAL timestamp as written
+        on master and the time on the current standby. Delays
+        in transfer because of networks or cascading replication configurations
+        may reduce the actual wait time significantly. If the system
+        clocks on master and standby are not synchronised, this may lead to
+        recovery applying records earlier than expected but is not a major issue
+        because the useful settings of the parameter are much larger than
+        typical time deviation between the servers. Be careful to allow for
+        different timezone settings on master and standby.
+       </para>
+       <para>
+        The delay occurs only on WAL records for COMMIT and Restore Points.
+        Other records may be replayed earlier than the specified delay, which
+        is not an issue for MVCC though may potentially increase the number
+        of recovery conflicts generated.
+       </para>
+       <para>
+        The delay occurs until the standby is promoted or triggered. After that
+        the standby will end recovery without further waiting.
+       </para>
+       <para>
+        This parameter is intended for use with streaming replication deployments,
+        however, if the parameter is specified it will be honoured in all cases.
+        Synchronous replication is not affected by this setting because there is
+        not yet any setting to request synchronous apply of transaction commits.
+        <varname>hot_standby_feedback</> will be delayed by use of this feature
+        which could lead to bloat on the master; use both together with care.
+       </para>
+      </listitem>
+     </varlistentry>
+
     </variablelist>
 
   </sect1>

src/backend/access/transam/recovery.conf.sample

@@ -123,6 +123,15 @@
 #
 #trigger_file = ''
 #
+# By default, a standby server keeps restoring XLOG records from the
+# primary as soon as possible. If you want to explicitly delay the replay of
+# committed transactions from the master, specify a recovery apply delay.
+# For example, if you set this parameter to 5min, the standby will replay
+# each transaction commit only when the system time on the standby is least
+# five minutes past the commit time reported by the master.
+#
+#min_recovery_apply_delay = 0
+#
 #---------------------------------------------------------------------------
 # HOT STANDBY PARAMETERS
 #---------------------------------------------------------------------------

src/backend/access/transam/xlog.c

@@ -218,6 +218,8 @@ static bool recoveryPauseAtTarget = true;
 static TransactionId recoveryTargetXid;
 static TimestampTz recoveryTargetTime;
 static char *recoveryTargetName;
+static int min_recovery_apply_delay = 0;
+static TimestampTz recoveryDelayUntilTime;
 
 /* options taken from recovery.conf for XLOG streaming */
 static bool StandbyModeRequested = false;
@@ -728,8 +730,10 @@ static bool holdingAllSlots = false;
 
 static void readRecoveryCommandFile(void);
 static void exitArchiveRecovery(TimeLineID endTLI, XLogSegNo endLogSegNo);
-static bool recoveryStopsHere(XLogRecord *record, bool *includeThis);
+static bool recoveryStopsHere(XLogRecord *record, bool *includeThis, bool *delayThis);
 static void recoveryPausesHere(void);
+static void recoveryApplyDelay(void);
+static bool SetRecoveryDelayUntilTime(TimestampTz xtime);
 static void SetLatestXTime(TimestampTz xtime);
 static void SetCurrentChunkStartTime(TimestampTz xtime);
 static void CheckRequiredParameterValues(void);
@@ -5476,6 +5480,19 @@ readRecoveryCommandFile(void)
 					(errmsg_internal("trigger_file = '%s'",
 									 TriggerFile)));
 		}
+		else if (strcmp(item->name, "min_recovery_apply_delay") == 0)
+		{
+			const char *hintmsg;
+
+			if (!parse_int(item->value, &min_recovery_apply_delay, GUC_UNIT_MS,
+					&hintmsg))
+				ereport(ERROR,
+						(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+						 errmsg("parameter \"%s\" requires a temporal value", "min_recovery_apply_delay"),
+						 hintmsg ? errhint("%s", _(hintmsg)) : 0));
+			ereport(DEBUG2,
+					(errmsg("min_recovery_apply_delay = '%s'", item->value)));
+		}
 		else
 			ereport(FATAL,
 					(errmsg("unrecognized recovery parameter \"%s\"",
@@ -5625,10 +5642,11 @@ exitArchiveRecovery(TimeLineID endTLI, XLogSegNo endLogSegNo)
  * We also track the timestamp of the latest applied COMMIT/ABORT
  * record in XLogCtl->recoveryLastXTime, for logging purposes.
  * Also, some information is saved in recoveryStopXid et al for use in
- * annotating the new timeline's history file.
+ * annotating the new timeline's history file; and recoveryDelayUntilTime
+ * is updated, for time-delayed standbys.
  */
 static bool
-recoveryStopsHere(XLogRecord *record, bool *includeThis)
+recoveryStopsHere(XLogRecord *record, bool *includeThis, bool *delayThis)
 {
 	bool		stopsHere;
 	uint8		record_info;
@@ -5645,20 +5663,31 @@ recoveryStopsHere(XLogRecord *record, bool *includeThis)
 
 		recordXactCommitData = (xl_xact_commit_compact *) XLogRecGetData(record);
 		recordXtime = recordXactCommitData->xact_time;
+
+		*delayThis = SetRecoveryDelayUntilTime(recordXactCommitData->xact_time);
 	}
 	else if (record->xl_rmid == RM_XACT_ID && record_info == XLOG_XACT_COMMIT)
 	{
 		xl_xact_commit *recordXactCommitData;
 
 		recordXactCommitData = (xl_xact_commit *) XLogRecGetData(record);
 		recordXtime = recordXactCommitData->xact_time;
+
+		*delayThis = SetRecoveryDelayUntilTime(recordXactCommitData->xact_time);
 	}
 	else if (record->xl_rmid == RM_XACT_ID && record_info == XLOG_XACT_ABORT)
 	{
 		xl_xact_abort *recordXactAbortData;
 
 		recordXactAbortData = (xl_xact_abort *) XLogRecGetData(record);
 		recordXtime = recordXactAbortData->xact_time;
+
+		/*
+		 * We deliberately choose not to delay aborts since they have no
+		 * effect on MVCC. We already allow replay of records that don't
+		 * have a timestamp, so there is already opportunity for issues
+		 * caused by early conflicts on standbys.
+		 */
 	}
 	else if (record->xl_rmid == RM_XLOG_ID && record_info == XLOG_RESTORE_POINT)
 	{
@@ -5667,6 +5696,8 @@ recoveryStopsHere(XLogRecord *record, bool *includeThis)
 		recordRestorePointData = (xl_restore_point *) XLogRecGetData(record);
 		recordXtime = recordRestorePointData->rp_time;
 		strncpy(recordRPName, recordRestorePointData->rp_name, MAXFNAMELEN);
+
+		*delayThis = SetRecoveryDelayUntilTime(recordRestorePointData->rp_time);
 	}
 	else
 		return false;
@@ -5833,6 +5864,66 @@ SetRecoveryPause(bool recoveryPause)
 	SpinLockRelease(&xlogctl->info_lck);
 }
 
+static bool
+SetRecoveryDelayUntilTime(TimestampTz xtime)
+{
+	if (min_recovery_apply_delay != 0)
+	{
+		recoveryDelayUntilTime =
+			TimestampTzPlusMilliseconds(xtime, min_recovery_apply_delay);
+
+		return true;
+	}
+
+	return false;
+}
+/*
+ * When min_recovery_apply_delay is set, we wait long enough to make sure
+ * certain record types are applied at least that interval behind the master.
+ * See recoveryStopsHere().
+ *
+ * Note that the delay is calculated between the WAL record log time and
+ * the current time on standby. We would prefer to keep track of when this
+ * standby received each WAL record, which would allow a more consistent
+ * approach and one not affected by time synchronisation issues, but that
+ * is significantly more effort and complexity for little actual gain in
+ * usability.
+ */
+static void
+recoveryApplyDelay(void)
+{
+	while (true)
+	{
+		long	secs;
+		int		microsecs;
+
+		ResetLatch(&XLogCtl->recoveryWakeupLatch);
+
+		/* might change the trigger file's location */
+		HandleStartupProcInterrupts();
+
+		if (CheckForStandbyTrigger())
+			break;
+
+		/*
+		 * Wait for difference between GetCurrentTimestamp() and
+		 * recoveryDelayUntilTime
+		 */
+		TimestampDifference(GetCurrentTimestamp(), recoveryDelayUntilTime,
+							&secs, &microsecs);
+
+		if (secs <= 0 && microsecs <=0)
+			break;
+
+		elog(DEBUG2, "recovery apply delay %ld seconds, %d milliseconds",
+			secs, microsecs / 1000);
+
+		WaitLatch(&XLogCtl->recoveryWakeupLatch,
+					WL_LATCH_SET | WL_TIMEOUT | WL_POSTMASTER_DEATH,
+					secs * 1000L + microsecs / 1000);
+	}
+}
+
 /*
  * Save timestamp of latest processed commit/abort record.
  *
@@ -6660,6 +6751,7 @@ StartupXLOG(void)
 		{
 			bool		recoveryContinue = true;
 			bool		recoveryApply = true;
+			bool		recoveryDelay = false;
 			ErrorContextCallback errcallback;
 			TimestampTz xtime;
 
@@ -6719,7 +6811,7 @@ StartupXLOG(void)
 				/*
 				 * Have we reached our recovery target?
 				 */
-				if (recoveryStopsHere(record, &recoveryApply))
+				if (recoveryStopsHere(record, &recoveryApply, &recoveryDelay))
 				{
 					if (recoveryPauseAtTarget)
 					{
@@ -6734,6 +6826,25 @@ StartupXLOG(void)
 						break;
 				}
 
+				/*
+				 * If we've been asked to lag the master, wait on
+				 * latch until enough time has passed.
+				 */
+				if (recoveryDelay)
+				{
+					recoveryApplyDelay();
+
+					/*
+					 * We test for paused recovery again here. If
+					 * user sets delayed apply, it may be because
+					 * they expect to pause recovery in case of
+					 * problems, so we must test again here otherwise
+					 * pausing during the delay-wait wouldn't work.
+					 */
+					if (xlogctl->recoveryPause)
+						recoveryPausesHere();
+				}
+
 				/* Setup error traceback support for ereport() */
 				errcallback.callback = rm_redo_error_callback;
 				errcallback.arg = (void *) record;