Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commit36da3cf

Browse files
Allow time delayed standbys and recovery
Set min_recovery_apply_delay to force a delay in recovery apply for commit andrestore point WAL records. Other records are replayed immediately. Delay ismeasured between WAL record time and local standby time.Robert Haas, Fabrízio de Royes Mello and Simon RiggsDetailed review by Mitsumasa Kondo
1 parent841a654 commit36da3cf

File tree

3 files changed

+174
-4
lines changed

3 files changed

+174
-4
lines changed

‎doc/src/sgml/recovery-config.sgml‎

Lines changed: 50 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -142,6 +142,56 @@ restore_command = 'copy "C:\\server\\archivedir\\%f" "%p"' # Windows
142142
</listitem>
143143
</varlistentry>
144144

145+
<varlistentry id="min-recovery-apply-delay" xreflabel="min_recovery_apply_delay">
146+
<term><varname>min_recovery_apply_delay</varname> (<type>integer</type>)</term>
147+
<indexterm>
148+
<primary><varname>min_recovery_apply_delay</> recovery parameter</primary>
149+
</indexterm>
150+
<listitem>
151+
<para>
152+
By default, a standby server keeps restoring WAL records from the
153+
primary as soon as possible. It may be useful to have a time-delayed
154+
copy of the data, offering various options to correct data loss errors.
155+
This paramater allows you to delay recovery by a fixed period of time,
156+
specified in milliseconds if no unit is specified. For example, if
157+
you set this parameter to <literal>5min</literal>, the standby will
158+
replay each transaction commit only when the system time on the standby
159+
is at least five minutes past the commit time reported by the master.
160+
</para>
161+
<para>
162+
It is possible that the replication delay between servers exceeds the
163+
value of this parameter, in which case no delay is added.
164+
Note that the delay is calculated between the WAL timestamp as written
165+
on master and the time on the current standby. Delays
166+
in transfer because of networks or cascading replication configurations
167+
may reduce the actual wait time significantly. If the system
168+
clocks on master and standby are not synchronised, this may lead to
169+
recovery applying records earlier than expected but is not a major issue
170+
because the useful settings of the parameter are much larger than
171+
typical time deviation between the servers. Be careful to allow for
172+
different timezone settings on master and standby.
173+
</para>
174+
<para>
175+
The delay occurs only on WAL records for COMMIT and Restore Points.
176+
Other records may be replayed earlier than the specified delay, which
177+
is not an issue for MVCC though may potentially increase the number
178+
of recovery conflicts generated.
179+
</para>
180+
<para>
181+
The delay occurs until the standby is promoted or triggered. After that
182+
the standby will end recovery without further waiting.
183+
</para>
184+
<para>
185+
This parameter is intended for use with streaming replication deployments,
186+
however, if the parameter is specified it will be honoured in all cases.
187+
Synchronous replication is not affected by this setting because there is
188+
not yet any setting to request synchronous apply of transaction commits.
189+
<varname>hot_standby_feedback</> will be delayed by use of this feature
190+
which could lead to bloat on the master; use both together with care.
191+
</para>
192+
</listitem>
193+
</varlistentry>
194+
145195
</variablelist>
146196

147197
</sect1>

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

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -123,6 +123,15 @@
123123
#
124124
#trigger_file = ''
125125
#
126+
# By default, a standby server keeps restoring XLOG records from the
127+
# primary as soon as possible. If you want to explicitly delay the replay of
128+
# committed transactions from the master, specify a recovery apply delay.
129+
# For example, if you set this parameter to 5min, the standby will replay
130+
# each transaction commit only when the system time on the standby is least
131+
# five minutes past the commit time reported by the master.
132+
#
133+
#min_recovery_apply_delay = 0
134+
#
126135
#---------------------------------------------------------------------------
127136
# HOT STANDBY PARAMETERS
128137
#---------------------------------------------------------------------------

‎src/backend/access/transam/xlog.c‎

Lines changed: 115 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -218,6 +218,8 @@ static bool recoveryPauseAtTarget = true;
218218
staticTransactionIdrecoveryTargetXid;
219219
staticTimestampTzrecoveryTargetTime;
220220
staticchar*recoveryTargetName;
221+
staticintmin_recovery_apply_delay=0;
222+
staticTimestampTzrecoveryDelayUntilTime;
221223

222224
/* options taken from recovery.conf for XLOG streaming */
223225
staticboolStandbyModeRequested= false;
@@ -728,8 +730,10 @@ static bool holdingAllSlots = false;
728730

729731
staticvoidreadRecoveryCommandFile(void);
730732
staticvoidexitArchiveRecovery(TimeLineIDendTLI,XLogSegNoendLogSegNo);
731-
staticboolrecoveryStopsHere(XLogRecord*record,bool*includeThis);
733+
staticboolrecoveryStopsHere(XLogRecord*record,bool*includeThis,bool*delayThis);
732734
staticvoidrecoveryPausesHere(void);
735+
staticvoidrecoveryApplyDelay(void);
736+
staticboolSetRecoveryDelayUntilTime(TimestampTzxtime);
733737
staticvoidSetLatestXTime(TimestampTzxtime);
734738
staticvoidSetCurrentChunkStartTime(TimestampTzxtime);
735739
staticvoidCheckRequiredParameterValues(void);
@@ -5476,6 +5480,19 @@ readRecoveryCommandFile(void)
54765480
(errmsg_internal("trigger_file = '%s'",
54775481
TriggerFile)));
54785482
}
5483+
elseif (strcmp(item->name,"min_recovery_apply_delay")==0)
5484+
{
5485+
constchar*hintmsg;
5486+
5487+
if (!parse_int(item->value,&min_recovery_apply_delay,GUC_UNIT_MS,
5488+
&hintmsg))
5489+
ereport(ERROR,
5490+
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
5491+
errmsg("parameter \"%s\" requires a temporal value","min_recovery_apply_delay"),
5492+
hintmsg ?errhint("%s",_(hintmsg)) :0));
5493+
ereport(DEBUG2,
5494+
(errmsg("min_recovery_apply_delay = '%s'",item->value)));
5495+
}
54795496
else
54805497
ereport(FATAL,
54815498
(errmsg("unrecognized recovery parameter \"%s\"",
@@ -5625,10 +5642,11 @@ exitArchiveRecovery(TimeLineID endTLI, XLogSegNo endLogSegNo)
56255642
* We also track the timestamp of the latest applied COMMIT/ABORT
56265643
* record in XLogCtl->recoveryLastXTime, for logging purposes.
56275644
* Also, some information is saved in recoveryStopXid et al for use in
5628-
* annotating the new timeline's history file.
5645+
* annotating the new timeline's history file; and recoveryDelayUntilTime
5646+
* is updated, for time-delayed standbys.
56295647
*/
56305648
staticbool
5631-
recoveryStopsHere(XLogRecord*record,bool*includeThis)
5649+
recoveryStopsHere(XLogRecord*record,bool*includeThis,bool*delayThis)
56325650
{
56335651
boolstopsHere;
56345652
uint8record_info;
@@ -5645,20 +5663,31 @@ recoveryStopsHere(XLogRecord *record, bool *includeThis)
56455663

56465664
recordXactCommitData= (xl_xact_commit_compact*)XLogRecGetData(record);
56475665
recordXtime=recordXactCommitData->xact_time;
5666+
5667+
*delayThis=SetRecoveryDelayUntilTime(recordXactCommitData->xact_time);
56485668
}
56495669
elseif (record->xl_rmid==RM_XACT_ID&&record_info==XLOG_XACT_COMMIT)
56505670
{
56515671
xl_xact_commit*recordXactCommitData;
56525672

56535673
recordXactCommitData= (xl_xact_commit*)XLogRecGetData(record);
56545674
recordXtime=recordXactCommitData->xact_time;
5675+
5676+
*delayThis=SetRecoveryDelayUntilTime(recordXactCommitData->xact_time);
56555677
}
56565678
elseif (record->xl_rmid==RM_XACT_ID&&record_info==XLOG_XACT_ABORT)
56575679
{
56585680
xl_xact_abort*recordXactAbortData;
56595681

56605682
recordXactAbortData= (xl_xact_abort*)XLogRecGetData(record);
56615683
recordXtime=recordXactAbortData->xact_time;
5684+
5685+
/*
5686+
* We deliberately choose not to delay aborts since they have no
5687+
* effect on MVCC. We already allow replay of records that don't
5688+
* have a timestamp, so there is already opportunity for issues
5689+
* caused by early conflicts on standbys.
5690+
*/
56625691
}
56635692
elseif (record->xl_rmid==RM_XLOG_ID&&record_info==XLOG_RESTORE_POINT)
56645693
{
@@ -5667,6 +5696,8 @@ recoveryStopsHere(XLogRecord *record, bool *includeThis)
56675696
recordRestorePointData= (xl_restore_point*)XLogRecGetData(record);
56685697
recordXtime=recordRestorePointData->rp_time;
56695698
strncpy(recordRPName,recordRestorePointData->rp_name,MAXFNAMELEN);
5699+
5700+
*delayThis=SetRecoveryDelayUntilTime(recordRestorePointData->rp_time);
56705701
}
56715702
else
56725703
return false;
@@ -5833,6 +5864,66 @@ SetRecoveryPause(bool recoveryPause)
58335864
SpinLockRelease(&xlogctl->info_lck);
58345865
}
58355866

5867+
staticbool
5868+
SetRecoveryDelayUntilTime(TimestampTzxtime)
5869+
{
5870+
if (min_recovery_apply_delay!=0)
5871+
{
5872+
recoveryDelayUntilTime=
5873+
TimestampTzPlusMilliseconds(xtime,min_recovery_apply_delay);
5874+
5875+
return true;
5876+
}
5877+
5878+
return false;
5879+
}
5880+
/*
5881+
* When min_recovery_apply_delay is set, we wait long enough to make sure
5882+
* certain record types are applied at least that interval behind the master.
5883+
* See recoveryStopsHere().
5884+
*
5885+
* Note that the delay is calculated between the WAL record log time and
5886+
* the current time on standby. We would prefer to keep track of when this
5887+
* standby received each WAL record, which would allow a more consistent
5888+
* approach and one not affected by time synchronisation issues, but that
5889+
* is significantly more effort and complexity for little actual gain in
5890+
* usability.
5891+
*/
5892+
staticvoid
5893+
recoveryApplyDelay(void)
5894+
{
5895+
while (true)
5896+
{
5897+
longsecs;
5898+
intmicrosecs;
5899+
5900+
ResetLatch(&XLogCtl->recoveryWakeupLatch);
5901+
5902+
/* might change the trigger file's location */
5903+
HandleStartupProcInterrupts();
5904+
5905+
if (CheckForStandbyTrigger())
5906+
break;
5907+
5908+
/*
5909+
* Wait for difference between GetCurrentTimestamp() and
5910+
* recoveryDelayUntilTime
5911+
*/
5912+
TimestampDifference(GetCurrentTimestamp(),recoveryDelayUntilTime,
5913+
&secs,&microsecs);
5914+
5915+
if (secs <=0&&microsecs <=0)
5916+
break;
5917+
5918+
elog(DEBUG2,"recovery apply delay %ld seconds, %d milliseconds",
5919+
secs,microsecs /1000);
5920+
5921+
WaitLatch(&XLogCtl->recoveryWakeupLatch,
5922+
WL_LATCH_SET |WL_TIMEOUT |WL_POSTMASTER_DEATH,
5923+
secs*1000L+microsecs /1000);
5924+
}
5925+
}
5926+
58365927
/*
58375928
* Save timestamp of latest processed commit/abort record.
58385929
*
@@ -6660,6 +6751,7 @@ StartupXLOG(void)
66606751
{
66616752
boolrecoveryContinue= true;
66626753
boolrecoveryApply= true;
6754+
boolrecoveryDelay= false;
66636755
ErrorContextCallbackerrcallback;
66646756
TimestampTzxtime;
66656757

@@ -6719,7 +6811,7 @@ StartupXLOG(void)
67196811
/*
67206812
* Have we reached our recovery target?
67216813
*/
6722-
if (recoveryStopsHere(record,&recoveryApply))
6814+
if (recoveryStopsHere(record,&recoveryApply,&recoveryDelay))
67236815
{
67246816
if (recoveryPauseAtTarget)
67256817
{
@@ -6734,6 +6826,25 @@ StartupXLOG(void)
67346826
break;
67356827
}
67366828

6829+
/*
6830+
* If we've been asked to lag the master, wait on
6831+
* latch until enough time has passed.
6832+
*/
6833+
if (recoveryDelay)
6834+
{
6835+
recoveryApplyDelay();
6836+
6837+
/*
6838+
* We test for paused recovery again here. If
6839+
* user sets delayed apply, it may be because
6840+
* they expect to pause recovery in case of
6841+
* problems, so we must test again here otherwise
6842+
* pausing during the delay-wait wouldn't work.
6843+
*/
6844+
if (xlogctl->recoveryPause)
6845+
recoveryPausesHere();
6846+
}
6847+
67376848
/* Setup error traceback support for ereport() */
67386849
errcallback.callback=rm_redo_error_callback;
67396850
errcallback.arg= (void*)record;

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp