Skip to content

Commit 88e9823

Browse files
hlinnakahlinnaka
committed
Replace checkpoint_segments with min_wal_size and max_wal_size.
Instead of having a single knob (checkpoint_segments) that both triggers checkpoints, and determines how many checkpoints to recycle, they are now separate concerns. There is still an internal variable called CheckpointSegments, which triggers checkpoints. But it no longer determines how many segments to recycle at a checkpoint. That is now auto-tuned by keeping a moving average of the distance between checkpoints (in bytes), and trying to keep that many segments in reserve. The advantage of this is that you can set max_wal_size very high, but the system won't actually consume that much space if there isn't any need for it. The min_wal_size sets a floor for that; you can effectively disable the auto-tuning behavior by setting min_wal_size equal to max_wal_size. The max_wal_size setting is now the actual target size of WAL at which a new checkpoint is triggered, instead of the distance between checkpoints. Previously, you could calculate the actual WAL usage with the formula "(2 + checkpoint_completion_target) * checkpoint_segments + 1". With this patch, you set the desired WAL usage with max_wal_size, and the system calculates the appropriate CheckpointSegments with the reverse of that formula. That's a lot more intuitive for administrators to set. Reviewed by Amit Kapila and Venkata Balaji N.
1 parent 0fec000 commit 88e9823

File tree

9 files changed

+327
-108
lines changed

9 files changed

+327
-108
lines changed

‎doc/src/sgml/config.sgml

Lines changed: 30 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -1325,7 +1325,7 @@ include_dir 'conf.d'
13251325
40% of RAM to <varname>shared_buffers</varname> will work better than a
13261326
smaller amount. Larger settings for <varname>shared_buffers</varname>
13271327
usually require a corresponding increase in
1328-
<varname>checkpoint_segments</varname>, in order to spread out the
1328+
<varname>max_wal_size</varname>, in order to spread out the
13291329
process of writing large quantities of new or changed data over a
13301330
longer period of time.
13311331
</para>
@@ -2394,18 +2394,20 @@ include_dir 'conf.d'
23942394
<title>Checkpoints</title>
23952395

23962396
<variablelist>
2397-
<varlistentry id="guc-checkpoint-segments" xreflabel="checkpoint_segments">
2398-
<term><varname>checkpoint_segments</varname> (<type>integer</type>)
2397+
<varlistentry id="guc-max-wal-size" xreflabel="max_wal_size">
2398+
<term><varname>max_wal_size</varname> (<type>integer</type>)</term>
23992399
<indexterm>
2400-
<primary><varname>checkpoint_segments</> configuration parameter</primary>
2400+
<primary><varname>max_wal_size</> configuration parameter</primary>
24012401
</indexterm>
2402-
</term>
24032402
<listitem>
24042403
<para>
2405-
Maximum number of log file segments between automatic WAL
2406-
checkpoints (each segment is normally 16 megabytes). The default
2407-
is three segments. Increasing this parameter can increase the
2408-
amount of time needed for crash recovery.
2404+
Maximum size to let the WAL grow to between automatic WAL
2405+
checkpoints. This is a soft limit; WAL size can exceed
2406+
<varname>max_wal_size</> under special circumstances, like
2407+
under heavy load, a failing <varname>archive_command</>, or a high
2408+
<varname>wal_keep_segments</> setting. The default is 128 MB.
2409+
Increasing this parameter can increase the amount of time needed for
2410+
crash recovery.
24092411
This parameter can only be set in the <filename>postgresql.conf</>
24102412
file or on the server command line.
24112413
</para>
@@ -2458,7 +2460,7 @@ include_dir 'conf.d'
24582460
Write a message to the server log if checkpoints caused by
24592461
the filling of checkpoint segment files happen closer together
24602462
than this many seconds (which suggests that
2461-
<varname>checkpoint_segments</> ought to be raised). The default is
2463+
<varname>max_wal_size</> ought to be raised). The default is
24622464
30 seconds (<literal>30s</>). Zero disables the warning.
24632465
No warnings will be generated if <varname>checkpoint_timeout</varname>
24642466
is less than <varname>checkpoint_warning</varname>.
@@ -2468,6 +2470,24 @@ include_dir 'conf.d'
24682470
</listitem>
24692471
</varlistentry>
24702472

2473+
<varlistentry id="guc-min-wal-size" xreflabel="min_wal_size">
2474+
<term><varname>min_wal_size</varname> (<type>integer</type>)</term>
2475+
<indexterm>
2476+
<primary><varname>min_wal_size</> configuration parameter</primary>
2477+
</indexterm>
2478+
<listitem>
2479+
<para>
2480+
As long as WAL disk usage stays below this setting, old WAL files are
2481+
always recycled for future use at a checkpoint, rather than removed.
2482+
This can be used to ensure that enough WAL space is reserved to
2483+
handle spikes in WAL usage, for example when running large batch
2484+
jobs. The default is 80 MB.
2485+
This parameter can only be set in the <filename>postgresql.conf</>
2486+
file or on the server command line.
2487+
</para>
2488+
</listitem>
2489+
</varlistentry>
2490+
24712491
</variablelist>
24722492
</sect2>
24732493
<sect2 id="runtime-config-wal-archiving">

‎doc/src/sgml/perform.sgml

Lines changed: 8 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -1328,19 +1328,19 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
13281328
</para>
13291329
</sect2>
13301330

1331-
<sect2 id="populate-checkpoint-segments">
1332-
<title>Increase <varname>checkpoint_segments</varname></title>
1331+
<sect2 id="populate-max-wal-size">
1332+
<title>Increase <varname>max_wal_size</varname></title>
13331333

13341334
<para>
1335-
Temporarily increasing the <xref
1336-
linkend="guc-checkpoint-segments"> configuration variable can also
1335+
Temporarily increasing the <xref linkend="guc-max-wal-size">
1336+
configuration variable can also
13371337
make large data loads faster. This is because loading a large
13381338
amount of data into <productname>PostgreSQL</productname> will
13391339
cause checkpoints to occur more often than the normal checkpoint
13401340
frequency (specified by the <varname>checkpoint_timeout</varname>
13411341
configuration variable). Whenever a checkpoint occurs, all dirty
13421342
pages must be flushed to disk. By increasing
1343-
<varname>checkpoint_segments</varname> temporarily during bulk
1343+
<varname>max_wal_size</varname> temporarily during bulk
13441344
data loads, the number of checkpoints that are required can be
13451345
reduced.
13461346
</para>
@@ -1445,7 +1445,7 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
14451445
<para>
14461446
Set appropriate (i.e., larger than normal) values for
14471447
<varname>maintenance_work_mem</varname> and
1448-
<varname>checkpoint_segments</varname>.
1448+
<varname>max_wal_size</varname>.
14491449
</para>
14501450
</listitem>
14511451
<listitem>
@@ -1512,7 +1512,7 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
15121512

15131513
So when loading a data-only dump, it is up to you to drop and recreate
15141514
indexes and foreign keys if you wish to use those techniques.
1515-
It's still useful to increase <varname>checkpoint_segments</varname>
1515+
It's still useful to increase <varname>max_wal_size</varname>
15161516
while loading the data, but don't bother increasing
15171517
<varname>maintenance_work_mem</varname>; rather, you'd do that while
15181518
manually recreating indexes and foreign keys afterwards.
@@ -1577,7 +1577,7 @@ SELECT * FROM x, y, a, b, c WHERE something AND somethingelse;
15771577

15781578
<listitem>
15791579
<para>
1580-
Increase <xref linkend="guc-checkpoint-segments"> and <xref
1580+
Increase <xref linkend="guc-max-wal-size"> and <xref
15811581
linkend="guc-checkpoint-timeout"> ; this reduces the frequency
15821582
of checkpoints, but increases the storage requirements of
15831583
<filename>/pg_xlog</>.

‎doc/src/sgml/wal.sgml

Lines changed: 43 additions & 26 deletions
Original file line numberDiff line numberDiff line change
@@ -472,9 +472,10 @@
472472
<para>
473473
The server's checkpointer process automatically performs
474474
a checkpoint every so often. A checkpoint is begun every <xref
475-
linkend="guc-checkpoint-segments"> log segments, or every <xref
476-
linkend="guc-checkpoint-timeout"> seconds, whichever comes first.
477-
The default settings are 3 segments and 300 seconds (5 minutes), respectively.
475+
linkend="guc-checkpoint-timeout"> seconds, or if
476+
<xref linkend="guc-max-wal-size"> is about to be exceeded,
477+
whichever comes first.
478+
The default settings are 5 minutes and 128 MB, respectively.
478479
If no WAL has been written since the previous checkpoint, new checkpoints
479480
will be skipped even if <varname>checkpoint_timeout</> has passed.
480481
(If WAL archiving is being used and you want to put a lower limit on how
@@ -486,8 +487,8 @@
486487
</para>
487488

488489
<para>
489-
Reducing <varname>checkpoint_segments</varname> and/or
490-
<varname>checkpoint_timeout</varname> causes checkpoints to occur
490+
Reducing <varname>checkpoint_timeout</varname> and/or
491+
<varname>max_wal_size</varname> causes checkpoints to occur
491492
more often. This allows faster after-crash recovery, since less work
492493
will need to be redone. However, one must balance this against the
493494
increased cost of flushing dirty data pages more often. If
@@ -510,11 +511,11 @@
510511
parameter. If checkpoints happen closer together than
511512
<varname>checkpoint_warning</> seconds,
512513
a message will be output to the server log recommending increasing
513-
<varname>checkpoint_segments</varname>. Occasional appearance of such
514+
<varname>max_wal_size</varname>. Occasional appearance of such
514515
a message is not cause for alarm, but if it appears often then the
515516
checkpoint control parameters should be increased. Bulk operations such
516517
as large <command>COPY</> transfers might cause a number of such warnings
517-
to appear if you have not set <varname>checkpoint_segments</> high
518+
to appear if you have not set <varname>max_wal_size</> high
518519
enough.
519520
</para>
520521

@@ -525,10 +526,10 @@
525526
<xref linkend="guc-checkpoint-completion-target">, which is
526527
given as a fraction of the checkpoint interval.
527528
The I/O rate is adjusted so that the checkpoint finishes when the
528-
given fraction of <varname>checkpoint_segments</varname> WAL segments
529-
have been consumed since checkpoint start, or the given fraction of
530-
<varname>checkpoint_timeout</varname> seconds have elapsed,
531-
whichever is sooner. With the default value of 0.5,
529+
given fraction of
530+
<varname>checkpoint_timeout</varname> seconds have elapsed, or before
531+
<varname>max_wal_size</varname> is exceeded, whichever is sooner.
532+
With the default value of 0.5,
532533
<productname>PostgreSQL</> can be expected to complete each checkpoint
533534
in about half the time before the next checkpoint starts. On a system
534535
that's very close to maximum I/O throughput during normal operation,
@@ -545,18 +546,35 @@
545546
</para>
546547

547548
<para>
548-
There will always be at least one WAL segment file, and will normally
549-
not be more than (2 + <varname>checkpoint_completion_target</varname>) * <varname>checkpoint_segments</varname> + 1
550-
or <varname>checkpoint_segments</> + <xref linkend="guc-wal-keep-segments"> + 1
551-
files. Each segment file is normally 16 MB (though this size can be
552-
altered when building the server). You can use this to estimate space
553-
requirements for <acronym>WAL</acronym>.
554-
Ordinarily, when old log segment files are no longer needed, they
555-
are recycled (that is, renamed to become future segments in the numbered
556-
sequence). If, due to a short-term peak of log output rate, there
557-
are more than 3 * <varname>checkpoint_segments</varname> + 1
558-
segment files, the unneeded segment files will be deleted instead
559-
of recycled until the system gets back under this limit.
549+
The number of WAL segment files in <filename>pg_xlog</> directory depends on
550+
<varname>min_wal_size</>, <varname>max_wal_size</> and
551+
the amount of WAL generated in previous checkpoint cycles. When old log
552+
segment files are no longer needed, they are removed or recycled (that is,
553+
renamed to become future segments in the numbered sequence). If, due to a
554+
short-term peak of log output rate, <varname>max_wal_size</> is
555+
exceeded, the unneeded segment files will be removed until the system
556+
gets back under this limit. Below that limit, the system recycles enough
557+
WAL files to cover the estimated need until the next checkpoint, and
558+
removes the rest. The estimate is based on a moving average of the number
559+
of WAL files used in previous checkpoint cycles. The moving average
560+
is increased immediately if the actual usage exceeds the estimate, so it
561+
accommodates peak usage rather average usage to some extent.
562+
<varname>min_wal_size</> puts a minimum on the amount of WAL files
563+
recycled for future usage; that much WAL is always recycled for future use,
564+
even if the system is idle and the WAL usage estimate suggests that little
565+
WAL is needed.
566+
</para>
567+
568+
<para>
569+
Independently of <varname>max_wal_size</varname>,
570+
<xref linkend="guc-wal-keep-segments"> + 1 most recent WAL files are
571+
kept at all times. Also, if WAL archiving is used, old segments can not be
572+
removed or recycled until they are archived. If WAL archiving cannot keep up
573+
with the pace that WAL is generated, or if <varname>archive_command</varname>
574+
fails repeatedly, old WAL files will accumulate in <filename>pg_xlog</>
575+
until the situation is resolved. A slow or failed standby server that
576+
uses a replication slot will have the same effect (see
577+
<xref linkend="streaming-replication-slots">).
560578
</para>
561579

562580
<para>
@@ -571,9 +589,8 @@
571589
master because restartpoints can only be performed at checkpoint records.
572590
A restartpoint is triggered when a checkpoint record is reached if at
573591
least <varname>checkpoint_timeout</> seconds have passed since the last
574-
restartpoint. In standby mode, a restartpoint is also triggered if at
575-
least <varname>checkpoint_segments</> log segments have been replayed
576-
since the last restartpoint.
592+
restartpoint, or if WAL size is about to exceed
593+
<varname>max_wal_size</>.
577594
</para>
578595

579596
<para>

0 commit comments

Comments
 (0)