1 <?xml version="1.0" encoding="UTF-8" standalone="no"?>
2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd"><html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>25.3. Continuous Archiving and Point-in-Time Recovery (PITR)</title><link rel="stylesheet" type="text/css" href="stylesheet.css" /><link rev="made" href="pgsql-docs@lists.postgresql.org" /><meta name="generator" content="DocBook XSL Stylesheets Vsnapshot" /><link rel="prev" href="backup-file.html" title="25.2. File System Level Backup" /><link rel="next" href="high-availability.html" title="Chapter 26. High Availability, Load Balancing, and Replication" /></head><body id="docContent" class="container-fluid col-10"><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="5" align="center">25.3. Continuous Archiving and Point-in-Time Recovery (PITR)</th></tr><tr><td width="10%" align="left"><a accesskey="p" href="backup-file.html" title="25.2. File System Level Backup">Prev</a> </td><td width="10%" align="left"><a accesskey="u" href="backup.html" title="Chapter 25. Backup and Restore">Up</a></td><th width="60%" align="center">Chapter 25. Backup and Restore</th><td width="10%" align="right"><a accesskey="h" href="index.html" title="PostgreSQL 18.0 Documentation">Home</a></td><td width="10%" align="right"> <a accesskey="n" href="high-availability.html" title="Chapter 26. High Availability, Load Balancing, and Replication">Next</a></td></tr></table><hr /></div><div class="sect1" id="CONTINUOUS-ARCHIVING"><div class="titlepage"><div><div><h2 class="title" style="clear: both">25.3. Continuous Archiving and Point-in-Time Recovery (PITR) <a href="#CONTINUOUS-ARCHIVING" class="id_link">#</a></h2></div></div></div><div class="toc"><dl class="toc"><dt><span class="sect2"><a href="continuous-archiving.html#BACKUP-ARCHIVING-WAL">25.3.1. Setting Up WAL Archiving</a></span></dt><dt><span class="sect2"><a href="continuous-archiving.html#BACKUP-BASE-BACKUP">25.3.2. Making a Base Backup</a></span></dt><dt><span class="sect2"><a href="continuous-archiving.html#BACKUP-INCREMENTAL-BACKUP">25.3.3. Making an Incremental Backup</a></span></dt><dt><span class="sect2"><a href="continuous-archiving.html#BACKUP-LOWLEVEL-BASE-BACKUP">25.3.4. Making a Base Backup Using the Low Level API</a></span></dt><dt><span class="sect2"><a href="continuous-archiving.html#BACKUP-PITR-RECOVERY">25.3.5. Recovering Using a Continuous Archive Backup</a></span></dt><dt><span class="sect2"><a href="continuous-archiving.html#BACKUP-TIMELINES">25.3.6. Timelines</a></span></dt><dt><span class="sect2"><a href="continuous-archiving.html#BACKUP-TIPS">25.3.7. Tips and Examples</a></span></dt><dt><span class="sect2"><a href="continuous-archiving.html#CONTINUOUS-ARCHIVING-CAVEATS">25.3.8. Caveats</a></span></dt></dl></div><a id="id-1.6.12.7.2" class="indexterm"></a><a id="id-1.6.12.7.3" class="indexterm"></a><a id="id-1.6.12.7.4" class="indexterm"></a><p>
3 At all times, <span class="productname">PostgreSQL</span> maintains a
4 <em class="firstterm">write ahead log</em> (WAL) in the <code class="filename">pg_wal/</code>
5 subdirectory of the cluster's data directory. The log records
6 every change made to the database's data files. This log exists
7 primarily for crash-safety purposes: if the system crashes, the
8 database can be restored to consistency by <span class="quote">“<span class="quote">replaying</span>”</span> the
9 log entries made since the last checkpoint. However, the existence
10 of the log makes it possible to use a third strategy for backing up
11 databases: we can combine a file-system-level backup with backup of
12 the WAL files. If recovery is needed, we restore the file system backup and
13 then replay from the backed-up WAL files to bring the system to a
14 current state. This approach is more complex to administer than
15 either of the previous approaches, but it has some significant
17 </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>
18 We do not need a perfectly consistent file system backup as the starting point.
19 Any internal inconsistency in the backup will be corrected by log
20 replay (this is not significantly different from what happens during
21 crash recovery). So we do not need a file system snapshot capability,
22 just <span class="application">tar</span> or a similar archiving tool.
23 </p></li><li class="listitem"><p>
24 Since we can combine an indefinitely long sequence of WAL files
25 for replay, continuous backup can be achieved simply by continuing to archive
26 the WAL files. This is particularly valuable for large databases, where
27 it might not be convenient to take a full backup frequently.
28 </p></li><li class="listitem"><p>
29 It is not necessary to replay the WAL entries all the
30 way to the end. We could stop the replay at any point and have a
31 consistent snapshot of the database as it was at that time. Thus,
32 this technique supports <em class="firstterm">point-in-time recovery</em>: it is
33 possible to restore the database to its state at any time since your base
35 </p></li><li class="listitem"><p>
36 If we continuously feed the series of WAL files to another
37 machine that has been loaded with the same base backup file, we
38 have a <em class="firstterm">warm standby</em> system: at any point we can bring up
39 the second machine and it will have a nearly-current copy of the
41 </p></li></ul></div><p>
42 </p><div class="note"><h3 class="title">Note</h3><p>
43 <span class="application">pg_dump</span> and
44 <span class="application">pg_dumpall</span> do not produce file-system-level
45 backups and cannot be used as part of a continuous-archiving solution.
46 Such dumps are <span class="emphasis"><em>logical</em></span> and do not contain enough
47 information to be used by WAL replay.
49 As with the plain file-system-backup technique, this method can only
50 support restoration of an entire database cluster, not a subset.
51 Also, it requires a lot of archival storage: the base backup might be bulky,
52 and a busy system will generate many megabytes of WAL traffic that
53 have to be archived. Still, it is the preferred backup technique in
54 many situations where high reliability is needed.
56 To recover successfully using continuous archiving (also called
57 <span class="quote">“<span class="quote">online backup</span>”</span> by many database vendors), you need a continuous
58 sequence of archived WAL files that extends back at least as far as the
59 start time of your backup. So to get started, you should set up and test
60 your procedure for archiving WAL files <span class="emphasis"><em>before</em></span> you take your
61 first base backup. Accordingly, we first discuss the mechanics of
63 </p><div class="sect2" id="BACKUP-ARCHIVING-WAL"><div class="titlepage"><div><div><h3 class="title">25.3.1. Setting Up WAL Archiving <a href="#BACKUP-ARCHIVING-WAL" class="id_link">#</a></h3></div></div></div><p>
64 In an abstract sense, a running <span class="productname">PostgreSQL</span> system
65 produces an indefinitely long sequence of WAL records. The system
66 physically divides this sequence into WAL <em class="firstterm">segment
67 files</em>, which are normally 16MB apiece (although the segment size
68 can be altered during <span class="application">initdb</span>). The segment
69 files are given numeric names that reflect their position in the
70 abstract WAL sequence. When not using WAL archiving, the system
71 normally creates just a few segment files and then
72 <span class="quote">“<span class="quote">recycles</span>”</span> them by renaming no-longer-needed segment files
73 to higher segment numbers. It's assumed that segment files whose
74 contents precede the last checkpoint are no longer of
75 interest and can be recycled.
77 When archiving WAL data, we need to capture the contents of each segment
78 file once it is filled, and save that data somewhere before the segment
79 file is recycled for reuse. Depending on the application and the
80 available hardware, there could be many different ways of <span class="quote">“<span class="quote">saving
81 the data somewhere</span>”</span>: we could copy the segment files to an NFS-mounted
82 directory on another machine, write them onto a tape drive (ensuring that
83 you have a way of identifying the original name of each file), or batch
84 them together and burn them onto CDs, or something else entirely. To
85 provide the database administrator with flexibility,
86 <span class="productname">PostgreSQL</span> tries not to make any assumptions about how
87 the archiving will be done. Instead, <span class="productname">PostgreSQL</span> lets
88 the administrator specify a shell command or an archive library to be executed to copy a
89 completed segment file to wherever it needs to go. This could be as simple
90 as a shell command that uses <code class="literal">cp</code>, or it could invoke a
91 complex C function — it's all up to you.
93 To enable WAL archiving, set the <a class="xref" href="runtime-config-wal.html#GUC-WAL-LEVEL">wal_level</a>
94 configuration parameter to <code class="literal">replica</code> or higher,
95 <a class="xref" href="runtime-config-wal.html#GUC-ARCHIVE-MODE">archive_mode</a> to <code class="literal">on</code>,
96 specify the shell command to use in the <a class="xref" href="runtime-config-wal.html#GUC-ARCHIVE-COMMAND">archive_command</a> configuration parameter
97 or specify the library to use in the <a class="xref" href="runtime-config-wal.html#GUC-ARCHIVE-LIBRARY">archive_library</a> configuration parameter. In practice
98 these settings will always be placed in the
99 <code class="filename">postgresql.conf</code> file.
101 In <code class="varname">archive_command</code>,
102 <code class="literal">%p</code> is replaced by the path name of the file to
103 archive, while <code class="literal">%f</code> is replaced by only the file name.
104 (The path name is relative to the current working directory,
105 i.e., the cluster's data directory.)
106 Use <code class="literal">%%</code> if you need to embed an actual <code class="literal">%</code>
107 character in the command. The simplest useful command is something
109 </p><pre class="programlisting">
110 archive_command = 'test ! -f /mnt/server/archivedir/%f && cp %p /mnt/server/archivedir/%f' # Unix
111 archive_command = 'copy "%p" "C:\\server\\archivedir\\%f"' # Windows
113 which will copy archivable WAL segments to the directory
114 <code class="filename">/mnt/server/archivedir</code>. (This is an example, not a
115 recommendation, and might not work on all platforms.) After the
116 <code class="literal">%p</code> and <code class="literal">%f</code> parameters have been replaced,
117 the actual command executed might look like this:
118 </p><pre class="programlisting">
119 test ! -f /mnt/server/archivedir/00000001000000A900000065 && cp pg_wal/00000001000000A900000065 /mnt/server/archivedir/00000001000000A900000065
121 A similar command will be generated for each new file to be archived.
123 The archive command will be executed under the ownership of the same
124 user that the <span class="productname">PostgreSQL</span> server is running as. Since
125 the series of WAL files being archived contains effectively everything
126 in your database, you will want to be sure that the archived data is
127 protected from prying eyes; for example, archive into a directory that
128 does not have group or world read access.
130 It is important that the archive command return zero exit status if and
131 only if it succeeds. Upon getting a zero result,
132 <span class="productname">PostgreSQL</span> will assume that the file has been
133 successfully archived, and will remove or recycle it. However, a nonzero
134 status tells <span class="productname">PostgreSQL</span> that the file was not archived;
135 it will try again periodically until it succeeds.
137 Another way to archive is to use a custom archive module as the
138 <code class="varname">archive_library</code>. Since such modules are written in
139 <code class="literal">C</code>, creating your own may require considerably more effort
140 than writing a shell command. However, archive modules can be more
141 performant than archiving via shell, and they will have access to many
142 useful server resources. For more information about archive modules, see
143 <a class="xref" href="archive-modules.html" title="Chapter 49. Archive Modules">Chapter 49</a>.
145 When the archive command is terminated by a signal (other than
146 <span class="systemitem">SIGTERM</span> that is used as part of a server
147 shutdown) or an error by the shell with an exit status greater than
148 125 (such as command not found), or if the archive function emits an
149 <code class="literal">ERROR</code> or <code class="literal">FATAL</code>, the archiver process
150 aborts and gets restarted by the postmaster. In such cases, the failure is
151 not reported in <a class="xref" href="monitoring-stats.html#PG-STAT-ARCHIVER-VIEW" title="Table 27.22. pg_stat_archiver View">pg_stat_archiver</a>.
153 Archive commands and libraries should generally be designed to refuse to overwrite
154 any pre-existing archive file. This is an important safety feature to
155 preserve the integrity of your archive in case of administrator error
156 (such as sending the output of two different servers to the same archive
157 directory). It is advisable to test your proposed archive library to ensure
158 that it does not overwrite an existing file.
160 In rare cases, <span class="productname">PostgreSQL</span> may attempt to
161 re-archive a WAL file that was previously archived. For example, if the
162 system crashes before the server makes a durable record of archival
163 success, the server will attempt to archive the file again after
164 restarting (provided archiving is still enabled). When an archive command or library
165 encounters a pre-existing file, it should return a zero status or <code class="literal">true</code>, respectively,
166 if the WAL file has identical contents to the pre-existing archive and the
167 pre-existing archive is fully persisted to storage. If a pre-existing
168 file contains different contents than the WAL file being archived, the
169 archive command or library <span class="emphasis"><em>must</em></span> return a nonzero status or
170 <code class="literal">false</code>, respectively.
172 The example command above for Unix avoids overwriting a pre-existing archive
173 by including a separate
174 <code class="command">test</code> step. On some Unix platforms, <code class="command">cp</code> has
175 switches such as <code class="option">-i</code> that can be used to do the same thing
176 less verbosely, but you should not rely on these without verifying that
177 the right exit status is returned. (In particular, GNU <code class="command">cp</code>
178 will return status zero when <code class="option">-i</code> is used and the target file
179 already exists, which is <span class="emphasis"><em>not</em></span> the desired behavior.)
181 While designing your archiving setup, consider what will happen if
182 the archive command or library fails repeatedly because some aspect requires
183 operator intervention or the archive runs out of space. For example, this
184 could occur if you write to tape without an autochanger; when the tape
185 fills, nothing further can be archived until the tape is swapped.
186 You should ensure that any error condition or request to a human operator
187 is reported appropriately so that the situation can be
188 resolved reasonably quickly. The <code class="filename">pg_wal/</code> directory will
189 continue to fill with WAL segment files until the situation is resolved.
190 (If the file system containing <code class="filename">pg_wal/</code> fills up,
191 <span class="productname">PostgreSQL</span> will do a PANIC shutdown. No committed
192 transactions will be lost, but the database will remain offline until
193 you free some space.)
195 The speed of the archive command or library is unimportant as long as it can keep up
196 with the average rate at which your server generates WAL data. Normal
197 operation continues even if the archiving process falls a little behind.
198 If archiving falls significantly behind, this will increase the amount of
199 data that would be lost in the event of a disaster. It will also mean that
200 the <code class="filename">pg_wal/</code> directory will contain large numbers of
201 not-yet-archived segment files, which could eventually exceed available
202 disk space. You are advised to monitor the archiving process to ensure that
203 it is working as you intend.
205 In writing your archive command or library, you should assume that the file names to
206 be archived can be up to 64 characters long and can contain any
207 combination of ASCII letters, digits, and dots. It is not necessary to
208 preserve the original relative path (<code class="literal">%p</code>) but it is necessary to
209 preserve the file name (<code class="literal">%f</code>).
211 Note that although WAL archiving will allow you to restore any
212 modifications made to the data in your <span class="productname">PostgreSQL</span> database,
213 it will not restore changes made to configuration files (that is,
214 <code class="filename">postgresql.conf</code>, <code class="filename">pg_hba.conf</code> and
215 <code class="filename">pg_ident.conf</code>), since those are edited manually rather
216 than through SQL operations.
217 You might wish to keep the configuration files in a location that will
218 be backed up by your regular file system backup procedures. See
219 <a class="xref" href="runtime-config-file-locations.html" title="19.2. File Locations">Section 19.2</a> for how to relocate the
222 The archive command or function is only invoked on completed WAL segments. Hence,
223 if your server generates only little WAL traffic (or has slack periods
224 where it does so), there could be a long delay between the completion
225 of a transaction and its safe recording in archive storage. To put
226 a limit on how old unarchived data can be, you can set
227 <a class="xref" href="runtime-config-wal.html#GUC-ARCHIVE-TIMEOUT">archive_timeout</a> to force the server to switch
228 to a new WAL segment file at least that often. Note that archived
229 files that are archived early due to a forced switch are still the same
230 length as completely full files. It is therefore unwise to set a very
231 short <code class="varname">archive_timeout</code> — it will bloat your archive
232 storage. <code class="varname">archive_timeout</code> settings of a minute or so are
235 Also, you can force a segment switch manually with
236 <code class="function">pg_switch_wal</code> if you want to ensure that a
237 just-finished transaction is archived as soon as possible. Other utility
238 functions related to WAL management are listed in <a class="xref" href="functions-admin.html#FUNCTIONS-ADMIN-BACKUP-TABLE" title="Table 9.97. Backup Control Functions">Table 9.97</a>.
240 When <code class="varname">wal_level</code> is <code class="literal">minimal</code> some SQL commands
241 are optimized to avoid WAL logging, as described in <a class="xref" href="populate.html#POPULATE-PITR" title="14.4.7. Disable WAL Archival and Streaming Replication">Section 14.4.7</a>. If archiving or streaming replication were
242 turned on during execution of one of these statements, WAL would not
243 contain enough information for archive recovery. (Crash recovery is
244 unaffected.) For this reason, <code class="varname">wal_level</code> can only be changed at
245 server start. However, <code class="varname">archive_command</code> and <code class="varname">archive_library</code> can be changed with a
246 configuration file reload. If you are archiving via shell and wish to
247 temporarily stop archiving,
248 one way to do it is to set <code class="varname">archive_command</code> to the empty
249 string (<code class="literal">''</code>).
250 This will cause WAL files to accumulate in <code class="filename">pg_wal/</code> until a
251 working <code class="varname">archive_command</code> is re-established.
252 </p></div><div class="sect2" id="BACKUP-BASE-BACKUP"><div class="titlepage"><div><div><h3 class="title">25.3.2. Making a Base Backup <a href="#BACKUP-BASE-BACKUP" class="id_link">#</a></h3></div></div></div><p>
253 The easiest way to perform a base backup is to use the
254 <a class="xref" href="app-pgbasebackup.html" title="pg_basebackup"><span class="refentrytitle"><span class="application">pg_basebackup</span></span></a> tool. It can create
255 a base backup either as regular files or as a tar archive. If more
256 flexibility than <a class="xref" href="app-pgbasebackup.html" title="pg_basebackup"><span class="refentrytitle"><span class="application">pg_basebackup</span></span></a> can provide is
257 required, you can also make a base backup using the low level API
258 (see <a class="xref" href="continuous-archiving.html#BACKUP-LOWLEVEL-BASE-BACKUP" title="25.3.4. Making a Base Backup Using the Low Level API">Section 25.3.4</a>).
260 It is not necessary to be concerned about the amount of time it takes
261 to make a base backup. However, if you normally run the
262 server with <code class="varname">full_page_writes</code> disabled, you might notice a drop
263 in performance while the backup runs since <code class="varname">full_page_writes</code> is
264 effectively forced on during backup mode.
266 To make use of the backup, you will need to keep all the WAL
267 segment files generated during and after the file system backup.
268 To aid you in doing this, the base backup process
269 creates a <em class="firstterm">backup history file</em> that is immediately
270 stored into the WAL archive area. This file is named after the first
271 WAL segment file that you need for the file system backup.
272 For example, if the starting WAL file is
273 <code class="literal">0000000100001234000055CD</code> the backup history file will be
275 <code class="literal">0000000100001234000055CD.007C9330.backup</code>. (The second
276 part of the file name stands for an exact position within the WAL
277 file, and can ordinarily be ignored.) Once you have safely archived
278 the file system backup and the WAL segment files used during the
279 backup (as specified in the backup history file), all archived WAL
280 segments with names numerically less are no longer needed to recover
281 the file system backup and can be deleted. However, you should
282 consider keeping several backup sets to be absolutely certain that
283 you can recover your data.
285 The backup history file is just a small text file. It contains the
286 label string you gave to <a class="xref" href="app-pgbasebackup.html" title="pg_basebackup"><span class="refentrytitle"><span class="application">pg_basebackup</span></span></a>, as well as
287 the starting and ending times and WAL segments of the backup.
288 If you used the label to identify the associated dump file,
289 then the archived history file is enough to tell you which dump file to
292 Since you have to keep around all the archived WAL files back to your
293 last base backup, the interval between base backups should usually be
294 chosen based on how much storage you want to expend on archived WAL
295 files. You should also consider how long you are prepared to spend
296 recovering, if recovery should be necessary — the system will have to
297 replay all those WAL segments, and that could take awhile if it has
298 been a long time since the last base backup.
299 </p></div><div class="sect2" id="BACKUP-INCREMENTAL-BACKUP"><div class="titlepage"><div><div><h3 class="title">25.3.3. Making an Incremental Backup <a href="#BACKUP-INCREMENTAL-BACKUP" class="id_link">#</a></h3></div></div></div><p>
300 You can use <a class="xref" href="app-pgbasebackup.html" title="pg_basebackup"><span class="refentrytitle"><span class="application">pg_basebackup</span></span></a> to take an incremental
301 backup by specifying the <code class="literal">--incremental</code> option. You must
302 supply, as an argument to <code class="literal">--incremental</code>, the backup
303 manifest to an earlier backup from the same server. In the resulting
304 backup, non-relation files will be included in their entirety, but some
305 relation files may be replaced by smaller incremental files which contain
306 only the blocks which have been changed since the earlier backup and enough
307 metadata to reconstruct the current version of the file.
309 To figure out which blocks need to be backed up, the server uses WAL
310 summaries, which are stored in the data directory, inside the directory
311 <code class="literal">pg_wal/summaries</code>. If the required summary files are not
312 present, an attempt to take an incremental backup will fail. The summaries
313 present in this directory must cover all LSNs from the start LSN of the
314 prior backup to the start LSN of the current backup. Since the server looks
315 for WAL summaries just after establishing the start LSN of the current
316 backup, the necessary summary files probably won't be instantly present
317 on disk, but the server will wait for any missing files to show up.
318 This also helps if the WAL summarization process has fallen behind.
319 However, if the necessary files have already been removed, or if the WAL
320 summarizer doesn't catch up quickly enough, the incremental backup will
323 When restoring an incremental backup, it will be necessary to have not
324 only the incremental backup itself but also all earlier backups that
325 are required to supply the blocks omitted from the incremental backup.
326 See <a class="xref" href="app-pgcombinebackup.html" title="pg_combinebackup"><span class="refentrytitle"><span class="application">pg_combinebackup</span></span></a> for further information about
327 this requirement. Note that there are restrictions on the use of
328 <code class="literal">pg_combinebackup</code> when the checksum status of the
329 cluster has been changed; see
330 <a class="link" href="app-pgcombinebackup.html#APP-PGCOMBINEBACKUP-LIMITATIONS" title="Limitations">pg_combinebackup
333 Note that all of the requirements for making use of a full backup also
334 apply to an incremental backup. For instance, you still need all of the
335 WAL segment files generated during and after the file system backup, and
336 any relevant WAL history files. And you still need to create a
337 <code class="literal">recovery.signal</code> (or <code class="literal">standby.signal</code>)
338 and perform recovery, as described in
339 <a class="xref" href="continuous-archiving.html#BACKUP-PITR-RECOVERY" title="25.3.5. Recovering Using a Continuous Archive Backup">Section 25.3.5</a>. The requirement to have earlier
340 backups available at restore time and to use
341 <code class="literal">pg_combinebackup</code> is an additional requirement on top of
342 everything else. Keep in mind that <span class="application">PostgreSQL</span>
343 has no built-in mechanism to figure out which backups are still needed as
344 a basis for restoring later incremental backups. You must keep track of
345 the relationships between your full and incremental backups on your own,
346 and be certain not to remove earlier backups if they might be needed when
347 restoring later incremental backups.
349 Incremental backups typically only make sense for relatively large
350 databases where a significant portion of the data does not change, or only
351 changes slowly. For a small database, it's simpler to ignore the existence
352 of incremental backups and simply take full backups, which are simpler
353 to manage. For a large database all of which is heavily modified,
354 incremental backups won't be much smaller than full backups.
356 An incremental backup is only possible if replay would begin from a later
357 checkpoint than for the previous backup upon which it depends. If you
358 take the incremental backup on the primary, this condition is always
359 satisfied, because each backup triggers a new checkpoint. On a standby,
360 replay begins from the most recent restartpoint. Therefore, an
361 incremental backup of a standby server can fail if there has been very
362 little activity since the previous backup, since no new restartpoint might
364 </p></div><div class="sect2" id="BACKUP-LOWLEVEL-BASE-BACKUP"><div class="titlepage"><div><div><h3 class="title">25.3.4. Making a Base Backup Using the Low Level API <a href="#BACKUP-LOWLEVEL-BASE-BACKUP" class="id_link">#</a></h3></div></div></div><p>
365 Instead of taking a full or incremental base backup using
366 <a class="xref" href="app-pgbasebackup.html" title="pg_basebackup"><span class="refentrytitle"><span class="application">pg_basebackup</span></span></a>, you can take a base backup using the
367 low-level API. This procedure contains a few more steps than
368 the <span class="application">pg_basebackup</span> method, but is relatively
369 simple. It is very important that these steps are executed in
370 sequence, and that the success of a step is verified before
371 proceeding to the next step.
373 Multiple backups are able to be run concurrently (both those
374 started using this backup API and those started using
375 <a class="xref" href="app-pgbasebackup.html" title="pg_basebackup"><span class="refentrytitle"><span class="application">pg_basebackup</span></span></a>).
377 </p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
378 Ensure that WAL archiving is enabled and working.
379 </p></li><li class="listitem"><p>
380 Connect to the server (it does not matter which database) as a user with
381 rights to run <code class="function">pg_backup_start</code> (superuser,
382 or a user who has been granted <code class="literal">EXECUTE</code> on the
383 function) and issue the command:
384 </p><pre class="programlisting">
385 SELECT pg_backup_start(label => 'label', fast => false);
387 where <code class="literal">label</code> is any string you want to use to uniquely
388 identify this backup operation. The connection
389 calling <code class="function">pg_backup_start</code> must be maintained until the end of
390 the backup, or the backup will be automatically aborted.
392 Online backups are always started at the beginning of a checkpoint.
393 By default, <code class="function">pg_backup_start</code> will wait for the next
394 regularly scheduled checkpoint to complete, which may take a long time (see the
395 configuration parameters <a class="xref" href="runtime-config-wal.html#GUC-CHECKPOINT-TIMEOUT">checkpoint_timeout</a> and
396 <a class="xref" href="runtime-config-wal.html#GUC-CHECKPOINT-COMPLETION-TARGET">checkpoint_completion_target</a>). This is
397 usually preferable as it minimizes the impact on the running system. If you
398 want to start the backup as soon as possible, pass <code class="literal">true</code> as
399 the second parameter to <code class="function">pg_backup_start</code> and it will
400 request an immediate checkpoint, which will finish as fast as possible using
401 as much I/O as possible.
402 </p></li><li class="listitem"><p>
403 Perform the backup, using any convenient file-system-backup tool
404 such as <span class="application">tar</span> or <span class="application">cpio</span> (not
405 <span class="application">pg_dump</span> or
406 <span class="application">pg_dumpall</span>). It is neither
407 necessary nor desirable to stop normal operation of the database
408 while you do this. See
409 <a class="xref" href="continuous-archiving.html#BACKUP-LOWLEVEL-BASE-BACKUP-DATA" title="25.3.4.1. Backing Up the Data Directory">Section 25.3.4.1</a> for things to
410 consider during this backup.
411 </p></li><li class="listitem"><p>
412 In the same connection as before, issue the command:
413 </p><pre class="programlisting">
414 SELECT * FROM pg_backup_stop(wait_for_archive => true);
416 This terminates backup mode. On a primary, it also performs an automatic
417 switch to the next WAL segment. On a standby, it is not possible to
418 automatically switch WAL segments, so you may wish to run
419 <code class="function">pg_switch_wal</code> on the primary to perform a manual
420 switch. The reason for the switch is to arrange for
421 the last WAL segment file written during the backup interval to be
424 <code class="function">pg_backup_stop</code> will return one row with three
425 values. The second of these fields should be written to a file named
426 <code class="filename">backup_label</code> in the root directory of the backup. The
427 third field should be written to a file named
428 <code class="filename">tablespace_map</code> unless the field is empty. These files are
429 vital to the backup working and must be written byte for byte without
430 modification, which may require opening the file in binary mode.
431 </p></li><li class="listitem"><p>
432 Once the WAL segment files active during the backup are archived, you are
433 done. The file identified by <code class="function">pg_backup_stop</code>'s first return
434 value is the last segment that is required to form a complete set of
435 backup files. On a primary, if <code class="varname">archive_mode</code> is enabled and the
436 <code class="literal">wait_for_archive</code> parameter is <code class="literal">true</code>,
437 <code class="function">pg_backup_stop</code> does not return until the last segment has
439 On a standby, <code class="varname">archive_mode</code> must be <code class="literal">always</code> in order
440 for <code class="function">pg_backup_stop</code> to wait.
441 Archiving of these files happens automatically since you have
442 already configured <code class="varname">archive_command</code> or <code class="varname">archive_library</code>.
443 In most cases this happens quickly, but you are advised to monitor your
444 archive system to ensure there are no delays.
445 If the archive process has fallen behind because of failures of the
446 archive command or library, it will keep retrying
447 until the archive succeeds and the backup is complete.
448 If you wish to place a time limit on the execution of
449 <code class="function">pg_backup_stop</code>, set an appropriate
450 <code class="varname">statement_timeout</code> value, but make note that if
451 <code class="function">pg_backup_stop</code> terminates because of this your backup
454 If the backup process monitors and ensures that all WAL segment files
455 required for the backup are successfully archived then the
456 <code class="literal">wait_for_archive</code> parameter (which defaults to true) can be set
458 <code class="function">pg_backup_stop</code> return as soon as the stop backup record is
459 written to the WAL. By default, <code class="function">pg_backup_stop</code> will wait
460 until all WAL has been archived, which can take some time. This option
461 must be used with caution: if WAL archiving is not monitored correctly
462 then the backup might not include all of the WAL files and will
463 therefore be incomplete and not able to be restored.
464 </p></li></ol></div><p>
465 </p><div class="sect3" id="BACKUP-LOWLEVEL-BASE-BACKUP-DATA"><div class="titlepage"><div><div><h4 class="title">25.3.4.1. Backing Up the Data Directory <a href="#BACKUP-LOWLEVEL-BASE-BACKUP-DATA" class="id_link">#</a></h4></div></div></div><p>
466 Some file system backup tools emit warnings or errors
467 if the files they are trying to copy change while the copy proceeds.
468 When taking a base backup of an active database, this situation is normal
469 and not an error. However, you need to ensure that you can distinguish
470 complaints of this sort from real errors. For example, some versions
471 of <span class="application">rsync</span> return a separate exit code for
472 <span class="quote">“<span class="quote">vanished source files</span>”</span>, and you can write a driver script to
473 accept this exit code as a non-error case. Also, some versions of
474 GNU <span class="application">tar</span> return an error code indistinguishable from
475 a fatal error if a file was truncated while <span class="application">tar</span> was
476 copying it. Fortunately, GNU <span class="application">tar</span> versions 1.16 and
477 later exit with 1 if a file was changed during the backup,
478 and 2 for other errors. With GNU <span class="application">tar</span> version 1.23 and
479 later, you can use the warning options <code class="literal">--warning=no-file-changed
480 --warning=no-file-removed</code> to hide the related warning messages.
482 Be certain that your backup includes all of the files under
483 the database cluster directory (e.g., <code class="filename">/usr/local/pgsql/data</code>).
484 If you are using tablespaces that do not reside underneath this directory,
485 be careful to include them as well (and be sure that your backup
486 archives symbolic links as links, otherwise the restore will corrupt
489 You should, however, omit from the backup the files within the
490 cluster's <code class="filename">pg_wal/</code> subdirectory. This
491 slight adjustment is worthwhile because it reduces the risk
492 of mistakes when restoring. This is easy to arrange if
493 <code class="filename">pg_wal/</code> is a symbolic link pointing to someplace outside
494 the cluster directory, which is a common setup anyway for performance
495 reasons. You might also want to exclude <code class="filename">postmaster.pid</code>
496 and <code class="filename">postmaster.opts</code>, which record information
497 about the running <span class="application">postmaster</span>, not about the
498 <span class="application">postmaster</span> which will eventually use this backup.
499 (These files can confuse <span class="application">pg_ctl</span>.)
501 It is often a good idea to also omit from the backup the files
502 within the cluster's <code class="filename">pg_replslot/</code> directory, so that
503 replication slots that exist on the primary do not become part of the
504 backup. Otherwise, the subsequent use of the backup to create a standby
505 may result in indefinite retention of WAL files on the standby, and
506 possibly bloat on the primary if hot standby feedback is enabled, because
507 the clients that are using those replication slots will still be connecting
508 to and updating the slots on the primary, not the standby. Even if the
509 backup is only intended for use in creating a new primary, copying the
510 replication slots isn't expected to be particularly useful, since the
511 contents of those slots will likely be badly out of date by the time
512 the new primary comes on line.
514 The contents of the directories <code class="filename">pg_dynshmem/</code>,
515 <code class="filename">pg_notify/</code>, <code class="filename">pg_serial/</code>,
516 <code class="filename">pg_snapshots/</code>, <code class="filename">pg_stat_tmp/</code>,
517 and <code class="filename">pg_subtrans/</code> (but not the directories themselves) can be
518 omitted from the backup as they will be initialized on postmaster startup.
520 Any file or directory beginning with <code class="filename">pgsql_tmp</code> can be
521 omitted from the backup. These files are removed on postmaster start and
522 the directories will be recreated as needed.
524 <code class="filename">pg_internal.init</code> files can be omitted from the
525 backup whenever a file of that name is found. These files contain
526 relation cache data that is always rebuilt when recovering.
529 file includes the label string you gave to <code class="function">pg_backup_start</code>,
530 as well as the time at which <code class="function">pg_backup_start</code> was run, and
531 the name of the starting WAL file. In case of confusion it is therefore
532 possible to look inside a backup file and determine exactly which
533 backup session the dump file came from. The tablespace map file includes
534 the symbolic link names as they exist in the directory
535 <code class="filename">pg_tblspc/</code> and the full path of each symbolic link.
536 These files are not merely for your information; their presence and
537 contents are critical to the proper operation of the system's recovery
540 It is also possible to make a backup while the server is
541 stopped. In this case, you obviously cannot use
542 <code class="function">pg_backup_start</code> or <code class="function">pg_backup_stop</code>, and
543 you will therefore be left to your own devices to keep track of which
544 backup is which and how far back the associated WAL files go.
545 It is generally better to follow the continuous archiving procedure above.
546 </p></div></div><div class="sect2" id="BACKUP-PITR-RECOVERY"><div class="titlepage"><div><div><h3 class="title">25.3.5. Recovering Using a Continuous Archive Backup <a href="#BACKUP-PITR-RECOVERY" class="id_link">#</a></h3></div></div></div><p>
547 Okay, the worst has happened and you need to recover from your backup.
548 Here is the procedure:
549 </p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
550 Stop the server, if it's running.
551 </p></li><li class="listitem"><p>
552 If you have the space to do so,
553 copy the whole cluster data directory and any tablespaces to a temporary
554 location in case you need them later. Note that this precaution will
555 require that you have enough free space on your system to hold two
556 copies of your existing database. If you do not have enough space,
557 you should at least save the contents of the cluster's <code class="filename">pg_wal</code>
558 subdirectory, as it might contain WAL files which
559 were not archived before the system went down.
560 </p></li><li class="listitem"><p>
561 Remove all existing files and subdirectories under the cluster data
562 directory and under the root directories of any tablespaces you are using.
563 </p></li><li class="listitem"><p>
564 If you're restoring a full backup, you can restore the database files
565 directly into the target directories. Be sure that they
566 are restored with the right ownership (the database system user, not
567 <code class="literal">root</code>!) and with the right permissions. If you are using
569 you should verify that the symbolic links in <code class="filename">pg_tblspc/</code>
570 were correctly restored.
571 </p></li><li class="listitem"><p>
572 If you're restoring an incremental backup, you'll need to restore the
573 incremental backup and all earlier backups upon which it directly or
574 indirectly depends to the machine where you are performing the restore.
575 These backups will need to be placed in separate directories, not the
576 target directories where you want the running server to end up.
577 Once this is done, use <a class="xref" href="app-pgcombinebackup.html" title="pg_combinebackup"><span class="refentrytitle"><span class="application">pg_combinebackup</span></span></a> to pull
578 data from the full backup and all of the subsequent incremental backups
579 and write out a synthetic full backup to the target directories. As above,
580 verify that permissions and tablespace links are correct.
581 </p></li><li class="listitem"><p>
582 Remove any files present in <code class="filename">pg_wal/</code>; these came from the
583 file system backup and are therefore probably obsolete rather than current.
584 If you didn't archive <code class="filename">pg_wal/</code> at all, then recreate
585 it with proper permissions,
586 being careful to ensure that you re-establish it as a symbolic link
587 if you had it set up that way before.
588 </p></li><li class="listitem"><p>
589 If you have unarchived WAL segment files that you saved in step 2,
590 copy them into <code class="filename">pg_wal/</code>. (It is best to copy them,
591 not move them, so you still have the unmodified files if a
592 problem occurs and you have to start over.)
593 </p></li><li class="listitem"><p>
594 Set recovery configuration settings in
595 <code class="filename">postgresql.conf</code> (see <a class="xref" href="runtime-config-wal.html#RUNTIME-CONFIG-WAL-ARCHIVE-RECOVERY" title="19.5.5. Archive Recovery">Section 19.5.5</a>) and create a file
596 <code class="filename">recovery.signal</code> in the cluster
597 data directory. You might
598 also want to temporarily modify <code class="filename">pg_hba.conf</code> to prevent
599 ordinary users from connecting until you are sure the recovery was successful.
600 </p></li><li class="listitem"><p>
601 Start the server. The server will go into recovery mode and
602 proceed to read through the archived WAL files it needs. Should the
603 recovery be terminated because of an external error, the server can
604 simply be restarted and it will continue recovery. Upon completion
605 of the recovery process, the server will remove
606 <code class="filename">recovery.signal</code> (to prevent
607 accidentally re-entering recovery mode later) and then
608 commence normal database operations.
609 </p></li><li class="listitem"><p>
610 Inspect the contents of the database to ensure you have recovered to
611 the desired state. If not, return to step 1. If all is well,
612 allow your users to connect by restoring <code class="filename">pg_hba.conf</code> to normal.
613 </p></li></ol></div><p>
615 The key part of all this is to set up a recovery configuration that
616 describes how you want to recover and how far the recovery should
617 run. The one thing that you absolutely must specify is the <code class="varname">restore_command</code>,
618 which tells <span class="productname">PostgreSQL</span> how to retrieve archived
619 WAL file segments. Like the <code class="varname">archive_command</code>, this is
620 a shell command string. It can contain <code class="literal">%f</code>, which is
621 replaced by the name of the desired WAL file, and <code class="literal">%p</code>,
622 which is replaced by the path name to copy the WAL file to.
623 (The path name is relative to the current working directory,
624 i.e., the cluster's data directory.)
625 Write <code class="literal">%%</code> if you need to embed an actual <code class="literal">%</code>
626 character in the command. The simplest useful command is
628 </p><pre class="programlisting">
629 restore_command = 'cp /mnt/server/archivedir/%f %p'
631 which will copy previously archived WAL segments from the directory
632 <code class="filename">/mnt/server/archivedir</code>. Of course, you can use something
633 much more complicated, perhaps even a shell script that requests the
634 operator to mount an appropriate tape.
636 It is important that the command return nonzero exit status on failure.
637 The command <span class="emphasis"><em>will</em></span> be called requesting files that are not
638 present in the archive; it must return nonzero when so asked. This is not
639 an error condition. An exception is that if the command was terminated by
640 a signal (other than <span class="systemitem">SIGTERM</span>, which is used as
641 part of a database server shutdown) or an error by the shell (such as
642 command not found), then recovery will abort and the server will not start
645 Not all of the requested files will be WAL segment
646 files; you should also expect requests for files with a suffix of
647 <code class="literal">.history</code>. Also be aware that
648 the base name of the <code class="literal">%p</code> path will be different from
649 <code class="literal">%f</code>; do not expect them to be interchangeable.
651 WAL segments that cannot be found in the archive will be sought in
652 <code class="filename">pg_wal/</code>; this allows use of recent un-archived segments.
653 However, segments that are available from the archive will be used in
654 preference to files in <code class="filename">pg_wal/</code>.
656 Normally, recovery will proceed through all available WAL segments,
657 thereby restoring the database to the current point in time (or as
658 close as possible given the available WAL segments). Therefore, a normal
659 recovery will end with a <span class="quote">“<span class="quote">file not found</span>”</span> message, the exact text
660 of the error message depending upon your choice of
661 <code class="varname">restore_command</code>. You may also see an error message
662 at the start of recovery for a file named something like
663 <code class="filename">00000001.history</code>. This is also normal and does not
664 indicate a problem in simple recovery situations; see
665 <a class="xref" href="continuous-archiving.html#BACKUP-TIMELINES" title="25.3.6. Timelines">Section 25.3.6</a> for discussion.
667 If you want to recover to some previous point in time (say, right before
668 the junior DBA dropped your main transaction table), just specify the
669 required <a class="link" href="runtime-config-wal.html#RUNTIME-CONFIG-WAL-RECOVERY-TARGET" title="19.5.6. Recovery Target">stopping point</a>. You can specify
670 the stop point, known as the <span class="quote">“<span class="quote">recovery target</span>”</span>, either by
671 date/time, named restore point or by completion of a specific transaction
672 ID. As of this writing only the date/time and named restore point options
673 are very usable, since there are no tools to help you identify with any
674 accuracy which transaction ID to use.
675 </p><div class="note"><h3 class="title">Note</h3><p>
676 The stop point must be after the ending time of the base backup, i.e.,
677 the end time of <code class="function">pg_backup_stop</code>. You cannot use a base backup
678 to recover to a time when that backup was in progress. (To
679 recover to such a time, you must go back to your previous base backup
680 and roll forward from there.)
682 If recovery finds corrupted WAL data, recovery will
683 halt at that point and the server will not start. In such a case the
684 recovery process could be re-run from the beginning, specifying a
685 <span class="quote">“<span class="quote">recovery target</span>”</span> before the point of corruption so that recovery
686 can complete normally.
687 If recovery fails for an external reason, such as a system crash or
688 if the WAL archive has become inaccessible, then the recovery can simply
689 be restarted and it will restart almost from where it failed.
690 Recovery restart works much like checkpointing in normal operation:
691 the server periodically forces all its state to disk, and then updates
692 the <code class="filename">pg_control</code> file to indicate that the already-processed
693 WAL data need not be scanned again.
694 </p></div><div class="sect2" id="BACKUP-TIMELINES"><div class="titlepage"><div><div><h3 class="title">25.3.6. Timelines <a href="#BACKUP-TIMELINES" class="id_link">#</a></h3></div></div></div><a id="id-1.6.12.7.14.2" class="indexterm"></a><p>
695 The ability to restore the database to a previous point in time creates
696 some complexities that are akin to science-fiction stories about time
697 travel and parallel universes. For example, in the original history of the database,
698 suppose you dropped a critical table at 5:15PM on Tuesday evening, but
699 didn't realize your mistake until Wednesday noon.
700 Unfazed, you get out your backup, restore to the point-in-time 5:14PM
701 Tuesday evening, and are up and running. In <span class="emphasis"><em>this</em></span> history of
702 the database universe, you never dropped the table. But suppose
703 you later realize this wasn't such a great idea, and would like
704 to return to sometime Wednesday morning in the original history.
706 to if, while your database was up-and-running, it overwrote some of the
707 WAL segment files that led up to the time you now wish you
708 could get back to. Thus, to avoid this, you need to distinguish the series of
709 WAL records generated after you've done a point-in-time recovery from
710 those that were generated in the original database history.
712 To deal with this problem, <span class="productname">PostgreSQL</span> has a notion
713 of <em class="firstterm">timelines</em>. Whenever an archive recovery completes,
714 a new timeline is created to identify the series of WAL records
715 generated after that recovery. The timeline
716 ID number is part of WAL segment file names so a new timeline does
717 not overwrite the WAL data generated by previous timelines.
718 For example, in the WAL file name
719 <code class="filename">0000000100001234000055CD</code>, the leading
720 <code class="literal">00000001</code> is the timeline ID in hexadecimal. (Note that
721 in other contexts, such as server log messages, timeline IDs are
722 usually printed in decimal.)
725 in fact possible to archive many different timelines. While that might
726 seem like a useless feature, it's often a lifesaver. Consider the
727 situation where you aren't quite sure what point-in-time to recover to,
728 and so have to do several point-in-time recoveries by trial and error
729 until you find the best place to branch off from the old history. Without
730 timelines this process would soon generate an unmanageable mess. With
731 timelines, you can recover to <span class="emphasis"><em>any</em></span> prior state, including
732 states in timeline branches that you abandoned earlier.
734 Every time a new timeline is created, <span class="productname">PostgreSQL</span> creates
735 a <span class="quote">“<span class="quote">timeline history</span>”</span> file that shows which timeline it branched
736 off from and when. These history files are necessary to allow the system
737 to pick the right WAL segment files when recovering from an archive that
738 contains multiple timelines. Therefore, they are archived into the WAL
739 archive area just like WAL segment files. The history files are just
740 small text files, so it's cheap and appropriate to keep them around
741 indefinitely (unlike the segment files which are large). You can, if
742 you like, add comments to a history file to record your own notes about
743 how and why this particular timeline was created. Such comments will be
744 especially valuable when you have a thicket of different timelines as
745 a result of experimentation.
747 The default behavior of recovery is to recover to the latest timeline found
748 in the archive. If you wish to recover to the timeline that was current
749 when the base backup was taken or into a specific child timeline (that
750 is, you want to return to some state that was itself generated after a
751 recovery attempt), you need to specify <code class="literal">current</code> or the
752 target timeline ID in <a class="xref" href="runtime-config-wal.html#GUC-RECOVERY-TARGET-TIMELINE">recovery_target_timeline</a>. You
753 cannot recover into timelines that branched off earlier than the base backup.
754 </p></div><div class="sect2" id="BACKUP-TIPS"><div class="titlepage"><div><div><h3 class="title">25.3.7. Tips and Examples <a href="#BACKUP-TIPS" class="id_link">#</a></h3></div></div></div><p>
755 Some tips for configuring continuous archiving are given here.
756 </p><div class="sect3" id="BACKUP-STANDALONE"><div class="titlepage"><div><div><h4 class="title">25.3.7.1. Standalone Hot Backups <a href="#BACKUP-STANDALONE" class="id_link">#</a></h4></div></div></div><p>
757 It is possible to use <span class="productname">PostgreSQL</span>'s backup facilities to
758 produce standalone hot backups. These are backups that cannot be used
759 for point-in-time recovery, yet are typically much faster to backup and
760 restore than <span class="application">pg_dump</span> dumps. (They are also much larger
761 than <span class="application">pg_dump</span> dumps, so in some cases the speed advantage
764 As with base backups, the easiest way to produce a standalone
765 hot backup is to use the <a class="xref" href="app-pgbasebackup.html" title="pg_basebackup"><span class="refentrytitle"><span class="application">pg_basebackup</span></span></a>
766 tool. If you include the <code class="literal">-X</code> parameter when calling
767 it, all the write-ahead log required to use the backup will be
768 included in the backup automatically, and no special action is
769 required to restore the backup.
770 </p></div><div class="sect3" id="COMPRESSED-ARCHIVE-LOGS"><div class="titlepage"><div><div><h4 class="title">25.3.7.2. Compressed Archive Logs <a href="#COMPRESSED-ARCHIVE-LOGS" class="id_link">#</a></h4></div></div></div><p>
771 If archive storage size is a concern, you can use
772 <span class="application">gzip</span> to compress the archive files:
773 </p><pre class="programlisting">
774 archive_command = 'gzip < %p > /mnt/server/archivedir/%f.gz'
776 You will then need to use <span class="application">gunzip</span> during recovery:
777 </p><pre class="programlisting">
778 restore_command = 'gunzip < /mnt/server/archivedir/%f.gz > %p'
780 </p></div><div class="sect3" id="BACKUP-SCRIPTS"><div class="titlepage"><div><div><h4 class="title">25.3.7.3. <code class="varname">archive_command</code> Scripts <a href="#BACKUP-SCRIPTS" class="id_link">#</a></h4></div></div></div><p>
781 Many people choose to use scripts to define their
782 <code class="varname">archive_command</code>, so that their
783 <code class="filename">postgresql.conf</code> entry looks very simple:
784 </p><pre class="programlisting">
785 archive_command = 'local_backup_script.sh "%p" "%f"'
787 Using a separate script file is advisable any time you want to use
788 more than a single command in the archiving process.
789 This allows all complexity to be managed within the script, which
790 can be written in a popular scripting language such as
791 <span class="application">bash</span> or <span class="application">perl</span>.
793 Examples of requirements that might be solved within a script include:
794 </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>
795 Copying data to secure off-site data storage
796 </p></li><li class="listitem"><p>
797 Batching WAL files so that they are transferred every three hours,
798 rather than one at a time
799 </p></li><li class="listitem"><p>
800 Interfacing with other backup and recovery software
801 </p></li><li class="listitem"><p>
802 Interfacing with monitoring software to report errors
803 </p></li></ul></div><p>
804 </p><div class="tip"><h3 class="title">Tip</h3><p>
805 When using an <code class="varname">archive_command</code> script, it's desirable
806 to enable <a class="xref" href="runtime-config-logging.html#GUC-LOGGING-COLLECTOR">logging_collector</a>.
807 Any messages written to <span class="systemitem">stderr</span> from the script will then
808 appear in the database server log, allowing complex configurations to
809 be diagnosed easily if they fail.
810 </p></div></div></div><div class="sect2" id="CONTINUOUS-ARCHIVING-CAVEATS"><div class="titlepage"><div><div><h3 class="title">25.3.8. Caveats <a href="#CONTINUOUS-ARCHIVING-CAVEATS" class="id_link">#</a></h3></div></div></div><p>
811 At this writing, there are several limitations of the continuous archiving
812 technique. These will probably be fixed in future releases:
814 </p><div class="itemizedlist"><ul class="itemizedlist" style="list-style-type: disc; "><li class="listitem"><p>
815 If a <a class="link" href="sql-createdatabase.html" title="CREATE DATABASE"><code class="command">CREATE DATABASE</code></a>
816 command is executed while a base backup is being taken, and then
817 the template database that the <code class="command">CREATE DATABASE</code> copied
818 is modified while the base backup is still in progress, it is
819 possible that recovery will cause those modifications to be
820 propagated into the created database as well. This is of course
821 undesirable. To avoid this risk, it is best not to modify any
822 template databases while taking a base backup.
823 </p></li><li class="listitem"><p>
824 <a class="link" href="sql-createtablespace.html" title="CREATE TABLESPACE"><code class="command">CREATE TABLESPACE</code></a>
825 commands are WAL-logged with the literal absolute path, and will
826 therefore be replayed as tablespace creations with the same
827 absolute path. This might be undesirable if the WAL is being
828 replayed on a different machine. It can be dangerous even if the
829 WAL is being replayed on the same machine, but into a new data
830 directory: the replay will still overwrite the contents of the
831 original tablespace. To avoid potential gotchas of this sort,
832 the best practice is to take a new base backup after creating or
833 dropping tablespaces.
834 </p></li></ul></div><p>
836 It should also be noted that the default <acronym class="acronym">WAL</acronym>
837 format is fairly bulky since it includes many disk page snapshots.
838 These page snapshots are designed to support crash recovery, since
839 we might need to fix partially-written disk pages. Depending on
840 your system hardware and software, the risk of partial writes might
841 be small enough to ignore, in which case you can significantly
842 reduce the total volume of archived WAL files by turning off page
843 snapshots using the <a class="xref" href="runtime-config-wal.html#GUC-FULL-PAGE-WRITES">full_page_writes</a>
844 parameter. (Read the notes and warnings in <a class="xref" href="wal.html" title="Chapter 28. Reliability and the Write-Ahead Log">Chapter 28</a>
845 before you do so.) Turning off page snapshots does not prevent
846 use of the WAL for PITR operations. An area for future
847 development is to compress archived WAL data by removing
848 unnecessary page copies even when <code class="varname">full_page_writes</code> is
849 on. In the meantime, administrators might wish to reduce the number
850 of page snapshots included in WAL by increasing the checkpoint
851 interval parameters as much as feasible.
852 </p></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="backup-file.html" title="25.2. File System Level Backup">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="backup.html" title="Chapter 25. Backup and Restore">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="high-availability.html" title="Chapter 26. High Availability, Load Balancing, and Replication">Next</a></td></tr><tr><td width="40%" align="left" valign="top">25.2. File System Level Backup </td><td width="20%" align="center"><a accesskey="h" href="index.html" title="PostgreSQL 18.0 Documentation">Home</a></td><td width="40%" align="right" valign="top"> Chapter 26. High Availability, Load Balancing, and Replication</td></tr></table></div></body></html>
begriffs.com / projects / ai-pg / blob