1 <!-- $PostgreSQL: pgsql/doc/src/sgml/backup.sgml,v 2.86 2006/09/16 00:30:11 momjian Exp $ -->
4 <title>Backup and Restore</title>
6 <indexterm zone="backup"><primary>backup</></>
9 As with everything that contains valuable data, <productname>PostgreSQL</>
10 databases should be backed up regularly. While the procedure is
11 essentially simple, it is important to have a basic understanding of
12 the underlying techniques and assumptions.
16 There are three fundamentally different approaches to backing up
17 <productname>PostgreSQL</> data:
19 <listitem><para><acronym>SQL</> dump</para></listitem>
20 <listitem><para>File system level backup</para></listitem>
21 <listitem><para>Continuous Archiving</para></listitem>
23 Each has its own strengths and weaknesses.
26 <sect1 id="backup-dump">
27 <title><acronym>SQL</> Dump</title>
30 The idea behind the SQL-dump method is to generate a text file with SQL
31 commands that, when fed back to the server, will recreate the
32 database in the same state as it was at the time of the dump.
33 <productname>PostgreSQL</> provides the utility program
34 <xref linkend="app-pgdump"> for this purpose. The basic usage of this
37 pg_dump <replaceable class="parameter">dbname</replaceable> > <replaceable class="parameter">outfile</replaceable>
39 As you see, <application>pg_dump</> writes its results to the
40 standard output. We will see below how this can be useful.
44 <application>pg_dump</> is a regular <productname>PostgreSQL</>
45 client application (albeit a particularly clever one). This means
46 that you can do this backup procedure from any remote host that has
47 access to the database. But remember that <application>pg_dump</>
48 does not operate with special permissions. In particular, it must
49 have read access to all tables that you want to back up, so in
50 practice you almost always have to run it as a database superuser.
54 To specify which database server <application>pg_dump</> should
55 contact, use the command line options <option>-h
56 <replaceable>host</></> and <option>-p <replaceable>port</></>. The
57 default host is the local host or whatever your
58 <envar>PGHOST</envar> environment variable specifies. Similarly,
59 the default port is indicated by the <envar>PGPORT</envar>
60 environment variable or, failing that, by the compiled-in default.
61 (Conveniently, the server will normally have the same compiled-in
66 As any other <productname>PostgreSQL</> client application,
67 <application>pg_dump</> will by default connect with the database
68 user name that is equal to the current operating system user name. To override
69 this, either specify the <option>-U</option> option or set the
70 environment variable <envar>PGUSER</envar>. Remember that
71 <application>pg_dump</> connections are subject to the normal
72 client authentication mechanisms (which are described in <xref
73 linkend="client-authentication">).
77 Dumps created by <application>pg_dump</> are internally consistent,
78 that is, updates to the database while <application>pg_dump</> is
79 running will not be in the dump. <application>pg_dump</> does not
80 block other operations on the database while it is working.
81 (Exceptions are those operations that need to operate with an
82 exclusive lock, such as <command>VACUUM FULL</command>.)
87 When your database schema relies on OIDs (for instance as foreign
88 keys) you must instruct <application>pg_dump</> to dump the OIDs
89 as well. To do this, use the <option>-o</option> command line
94 <sect2 id="backup-dump-restore">
95 <title>Restoring the dump</title>
98 The text files created by <application>pg_dump</> are intended to
99 be read in by the <application>psql</application> program. The
100 general command form to restore a dump is
102 psql <replaceable class="parameter">dbname</replaceable> < <replaceable class="parameter">infile</replaceable>
104 where <replaceable class="parameter">infile</replaceable> is what
105 you used as <replaceable class="parameter">outfile</replaceable>
106 for the <application>pg_dump</> command. The database <replaceable
107 class="parameter">dbname</replaceable> will not be created by this
108 command, you must create it yourself from <literal>template0</> before executing
109 <application>psql</> (e.g., with <literal>createdb -T template0
110 <replaceable class="parameter">dbname</></literal>).
111 <application>psql</> supports options similar to <application>pg_dump</>
112 for controlling the database server location and the user name. See
113 <xref linkend="app-psql">'s reference page for more information.
117 Not only must the target database already exist before starting to
118 run the restore, but so must all the users who own objects in the
119 dumped database or were granted permissions on the objects. If they
120 do not, then the restore will fail to recreate the objects with the
121 original ownership and/or permissions. (Sometimes this is what you want,
122 but usually it is not.)
126 Once restored, it is wise to run <xref linkend="sql-analyze"
127 endterm="sql-analyze-title"> on each database so the optimizer has
128 useful statistics. An easy way to do this is to run
129 <command>vacuumdb -a -z</> to
130 <command>VACUUM ANALYZE</> all databases; this is equivalent to
131 running <command>VACUUM ANALYZE</command> manually.
135 The ability of <application>pg_dump</> and <application>psql</> to
136 write to or read from pipes makes it possible to dump a database
137 directly from one server to another; for example:
139 pg_dump -h <replaceable>host1</> <replaceable>dbname</> | psql -h <replaceable>host2</> <replaceable>dbname</>
145 The dumps produced by <application>pg_dump</> are relative to
146 <literal>template0</>. This means that any languages, procedures,
147 etc. added to <literal>template1</> will also be dumped by
148 <application>pg_dump</>. As a result, when restoring, if you are
149 using a customized <literal>template1</>, you must create the
150 empty database from <literal>template0</>, as in the example
156 For advice on how to load large amounts of data into
157 <productname>PostgreSQL</productname> efficiently, refer to <xref
162 <sect2 id="backup-dump-all">
163 <title>Using <application>pg_dumpall</></title>
166 The above mechanism is cumbersome and inappropriate when backing
167 up an entire database cluster. For this reason the <xref
168 linkend="app-pg-dumpall"> program is provided.
169 <application>pg_dumpall</> backs up each database in a given
170 cluster, and also preserves cluster-wide data such as users and
171 groups. The basic usage of this command is:
173 pg_dumpall > <replaceable>outfile</>
175 The resulting dump can be restored with <application>psql</>:
177 psql -f <replaceable class="parameter">infile</replaceable> postgres
179 (Actually, you can specify any existing database name to start from,
180 but if you are reloading in an empty cluster then <literal>postgres</>
181 should generally be used.) It is always necessary to have
182 database superuser access when restoring a <application>pg_dumpall</>
183 dump, as that is required to restore the user and group information.
187 <sect2 id="backup-dump-large">
188 <title>Handling large databases</title>
191 Since <productname>PostgreSQL</productname> allows tables larger
192 than the maximum file size on your system, it can be problematic
193 to dump such a table to a file, since the resulting file will likely
194 be larger than the maximum size allowed by your system. Since
195 <application>pg_dump</> can write to the standard output, you can
196 just use standard Unix tools to work around this possible problem.
200 <title>Use compressed dumps.</title>
202 You can use your favorite compression program, for example
203 <application>gzip</application>.
206 pg_dump <replaceable class="parameter">dbname</replaceable> | gzip > <replaceable class="parameter">filename</replaceable>.gz
212 createdb <replaceable class="parameter">dbname</replaceable>
213 gunzip -c <replaceable class="parameter">filename</replaceable>.gz | psql <replaceable class="parameter">dbname</replaceable>
219 cat <replaceable class="parameter">filename</replaceable>.gz | gunzip | psql <replaceable class="parameter">dbname</replaceable>
225 <title>Use <command>split</>.</title>
227 The <command>split</command> command
228 allows you to split the output into pieces that are
229 acceptable in size to the underlying file system. For example, to
230 make chunks of 1 megabyte:
233 pg_dump <replaceable class="parameter">dbname</replaceable> | split -b 1m - <replaceable class="parameter">filename</replaceable>
239 createdb <replaceable class="parameter">dbname</replaceable>
240 cat <replaceable class="parameter">filename</replaceable>* | psql <replaceable class="parameter">dbname</replaceable>
246 <title>Use the custom dump format.</title>
248 If <productname>PostgreSQL</productname> was built on a system with the
249 <application>zlib</> compression library installed, the custom dump
250 format will compress data as it writes it to the output file. This will
251 produce dump file sizes similar to using <command>gzip</command>, but it
252 has the added advantage that tables can be restored selectively. The
253 following command dumps a database using the custom dump format:
256 pg_dump -Fc <replaceable class="parameter">dbname</replaceable> > <replaceable class="parameter">filename</replaceable>
259 A custom-format dump is not a script for <application>psql</>, but
260 instead must be restored with <application>pg_restore</>.
261 See the <xref linkend="app-pgdump"> and <xref
262 linkend="app-pgrestore"> reference pages for details.
269 <sect1 id="backup-file">
270 <title>File system level backup</title>
273 An alternative backup strategy is to directly copy the files that
274 <productname>PostgreSQL</> uses to store the data in the database. In
275 <xref linkend="creating-cluster"> it is explained where these files
276 are located, but you have probably found them already if you are
277 interested in this method. You can use whatever method you prefer
278 for doing usual file system backups, for example
281 tar -cf backup.tar /usr/local/pgsql/data
286 There are two restrictions, however, which make this method
287 impractical, or at least inferior to the <application>pg_dump</>
293 The database server <emphasis>must</> be shut down in order to
294 get a usable backup. Half-way measures such as disallowing all
295 connections will <emphasis>not</emphasis> work
296 (mainly because <command>tar</command> and similar tools do not take an
297 atomic snapshot of the state of the file system at a point in
298 time). Information about stopping the server can be found in
299 <xref linkend="server-shutdown">. Needless to say that you
300 also need to shut down the server before restoring the data.
306 If you have dug into the details of the file system layout of the
307 database, you may be tempted to try to back up or restore only certain
308 individual tables or databases from their respective files or
309 directories. This will <emphasis>not</> work because the
310 information contained in these files contains only half the
311 truth. The other half is in the commit log files
312 <filename>pg_clog/*</filename>, which contain the commit status of
313 all transactions. A table file is only usable with this
314 information. Of course it is also impossible to restore only a
315 table and the associated <filename>pg_clog</filename> data
316 because that would render all other tables in the database
317 cluster useless. So file system backups only work for complete
318 restoration of an entire database cluster.
325 An alternative file-system backup approach is to make a
326 <quote>consistent snapshot</quote> of the data directory, if the
327 file system supports that functionality (and you are willing to
328 trust that it is implemented correctly). The typical procedure is
329 to make a <quote>frozen snapshot</> of the volume containing the
330 database, then copy the whole data directory (not just parts, see
331 above) from the snapshot to a backup device, then release the frozen
332 snapshot. This will work even while the database server is running.
333 However, a backup created in this way saves
334 the database files in a state where the database server was not
335 properly shut down; therefore, when you start the database server
336 on the backed-up data, it will think the server had crashed
337 and replay the WAL log. This is not a problem, just be aware of
338 it (and be sure to include the WAL files in your backup).
342 If your database is spread across multiple file systems, there may not
343 be any way to obtain exactly-simultaneous frozen snapshots of all
344 the volumes. For example, if your data files and WAL log are on different
345 disks, or if tablespaces are on different file systems, it might
346 not be possible to use snapshot backup because the snapshots must be
348 Read your file system documentation very carefully before trusting
349 to the consistent-snapshot technique in such situations. The safest
350 approach is to shut down the database server for long enough to
351 establish all the frozen snapshots.
355 Another option is to use <application>rsync</> to perform a file
356 system backup. This is done by first running <application>rsync</>
357 while the database server is running, then shutting down the database
358 server just long enough to do a second <application>rsync</>. The
359 second <application>rsync</> will be much quicker than the first,
360 because it has relatively little data to transfer, and the end result
361 will be consistent because the server was down. This method
362 allows a file system backup to be performed with minimal downtime.
366 Note that a file system backup will not necessarily be
367 smaller than an SQL dump. On the contrary, it will most likely be
368 larger. (<application>pg_dump</application> does not need to dump
369 the contents of indexes for example, just the commands to recreate
374 <sect1 id="continuous-archiving">
375 <title>Continuous Archiving and Point-In-Time Recovery (PITR)</title>
377 <indexterm zone="backup">
378 <primary>continuous archiving</primary>
381 <indexterm zone="backup">
382 <primary>point-in-time recovery</primary>
385 <indexterm zone="backup">
386 <primary>PITR</primary>
390 At all times, <productname>PostgreSQL</> maintains a
391 <firstterm>write ahead log</> (WAL) in the <filename>pg_xlog/</>
392 subdirectory of the cluster's data directory. The log describes
393 every change made to the database's data files. This log exists
394 primarily for crash-safety purposes: if the system crashes, the
395 database can be restored to consistency by <quote>replaying</> the
396 log entries made since the last checkpoint. However, the existence
397 of the log makes it possible to use a third strategy for backing up
398 databases: we can combine a file-system-level backup with backup of
399 the WAL files. If recovery is needed, we restore the backup and
400 then replay from the backed-up WAL files to bring the backup up to
401 current time. This approach is more complex to administer than
402 either of the previous approaches, but it has some significant
407 We do not need a perfectly consistent backup as the starting point.
408 Any internal inconsistency in the backup will be corrected by log
409 replay (this is not significantly different from what happens during
410 crash recovery). So we don't need file system snapshot capability,
411 just <application>tar</> or a similar archiving tool.
416 Since we can string together an indefinitely long sequence of WAL files
417 for replay, continuous backup can be achieved simply by continuing to archive
418 the WAL files. This is particularly valuable for large databases, where
419 it may not be convenient to take a full backup frequently.
424 There is nothing that says we have to replay the WAL entries all the
425 way to the end. We could stop the replay at any point and have a
426 consistent snapshot of the database as it was at that time. Thus,
427 this technique supports <firstterm>point-in-time recovery</>: it is
428 possible to restore the database to its state at any time since your base
434 If we continuously feed the series of WAL files to another
435 machine that has been loaded with the same base backup file, we
436 have a <quote>hot standby</> system: at any point we can bring up
437 the second machine and it will have a nearly-current copy of the
445 As with the plain file-system-backup technique, this method can only
446 support restoration of an entire database cluster, not a subset.
447 Also, it requires a lot of archival storage: the base backup may be bulky,
448 and a busy system will generate many megabytes of WAL traffic that
449 have to be archived. Still, it is the preferred backup technique in
450 many situations where high reliability is needed.
454 To recover successfully using continuous archiving (also called "online
455 backup" by many database vendors), you need a continuous
456 sequence of archived WAL files that extends back at least as far as the
457 start time of your backup. So to get started, you should set up and test
458 your procedure for archiving WAL files <emphasis>before</> you take your
459 first base backup. Accordingly, we first discuss the mechanics of
463 <sect2 id="backup-archiving-wal">
464 <title>Setting up WAL archiving</title>
467 In an abstract sense, a running <productname>PostgreSQL</> system
468 produces an indefinitely long sequence of WAL records. The system
469 physically divides this sequence into WAL <firstterm>segment
470 files</>, which are normally 16MB apiece (although the size can be
471 altered when building <productname>PostgreSQL</>). The segment
472 files are given numeric names that reflect their position in the
473 abstract WAL sequence. When not using WAL archiving, the system
474 normally creates just a few segment files and then
475 <quote>recycles</> them by renaming no-longer-needed segment files
476 to higher segment numbers. It's assumed that a segment file whose
477 contents precede the checkpoint-before-last is no longer of
478 interest and can be recycled.
482 When archiving WAL data, we want to capture the contents of each segment
483 file once it is filled, and save that data somewhere before the segment
484 file is recycled for reuse. Depending on the application and the
485 available hardware, there could be many different ways of <quote>saving
486 the data somewhere</>: we could copy the segment files to an NFS-mounted
487 directory on another machine, write them onto a tape drive (ensuring that
488 you have a way of restoring the file with its original file name), or batch
489 them together and burn them onto CDs, or something else entirely. To
490 provide the database administrator with as much flexibility as possible,
491 <productname>PostgreSQL</> tries not to make any assumptions about how
492 the archiving will be done. Instead, <productname>PostgreSQL</> lets
493 the administrator specify a shell command to be executed to copy a
494 completed segment file to wherever it needs to go. The command could be
495 as simple as a <literal>cp</>, or it could invoke a complex shell
496 script — it's all up to you.
500 The shell command to use is specified by the <xref
501 linkend="guc-archive-command"> configuration parameter, which in practice
502 will always be placed in the <filename>postgresql.conf</filename> file.
504 any <literal>%p</> is replaced by the absolute path of the file to
505 archive, while any <literal>%f</> is replaced by the file name only.
506 Write <literal>%%</> if you need to embed an actual <literal>%</>
507 character in the command. The simplest useful command is something
510 archive_command = 'cp -i %p /mnt/server/archivedir/%f </dev/null'
512 which will copy archivable WAL segments to the directory
513 <filename>/mnt/server/archivedir</>. (This is an example, not a
514 recommendation, and may not work on all platforms.)
518 The archive command will be executed under the ownership of the same
519 user that the <productname>PostgreSQL</> server is running as. Since
520 the series of WAL files being archived contains effectively everything
521 in your database, you will want to be sure that the archived data is
522 protected from prying eyes; for example, archive into a directory that
523 does not have group or world read access.
527 It is important that the archive command return zero exit status if and
528 only if it succeeded. Upon getting a zero result,
529 <productname>PostgreSQL</> will assume that the WAL segment file has been
530 successfully archived, and will remove or recycle it.
531 However, a nonzero status tells
532 <productname>PostgreSQL</> that the file was not archived; it will try
533 again periodically until it succeeds.
537 The archive command should generally be designed to refuse to overwrite
538 any pre-existing archive file. This is an important safety feature to
539 preserve the integrity of your archive in case of administrator error
540 (such as sending the output of two different servers to the same archive
542 It is advisable to test your proposed archive command to ensure that it
543 indeed does not overwrite an existing file, <emphasis>and that it returns
544 nonzero status in this case</>. We have found that <literal>cp -i</> does
545 this correctly on some platforms but not others. If the chosen command
546 does not itself handle this case correctly, you should add a command
547 to test for pre-existence of the archive file. For example, something
550 archive_command = 'test ! -f .../%f && cp %p .../%f'
552 works correctly on most Unix variants.
556 While designing your archiving setup, consider what will happen if
557 the archive command fails repeatedly because some aspect requires
558 operator intervention or the archive runs out of space. For example, this
559 could occur if you write to tape without an autochanger; when the tape
560 fills, nothing further can be archived until the tape is swapped.
561 You should ensure that any error condition or request to a human operator
562 is reported appropriately so that the situation can be
563 resolved relatively quickly. The <filename>pg_xlog/</> directory will
564 continue to fill with WAL segment files until the situation is resolved.
568 The speed of the archiving command is not important, so long as it can keep up
569 with the average rate at which your server generates WAL data. Normal
570 operation continues even if the archiving process falls a little behind.
571 If archiving falls significantly behind, this will increase the amount of
572 data that would be lost in the event of a disaster. It will also mean that
573 the <filename>pg_xlog/</> directory will contain large numbers of
574 not-yet-archived segment files, which could eventually exceed available
575 disk space. You are advised to monitor the archiving process to ensure that
576 it is working as you intend.
580 In writing your archive command, you should assume that the file names to
581 be archived may be up to 64 characters long and may contain any
582 combination of ASCII letters, digits, and dots. It is not necessary to
583 remember the original full path (<literal>%p</>) but it is necessary to
584 remember the file name (<literal>%f</>).
588 Note that although WAL archiving will allow you to restore any
589 modifications made to the data in your <productname>PostgreSQL</> database
590 it will not restore changes made to configuration files (that is,
591 <filename>postgresql.conf</>, <filename>pg_hba.conf</> and
592 <filename>pg_ident.conf</>), since those are edited manually rather
593 than through SQL operations.
594 You may wish to keep the configuration files in a location that will
595 be backed up by your regular file system backup procedures. See
596 <xref linkend="runtime-config-file-locations"> for how to relocate the
601 The archive command is only invoked on completed WAL segments. Hence,
602 if your server generates only little WAL traffic (or has slack periods
603 where it does so), there could be a long delay between the completion
604 of a transaction and its safe recording in archive storage. To put
605 a limit on how old unarchived data can be, you can set
606 <xref linkend="guc-archive-timeout"> to force the server to switch
607 to a new WAL segment file at least that often. Note that archived
608 files that are ended early due to a forced switch are still the same
609 length as completely full files. It is therefore unwise to set a very
610 short <varname>archive_timeout</> — it will bloat your archive
611 storage. <varname>archive_timeout</> settings of a minute or so are
616 Also, you can force a segment switch manually with
617 <function>pg_switch_xlog()</>,
618 if you want to ensure that a just-finished transaction is archived
619 immediately. Other utility functions related to WAL management are
620 listed in <xref linkend="functions-admin-backup-table">.
624 <sect2 id="backup-base-backup">
625 <title>Making a Base Backup</title>
628 The procedure for making a base backup is relatively simple:
632 Ensure that WAL archiving is enabled and working.
637 Connect to the database as a superuser, and issue the command
639 SELECT pg_start_backup('label');
641 where <literal>label</> is any string you want to use to uniquely
642 identify this backup operation. (One good practice is to use the
643 full path where you intend to put the backup dump file.)
644 <function>pg_start_backup</> creates a <firstterm>backup label</> file,
645 called <filename>backup_label</>, in the cluster directory with
646 information about your backup.
650 It does not matter which database within the cluster you connect to to
651 issue this command. You can ignore the result returned by the function;
652 but if it reports an error, deal with that before proceeding.
657 Perform the backup, using any convenient file-system-backup tool
658 such as <application>tar</> or <application>cpio</>. It is neither
659 necessary nor desirable to stop normal operation of the database
665 Again connect to the database as a superuser, and issue the command
667 SELECT pg_stop_backup();
669 This should return successfully.
674 Once the WAL segment files used during the backup are archived as part
675 of normal database activity, you are done. The file identified by
676 <function>pg_stop_backup</>'s result is the last segment that needs
677 to be archived to complete the backup.
684 Some backup tools that you might wish to use emit warnings or errors
685 if the files they are trying to copy change while the copy proceeds.
686 This situation is normal, and not an error, when taking a base backup of
687 an active database; so you need to ensure that you can distinguish
688 complaints of this sort from real errors. For example, some versions
689 of <application>rsync</> return a separate exit code for <quote>vanished
690 source files</>, and you can write a driver script to accept this exit
691 code as a non-error case. Also,
692 some versions of GNU <application>tar</> consider it an error if a file
693 is changed while <application>tar</> is copying it. There does not seem
694 to be any very convenient way to distinguish this error from other types
695 of errors, other than manual inspection of <application>tar</>'s messages.
696 GNU <application>tar</> is therefore not the best tool for making base
701 It is not necessary to be very concerned about the amount of time elapsed
702 between <function>pg_start_backup</> and the start of the actual backup,
703 nor between the end of the backup and <function>pg_stop_backup</>; a
704 few minutes' delay won't hurt anything. You
705 must however be quite sure that these operations are carried out in
706 sequence and do not overlap.
710 Be certain that your backup dump includes all of the files underneath
711 the database cluster directory (e.g., <filename>/usr/local/pgsql/data</>).
712 If you are using tablespaces that do not reside underneath this directory,
713 be careful to include them as well (and be sure that your backup dump
714 archives symbolic links as links, otherwise the restore will mess up
719 You may, however, omit from the backup dump the files within the
720 <filename>pg_xlog/</> subdirectory of the cluster directory. This
721 slight complication is worthwhile because it reduces the risk
722 of mistakes when restoring. This is easy to arrange if
723 <filename>pg_xlog/</> is a symbolic link pointing to someplace outside
724 the cluster directory, which is a common setup anyway for performance
729 To make use of this backup, you will need to keep around all the WAL
730 segment files generated during and after the file system backup.
731 To aid you in doing this, the <function>pg_stop_backup</> function
732 creates a <firstterm>backup history file</> that is immediately
733 stored into the WAL archive area. This file is named after the first
734 WAL segment file that you need to have to make use of the backup.
735 For example, if the starting WAL file is
736 <literal>0000000100001234000055CD</> the backup history file will be
738 <literal>0000000100001234000055CD.007C9330.backup</>. (The second
739 number in the file name stands for an exact position within the WAL
740 file, and can ordinarily be ignored.) Once you have safely archived
741 the file system backup and the WAL segment files used during the
742 backup (as specified in the backup history file), all archived WAL
743 segments with names numerically less are no longer needed to recover
744 the file system backup and may be deleted. However, you should
745 consider keeping several backup sets to be absolutely certain that
746 you can recover your data.
750 The backup history file is just a small text file. It contains the
751 label string you gave to <function>pg_start_backup</>, as well as
752 the starting and ending times and WAL segments of the backup.
753 If you used the label to identify where the associated dump file is kept,
754 then the archived history file is enough to tell you which dump file to
755 restore, should you need to do so.
759 Since you have to keep around all the archived WAL files back to your
760 last base backup, the interval between base backups should usually be
761 chosen based on how much storage you want to expend on archived WAL
762 files. You should also consider how long you are prepared to spend
763 recovering, if recovery should be necessary — the system will have to
764 replay all those WAL segments, and that could take awhile if it has
765 been a long time since the last base backup.
769 It's also worth noting that the <function>pg_start_backup</> function
770 makes a file named <filename>backup_label</> in the database cluster
771 directory, which is then removed again by <function>pg_stop_backup</>.
772 This file will of course be archived as a part of your backup dump file.
773 The backup label file includes the label string you gave to
774 <function>pg_start_backup</>, as well as the time at which
775 <function>pg_start_backup</> was run, and the name of the starting WAL
776 file. In case of confusion it will
777 therefore be possible to look inside a backup dump file and determine
778 exactly which backup session the dump file came from.
782 It is also possible to make a backup dump while the server is
783 stopped. In this case, you obviously cannot use
784 <function>pg_start_backup</> or <function>pg_stop_backup</>, and
785 you will therefore be left to your own devices to keep track of which
786 backup dump is which and how far back the associated WAL files go.
787 It is generally better to follow the continuous archiving procedure above.
791 <sect2 id="backup-pitr-recovery">
792 <title>Recovering using a Continuous Archive Backup</title>
795 Okay, the worst has happened and you need to recover from your backup.
796 Here is the procedure:
800 Stop the server, if it's running.
805 If you have the space to do so,
806 copy the whole cluster data directory and any tablespaces to a temporary
807 location in case you need them later. Note that this precaution will
808 require that you have enough free space on your system to hold two
809 copies of your existing database. If you do not have enough space,
810 you need at the least to copy the contents of the <filename>pg_xlog</>
811 subdirectory of the cluster data directory, as it may contain logs which
812 were not archived before the system went down.
817 Clean out all existing files and subdirectories under the cluster data
818 directory and under the root directories of any tablespaces you are using.
823 Restore the database files from your backup dump. Be careful that they
824 are restored with the right ownership (the database system user, not
825 root!) and with the right permissions. If you are using tablespaces,
826 you may want to verify that the symbolic links in <filename>pg_tblspc/</>
827 were correctly restored.
832 Remove any files present in <filename>pg_xlog/</>; these came from the
833 backup dump and are therefore probably obsolete rather than current.
834 If you didn't archive <filename>pg_xlog/</> at all, then re-create it,
835 and be sure to re-create the subdirectory
836 <filename>pg_xlog/archive_status/</> as well.
841 If you had unarchived WAL segment files that you saved in step 2,
842 copy them into <filename>pg_xlog/</>. (It is best to copy them,
843 not move them, so that you still have the unmodified files if a
844 problem occurs and you have to start over.)
849 Create a recovery command file <filename>recovery.conf</> in the cluster
850 data directory (see <xref linkend="recovery-config-settings">). You may
851 also want to temporarily modify <filename>pg_hba.conf</> to prevent
852 ordinary users from connecting until you are sure the recovery has worked.
857 Start the server. The server will go into recovery mode and
858 proceed to read through the archived WAL files it needs. Should the
859 recovery be terminated because of an external error, the server can
860 simply be restarted and it will continue recovery. Upon completion
861 of the recovery process, the server will rename
862 <filename>recovery.conf</> to <filename>recovery.done</> (to prevent
863 accidentally re-entering recovery mode in case of a crash later) and then
864 commence normal database operations.
869 Inspect the contents of the database to ensure you have recovered to
870 where you want to be. If not, return to step 1. If all is well,
871 let in your users by restoring <filename>pg_hba.conf</> to normal.
878 The key part of all this is to set up a recovery command file that
879 describes how you want to recover and how far the recovery should
880 run. You can use <filename>recovery.conf.sample</> (normally
881 installed in the installation <filename>share/</> directory) as a
882 prototype. The one thing that you absolutely must specify in
883 <filename>recovery.conf</> is the <varname>restore_command</>,
884 which tells <productname>PostgreSQL</> how to get back archived
885 WAL file segments. Like the <varname>archive_command</>, this is
886 a shell command string. It may contain <literal>%f</>, which is
887 replaced by the name of the desired log file, and <literal>%p</>,
888 which is replaced by the absolute path to copy the log file to.
889 Write <literal>%%</> if you need to embed an actual <literal>%</>
890 character in the command. The simplest useful command is
893 restore_command = 'cp /mnt/server/archivedir/%f %p'
895 which will copy previously archived WAL segments from the directory
896 <filename>/mnt/server/archivedir</>. You could of course use something
897 much more complicated, perhaps even a shell script that requests the
898 operator to mount an appropriate tape.
902 It is important that the command return nonzero exit status on failure.
903 The command <emphasis>will</> be asked for log files that are not present
904 in the archive; it must return nonzero when so asked. This is not an
905 error condition. Be aware also that the base name of the <literal>%p</>
906 path will be different from <literal>%f</>; do not expect them to be
911 WAL segments that cannot be found in the archive will be sought in
912 <filename>pg_xlog/</>; this allows use of recent un-archived segments.
913 However segments that are available from the archive will be used in
914 preference to files in <filename>pg_xlog/</>. The system will not
915 overwrite the existing contents of <filename>pg_xlog/</> when retrieving
920 Normally, recovery will proceed through all available WAL segments,
921 thereby restoring the database to the current point in time (or as
922 close as we can get given the available WAL segments). But if you want
923 to recover to some previous point in time (say, right before the junior
924 DBA dropped your main transaction table), just specify the required
925 stopping point in <filename>recovery.conf</>. You can specify the stop
926 point, known as the <quote>recovery target</>, either by date/time or
927 by completion of a specific transaction ID. As of this writing only
928 the date/time option is very usable, since there are no tools to help
929 you identify with any accuracy which transaction ID to use.
934 The stop point must be after the ending time of the base backup (the
935 time of <function>pg_stop_backup</>). You cannot use a base backup
936 to recover to a time when that backup was still going on. (To
937 recover to such a time, you must go back to your previous base backup
938 and roll forward from there.)
943 If recovery finds a corruption in the WAL data then recovery will
944 complete at that point and the server will not start. The recovery
945 process could be re-run from the beginning, specifying a
946 <quote>recovery target</> so that recovery can complete normally.
947 If recovery fails for an external reason, such as a system crash or
948 the WAL archive has become inaccessible, then the recovery can be
949 simply restarted and it will restart almost from where it failed.
950 Restartable recovery works by writing a restartpoint record to the control
951 file at the first safely usable checkpoint record found after
952 <varname>checkpoint_timeout</> seconds.
956 <sect3 id="recovery-config-settings" xreflabel="Recovery Settings">
957 <title>Recovery Settings</title>
960 These settings can only be made in the <filename>recovery.conf</>
961 file, and apply only for the duration of the recovery. They must be
962 reset for any subsequent recovery you wish to perform. They cannot be
963 changed once recovery has begun.
968 <varlistentry id="restore-command" xreflabel="restore_command">
969 <term><varname>restore_command</varname> (<type>string</type>)</term>
972 The shell command to execute to retrieve an archived segment of
973 the WAL file series. This parameter is required.
974 Any <literal>%f</> in the string is
975 replaced by the name of the file to retrieve from the archive,
976 and any <literal>%p</> is replaced by the absolute path to copy
978 Write <literal>%%</> to embed an actual <literal>%</> character
982 It is important for the command to return a zero exit status if and
983 only if it succeeds. The command <emphasis>will</> be asked for file
984 names that are not present in the archive; it must return nonzero
985 when so asked. Examples:
987 restore_command = 'cp /mnt/server/archivedir/%f "%p"'
988 restore_command = 'copy /mnt/server/archivedir/%f "%p"' # Windows
994 <varlistentry id="recovery-target-time" xreflabel="recovery_target_time">
995 <term><varname>recovery_target_time</varname>
996 (<type>timestamp</type>)
1000 This parameter specifies the time stamp up to which recovery
1002 At most one of <varname>recovery_target_time</> and
1003 <xref linkend="recovery-target-xid"> can be specified.
1004 The default is to recover to the end of the WAL log.
1005 The precise stopping point is also influenced by
1006 <xref linkend="recovery-target-inclusive">.
1011 <varlistentry id="recovery-target-xid" xreflabel="recovery_target_xid">
1012 <term><varname>recovery_target_xid</varname> (<type>string</type>)</term>
1015 This parameter specifies the transaction ID up to which recovery
1016 will proceed. Keep in mind
1017 that while transaction IDs are assigned sequentially at transaction
1018 start, transactions can complete in a different numeric order.
1019 The transactions that will be recovered are those that committed
1020 before (and optionally including) the specified one.
1021 At most one of <varname>recovery_target_xid</> and
1022 <xref linkend="recovery-target-time"> can be specified.
1023 The default is to recover to the end of the WAL log.
1024 The precise stopping point is also influenced by
1025 <xref linkend="recovery-target-inclusive">.
1030 <varlistentry id="recovery-target-inclusive"
1031 xreflabel="recovery_target_inclusive">
1032 <term><varname>recovery_target_inclusive</varname>
1033 (<type>boolean</type>)
1037 Specifies whether we stop just after the specified recovery target
1038 (<literal>true</literal>), or just before the recovery target
1039 (<literal>false</literal>).
1040 Applies to both <xref linkend="recovery-target-time">
1041 and <xref linkend="recovery-target-xid">, whichever one is
1042 specified for this recovery. This indicates whether transactions
1043 having exactly the target commit time or ID, respectively, will
1044 be included in the recovery. Default is <literal>true</>.
1049 <varlistentry id="recovery-target-timeline"
1050 xreflabel="recovery_target_timeline">
1051 <term><varname>recovery_target_timeline</varname>
1052 (<type>string</type>)
1056 Specifies recovering into a particular timeline. The default is
1057 to recover along the same timeline that was current when the
1058 base backup was taken. You would only need to set this parameter
1059 in complex re-recovery situations, where you need to return to
1060 a state that itself was reached after a point-in-time recovery.
1061 See <xref linkend="backup-timelines"> for discussion.
1072 <sect2 id="backup-timelines">
1073 <title>Timelines</title>
1075 <indexterm zone="backup">
1076 <primary>timelines</primary>
1080 The ability to restore the database to a previous point in time creates
1081 some complexities that are akin to science-fiction stories about time
1082 travel and parallel universes. In the original history of the database,
1083 perhaps you dropped a critical table at 5:15PM on Tuesday evening.
1084 Unfazed, you get out your backup, restore to the point-in-time 5:14PM
1085 Tuesday evening, and are up and running. In <emphasis>this</> history of
1086 the database universe, you never dropped the table at all. But suppose
1087 you later realize this wasn't such a great idea after all, and would like
1088 to return to some later point in the original history. You won't be able
1089 to if, while your database was up-and-running, it overwrote some of the
1090 sequence of WAL segment files that led up to the time you now wish you
1091 could get back to. So you really want to distinguish the series of
1092 WAL records generated after you've done a point-in-time recovery from
1093 those that were generated in the original database history.
1097 To deal with these problems, <productname>PostgreSQL</> has a notion
1098 of <firstterm>timelines</>. Each time you recover to a point-in-time
1099 earlier than the end of the WAL sequence, a new timeline is created
1100 to identify the series of WAL records generated after that recovery.
1101 (If recovery proceeds all the way to the end of WAL, however, we do not
1102 start a new timeline: we just extend the existing one.) The timeline
1103 ID number is part of WAL segment file names, and so a new timeline does
1104 not overwrite the WAL data generated by previous timelines. It is
1105 in fact possible to archive many different timelines. While that might
1106 seem like a useless feature, it's often a lifesaver. Consider the
1107 situation where you aren't quite sure what point-in-time to recover to,
1108 and so have to do several point-in-time recoveries by trial and error
1109 until you find the best place to branch off from the old history. Without
1110 timelines this process would soon generate an unmanageable mess. With
1111 timelines, you can recover to <emphasis>any</> prior state, including
1112 states in timeline branches that you later abandoned.
1116 Each time a new timeline is created, <productname>PostgreSQL</> creates
1117 a <quote>timeline history</> file that shows which timeline it branched
1118 off from and when. These history files are necessary to allow the system
1119 to pick the right WAL segment files when recovering from an archive that
1120 contains multiple timelines. Therefore, they are archived into the WAL
1121 archive area just like WAL segment files. The history files are just
1122 small text files, so it's cheap and appropriate to keep them around
1123 indefinitely (unlike the segment files which are large). You can, if
1124 you like, add comments to a history file to make your own notes about
1125 how and why this particular timeline came to be. Such comments will be
1126 especially valuable when you have a thicket of different timelines as
1127 a result of experimentation.
1131 The default behavior of recovery is to recover along the same timeline
1132 that was current when the base backup was taken. If you want to recover
1133 into some child timeline (that is, you want to return to some state that
1134 was itself generated after a recovery attempt), you need to specify the
1135 target timeline ID in <filename>recovery.conf</>. You cannot recover into
1136 timelines that branched off earlier than the base backup.
1140 <sect2 id="continuous-archiving-caveats">
1141 <title>Caveats</title>
1144 At this writing, there are several limitations of the continuous archiving
1145 technique. These will probably be fixed in future releases:
1150 Operations on hash indexes are
1151 not presently WAL-logged, so replay will not update these indexes.
1152 The recommended workaround is to manually <command>REINDEX</> each
1153 such index after completing a recovery operation.
1159 If a <command>CREATE DATABASE</> command is executed while a base
1160 backup is being taken, and then the template database that the
1161 <command>CREATE DATABASE</> copied is modified while the base backup
1162 is still in progress, it is possible that recovery will cause those
1163 modifications to be propagated into the created database as well.
1164 This is of course undesirable. To avoid this risk, it is best not to
1165 modify any template databases while taking a base backup.
1171 <command>CREATE TABLESPACE</> commands are WAL-logged with the literal
1172 absolute path, and will therefore be replayed as tablespace creations
1173 with the same absolute path. This might be undesirable if the log is
1174 being replayed on a different machine. It can be dangerous even if
1175 the log is being replayed on the same machine, but into a new data
1176 directory: the replay will still overwrite the contents of the original
1177 tablespace. To avoid potential gotchas of this sort, the best practice
1178 is to take a new base backup after creating or dropping tablespaces.
1185 It should also be noted that the default <acronym>WAL</acronym>
1186 format is fairly bulky since it includes many disk page snapshots.
1187 These page snapshots are designed to support crash recovery,
1188 since we may need to fix partially-written disk pages. Depending
1189 on your system hardware and software, the risk of partial writes may
1190 be small enough to ignore, in which case you can significantly reduce
1191 the total volume of archived logs by turning off page snapshots
1192 using the <xref linkend="guc-full-page-writes"> parameter.
1193 (Read the notes and warnings in
1194 <xref linkend="wal"> before you do so.)
1195 Turning off page snapshots does not prevent use of the logs for PITR
1197 An area for future development is to compress archived WAL data by
1198 removing unnecessary page copies even when <varname>full_page_writes</>
1199 is on. In the meantime, administrators
1200 may wish to reduce the number of page snapshots included in WAL by
1201 increasing the checkpoint interval parameters as much as feasible.
1206 <sect1 id="warm-standby">
1207 <title>Warm Standby Servers for High Availability</title>
1209 <indexterm zone="backup">
1210 <primary>Warm Standby</primary>
1213 <indexterm zone="backup">
1214 <primary>PITR Standby</primary>
1217 <indexterm zone="backup">
1218 <primary>Standby Server</primary>
1221 <indexterm zone="backup">
1222 <primary>Log Shipping</primary>
1225 <indexterm zone="backup">
1226 <primary>Witness Server</primary>
1229 <indexterm zone="backup">
1230 <primary>STONITH</primary>
1233 <indexterm zone="backup">
1234 <primary>High Availability</primary>
1238 Continuous Archiving can be used to create a High Availability (HA)
1239 cluster configuration with one or more Standby Servers ready to take
1240 over operations in the case that the Primary Server fails. This
1241 capability is more widely known as Warm Standby Log Shipping.
1245 The Primary and Standby Server work together to provide this capability,
1246 though the servers are only loosely coupled. The Primary Server operates
1247 in Continuous Archiving mode, while the Standby Server operates in a
1248 continuous Recovery mode, reading the WAL files from the Primary. No
1249 changes to the database tables are required to enable this capability,
1250 so it offers a low administration overhead in comparison with other
1251 replication approaches. This configuration also has a very low
1252 performance impact on the Primary server.
1256 Directly moving WAL or "log" records from one database server to another
1257 is typically described as Log Shipping. PostgreSQL implements file-based
1258 Log Shipping, meaning WAL records are batched one file at a time. WAL
1259 files can be shipped easily and cheaply over any distance, whether it be
1260 to an adjacent system, another system on the same site or another system
1261 on the far side of the globe. The bandwidth required for this technique
1262 varies according to the transaction rate of the Primary Server.
1263 Record-based Log Shipping is also possible with custom-developed
1264 procedures, discussed in a later section. Future developments are likely
1265 to include options for synchronous and/or integrated record-based log
1270 It should be noted that the log shipping is asynchronous, i.e. the WAL
1271 records are shipped after transaction commit. As a result there can be a
1272 small window of data loss, should the Primary Server suffer a
1273 catastrophic failure. The window of data loss is minimised by the use of
1274 the archive_timeout parameter, which can be set as low as a few seconds
1275 if required. A very low setting can increase the bandwidth requirements
1280 The Standby server is not available for access, since it is continually
1281 performing recovery processing. Recovery performance is sufficiently
1282 good that the Standby will typically be only minutes away from full
1283 availability once it has been activated. As a result, we refer to this
1284 capability as a Warm Standby configuration that offers High
1285 Availability. Restoring a server from an archived base backup and
1286 rollforward can take considerably longer and so that technique only
1287 really offers a solution for Disaster Recovery, not HA.
1291 Other mechanisms for High Availability replication are available, both
1292 commercially and as open-source software.
1296 In general, log shipping between servers running different release
1297 levels will not be possible. It is the policy of the PostgreSQL Worldwide
1298 Development Group not to make changes to disk formats during minor release
1299 upgrades, so it is likely that running different minor release levels
1300 on Primary and Standby servers will work successfully. However, no
1301 formal support for that is offered and you are advised not to allow this
1302 to occur over long periods.
1305 <sect2 id="warm-standby-planning">
1306 <title>Planning</title>
1309 On the Standby server all tablespaces and paths will refer to similarly
1310 named mount points, so it is important to create the Primary and Standby
1311 servers so that they are as similar as possible, at least from the
1312 perspective of the database server. Furthermore, any CREATE TABLESPACE
1313 commands will be passed across as-is, so any new mount points must be
1314 created on both servers before they are used on the Primary. Hardware
1315 need not be the same, but experience shows that maintaining two
1316 identical systems is easier than maintaining two dissimilar ones over
1317 the whole lifetime of the application and system.
1321 There is no special mode required to enable a Standby server. The
1322 operations that occur on both Primary and Standby servers are entirely
1323 normal continuous archiving and recovery tasks. The primary point of
1324 contact between the two database servers is the archive of WAL files
1325 that both share: Primary writing to the archive, Standby reading from
1326 the archive. Care must be taken to ensure that WAL archives for separate
1327 servers do not become mixed together or confused.
1331 The magic that makes the two loosely coupled servers work together is
1332 simply a restore_command that waits for the next WAL file to be archived
1333 from the Primary. The restore_command is specified in the recovery.conf
1334 file on the Standby Server. Normal recovery processing would request a
1335 file from the WAL archive, causing an error if the file was unavailable.
1336 For Standby processing it is normal for the next file to be unavailable,
1337 so we must be patient and wait for it to appear. A waiting
1338 restore_command can be written as a custom script that loops after
1339 polling for the existence of the next WAL file. There must also be some
1340 way to trigger failover, which should interrupt the restore_command,
1341 break the loop and return a file not found error to the Standby Server.
1342 This then ends recovery and the Standby will then come up as a normal
1347 Sample code for the C version of the restore_command would be be:
1350 while (!NextWALFileReady() && !triggered)
1352 sleep(100000L); // wait for ~0.1 sec
1353 if (CheckForExternalTrigger())
1357 CopyWALFileForRecovery();
1362 PostgreSQL does not provide the system software required to identify a
1363 failure on the Primary and notify the Standby system and then the
1364 Standby database server. Many such tools exist and are well integrated
1365 with other aspects of a system failover, such as ip address migration.
1369 Triggering failover is an important part of planning and design. The
1370 restore_command is executed in full once for each WAL file. The process
1371 running the restore_command is therefore created and dies for each file,
1372 so there is no daemon or server process and so we cannot use signals and
1373 a signal handler. A more permanent notification is required to trigger
1374 the failover. It is possible to use a simple timeout facility,
1375 especially if used in conjunction with a known archive_timeout setting
1376 on the Primary. This is somewhat error prone since a network or busy
1377 Primary server might be sufficient to initiate failover. A notification
1378 mechanism such as the explicit creation of a trigger file is less error
1379 prone, if this can be arranged.
1383 <sect2 id="warm-standby-config">
1384 <title>Implementation</title>
1387 The short procedure for configuring a Standby Server is as follows. For
1388 full details of each step, refer to previous sections as noted.
1392 Set up Primary and Standby systems as near identically as possible,
1393 including two identical copies of PostgreSQL at same release level.
1398 Set up Continuous Archiving from the Primary to a WAL archive located
1399 in a directory on the Standby Server. Ensure that both <xref
1400 linkend="guc-archive-command"> and <xref linkend="guc-archive-timeout">
1401 are set. (See <xref linkend="backup-archiving-wal">)
1406 Make a Base Backup of the Primary Server. (See <xref
1407 linkend="backup-base-backup">)
1412 Begin recovery on the Standby Server from the local WAL archive,
1413 using a recovery.conf that specifies a restore_command that waits as
1414 described previously. (See <xref linkend="backup-pitr-recovery">)
1421 Recovery treats the WAL Archive as read-only, so once a WAL file has
1422 been copied to the Standby system it can be copied to tape at the same
1423 time as it is being used by the Standby database server to recover.
1424 Thus, running a Standby Server for High Availability can be performed at
1425 the same time as files are stored for longer term Disaster Recovery
1430 For testing purposes, it is possible to run both Primary and Standby
1431 servers on the same system. This does not provide any worthwhile
1432 improvement on server robustness, nor would it be described as HA.
1436 <sect2 id="warm-standby-failover">
1437 <title>Failover</title>
1440 If the Primary Server fails then the Standby Server should take begin
1441 failover procedures.
1445 If the Standby Server fails then no failover need take place. If the
1446 Standby Server can be restarted, then the recovery process can also be
1447 immediately restarted, taking advantage of Restartable Recovery.
1451 If the Primary Server fails and then immediately restarts, you must have
1452 a mechanism for informing it that it is no longer the Primary. This is
1453 sometimes known as STONITH (Should the Other Node In The Head), which is
1454 necessary to avoid situations where both systems think they are the
1455 Primary, which can lead to confusion and ultimately data loss.
1459 Many failover systems use just two systems, the Primary and the Standby,
1460 connected by some kind of heartbeat mechanism to continually verify the
1461 connectivity between the two and the viability of the Primary. It is
1462 also possible to use a third system, known as a Witness Server to avoid
1463 some problems of inappropriate failover, but the additional complexity
1464 may not be worthwhile unless it is set-up with sufficient care and
1469 At the instant that failover takes place to the Standby, we have only a
1470 single server in operation. This is known as a degenerate state.
1471 The former Standby is now the Primary, but the former Primary is down
1472 and may stay down. We must now fully re-create a Standby server,
1473 either on the former Primary system when it comes up, or on a third,
1474 possibly new, system. Once complete the Primary and Standby can be
1475 considered to have switched roles. Some people choose to use a third
1476 server to provide additional protection across the failover interval,
1477 though clearly this complicates the system configuration and
1478 operational processes (and this can also act as a Witness Server).
1482 So, switching from Primary to Standby Server can be fast, but requires
1483 some time to re-prepare the failover cluster. Regular switching from
1484 Primary to Standby is encouraged, since it allows the regular downtime
1485 one each system required to maintain HA. This also acts as a test of the
1486 failover so that it definitely works when you really need it. Written
1487 administration procedures are advised.
1491 <sect2 id="warm-standby-record">
1492 <title>Implementing Record-based Log Shipping</title>
1495 The main features for Log Shipping in this release are based around the
1496 file-based Log Shipping described above. It is also possible to
1497 implement record-based Log Shipping using the pg_xlogfile_name_offset()
1498 function, though this requires custom development.
1502 An external program can call pg_xlogfile_name_offset() to find out the
1503 filename and the exact byte offset within it of the latest WAL pointer.
1504 If the external program regularly polls the server it can find out how
1505 far forward the pointer has moved. It can then access the WAL file
1506 directly and copy those bytes across to a less up-to-date copy on a
1512 <sect1 id="migration">
1513 <title>Migration Between Releases</title>
1515 <indexterm zone="migration">
1516 <primary>upgrading</primary>
1519 <indexterm zone="migration">
1520 <primary>version</primary>
1521 <secondary>compatibility</secondary>
1525 This section discusses how to migrate your database data from one
1526 <productname>PostgreSQL</> release to a newer one.
1527 The software installation procedure <foreignphrase>per se</> is not the
1528 subject of this section; those details are in <xref linkend="installation">.
1532 As a general rule, the internal data storage format is subject to
1533 change between major releases of <productname>PostgreSQL</> (where
1534 the number after the first dot changes). This does not apply to
1535 different minor releases under the same major release (where the
1536 number after the second dot changes); these always have compatible
1537 storage formats. For example, releases 7.2.1, 7.3.2, and 7.4 are
1538 not compatible, whereas 7.2.1 and 7.2.2 are. When you update
1539 between compatible versions, you can simply replace the executables
1540 and reuse the data directory on disk. Otherwise you need to back
1541 up your data and restore it on the new server. This has to be done
1542 using <application>pg_dump</>; file system level backup methods
1543 obviously won't work. There are checks in place that prevent you
1544 from using a data directory with an incompatible version of
1545 <productname>PostgreSQL</productname>, so no great harm can be done by
1546 trying to start the wrong server version on a data directory.
1550 It is recommended that you use the <application>pg_dump</> and
1551 <application>pg_dumpall</> programs from the newer version of
1552 <productname>PostgreSQL</>, to take advantage of any enhancements
1553 that may have been made in these programs. Current releases of the
1554 dump programs can read data from any server version back to 7.0.
1558 The least downtime can be achieved by installing the new server in
1559 a different directory and running both the old and the new servers
1560 in parallel, on different ports. Then you can use something like
1563 pg_dumpall -p 5432 | psql -d postgres -p 6543
1566 to transfer your data. Or use an intermediate file if you want.
1567 Then you can shut down the old server and start the new server at
1568 the port the old one was running at. You should make sure that the
1569 old database is not updated after you run <application>pg_dumpall</>,
1570 otherwise you will obviously lose that data. See <xref
1571 linkend="client-authentication"> for information on how to prohibit
1576 In practice you probably want to test your client
1577 applications on the new setup before switching over completely.
1578 This is another reason for setting up concurrent installations
1579 of old and new versions.
1583 If you cannot or do not want to run two servers in parallel you can
1584 do the backup step before installing the new version, bring down
1585 the server, move the old version out of the way, install the new
1586 version, start the new server, restore the data. For example:
1589 pg_dumpall > backup
1591 mv /usr/local/pgsql /usr/local/pgsql.old
1592 cd ~/postgresql-&version;
1594 initdb -D /usr/local/pgsql/data
1595 postgres -D /usr/local/pgsql/data
1596 psql -f backup postgres
1599 See <xref linkend="runtime"> about ways to start and stop the
1600 server and other details. The installation instructions will advise
1601 you of strategic places to perform these steps.
1606 When you <quote>move the old installation out of the way</quote>
1607 it may no longer be perfectly usable. Some of the executable programs
1608 contain absolute paths to various installed programs and data files.
1609 This is usually not a big problem but if you plan on using two
1610 installations in parallel for a while you should assign them
1611 different installation directories at build time. (This problem
1612 is rectified in <productname>PostgreSQL</> 8.0 and later, but you
1613 need to be wary of moving older installations.)