2 25.2. File System Level Backup #
4 An alternative backup strategy is to directly copy the files that
5 PostgreSQL uses to store the data in the database; Section 18.2
6 explains where these files are located. You can use whatever method you
7 prefer for doing file system backups; for example:
8 tar -cf backup.tar /usr/local/pgsql/data
10 There are two restrictions, however, which make this method
11 impractical, or at least inferior to the pg_dump method:
12 1. The database server must be shut down in order to get a usable
13 backup. Half-way measures such as disallowing all connections will
14 not work (in part because tar and similar tools do not take an
15 atomic snapshot of the state of the file system, but also because
16 of internal buffering within the server). Information about
17 stopping the server can be found in Section 18.5. Needless to say,
18 you also need to shut down the server before restoring the data.
19 2. If you have dug into the details of the file system layout of the
20 database, you might be tempted to try to back up or restore only
21 certain individual tables or databases from their respective files
22 or directories. This will not work because the information
23 contained in these files is not usable without the commit log
24 files, pg_xact/*, which contain the commit status of all
25 transactions. A table file is only usable with this information. Of
26 course it is also impossible to restore only a table and the
27 associated pg_xact data because that would render all other tables
28 in the database cluster useless. So file system backups only work
29 for complete backup and restoration of an entire database cluster.
31 An alternative file-system backup approach is to make a “consistent
32 snapshot” of the data directory, if the file system supports that
33 functionality (and you are willing to trust that it is implemented
34 correctly). The typical procedure is to make a “frozen snapshot” of the
35 volume containing the database, then copy the whole data directory (not
36 just parts, see above) from the snapshot to a backup device, then
37 release the frozen snapshot. This will work even while the database
38 server is running. However, a backup created in this way saves the
39 database files in a state as if the database server was not properly
40 shut down; therefore, when you start the database server on the
41 backed-up data, it will think the previous server instance crashed and
42 will replay the WAL log. This is not a problem; just be aware of it
43 (and be sure to include the WAL files in your backup). You can perform
44 a CHECKPOINT before taking the snapshot to reduce recovery time.
46 If your database is spread across multiple file systems, there might
47 not be any way to obtain exactly-simultaneous frozen snapshots of all
48 the volumes. For example, if your data files and WAL log are on
49 different disks, or if tablespaces are on different file systems, it
50 might not be possible to use snapshot backup because the snapshots must
51 be simultaneous. Read your file system documentation very carefully
52 before trusting the consistent-snapshot technique in such situations.
54 If simultaneous snapshots are not possible, one option is to shut down
55 the database server long enough to establish all the frozen snapshots.
56 Another option is to perform a continuous archiving base backup
57 (Section 25.3.2) because such backups are immune to file system changes
58 during the backup. This requires enabling continuous archiving just
59 during the backup process; restore is done using continuous archive
60 recovery (Section 25.3.5).
62 Another option is to use rsync to perform a file system backup. This is
63 done by first running rsync while the database server is running, then
64 shutting down the database server long enough to do an rsync
65 --checksum. (--checksum is necessary because rsync only has file
66 modification-time granularity of one second.) The second rsync will be
67 quicker than the first, because it has relatively little data to
68 transfer, and the end result will be consistent because the server was
69 down. This method allows a file system backup to be performed with
72 Note that a file system backup will typically be larger than an SQL
73 dump. (pg_dump does not need to dump the contents of indexes for
74 example, just the commands to recreate them.) However, taking a file
75 system backup might be faster.