risks of backup using filesystem snapshot/dump ?

[Jim Mercer]

hi,

i'm running pgsql 9.1.3 (soon to be upgraded to 9.2.x) on FreeBSD 8-STABLE.

i've got a db of some 550GB that i want to backup on a daily basis, as we
have had some nasty hardware/power/other issues of late.

with dump on FreeBSD, you can specify -L, which does a filesystem level
snapshot before the dump starts, and then clears it afterwards.

i did a backup using dump on the filesystems, except for $PGDATA, followed
by a pg_dumpall.
this took 4 hours, and ate some 30GB (after compression)

i'm currently doing a dump of the filesystems, including $PGDATA.
this is looking like it will take 6-7 hours, and looks like it will take
some 40-50GB (after compression)

if i were to attempt a restore by replaying the 'pg_dumpall', i suspect that
i would be looking at many hours, and possibly hiccups requiring re-runs, etc.

i'm alot more comfortable with the idea of just restoring the filesystem
dump.

so, the question becomes, if i restore a filesystem snapshot backup, the
DB will be in a 'crashed' state.

what, if any, risks are there with regards to corruption, or the inability to
actually use the restored DB?

[Albe Laurenz]

Post by Jim Mercer
i'm running pgsql 9.1.3 (soon to be upgraded to 9.2.x) on FreeBSD 8-STABLE.
i've got a db of some 550GB that i want to backup on a daily basis, as we
have had some nasty hardware/power/other issues of late.
with dump on FreeBSD, you can specify -L, which does a filesystem level
snapshot before the dump starts, and then clears it afterwards.
i did a backup using dump on the filesystems, except for $PGDATA, followed
by a pg_dumpall.
this took 4 hours, and ate some 30GB (after compression)
i'm currently doing a dump of the filesystems, including $PGDATA.
this is looking like it will take 6-7 hours, and looks like it will take
some 40-50GB (after compression)
if i were to attempt a restore by replaying the 'pg_dumpall', i suspect that
i would be looking at many hours, and possibly hiccups requiring re-runs, etc.
i'm alot more comfortable with the idea of just restoring the filesystem
dump.
so, the question becomes, if i restore a filesystem snapshot backup, the
DB will be in a 'crashed' state.
what, if any, risks are there with regards to corruption, or the inability to
actually use the restored DB?

If the filesystem snapshot is truly atomic, that should work fine:
http://www.postgresql.org/docs/current/static/backup-file.html

If you do a pg_start_backup before and a pg_stop_backup after the
snapshot, you'd gain the additional advantage of point-in-time-recovery
(if you archive WAL files).

Yours,
Laurenz Albe