We presently use a backup strategy based on doing an rsync from an LVM
snapshot of the $PGDATA filesystem that has first been frozen and flushed to
disk (the discussion at
http://marc.theaimsgroup.com/?l=postgresql-admin&w=2&r=1&s=LVM+snapshots&q=b
gives all the gory details). I'm contemplating putting the pg_xlog/ folder
on separate disks, to improve performance, and am wondering if the following
will give me "consistent" copies of the $PGDATA cluster:
1.) Freeze the filesystem on which $PGDATA/pg_xlog/ (the postgresql log
files) resides
2.) Freeze the filesystem on which the rest of $PGDATA resides
3.) Create the snapshot
4.) Unfreeze the rest of $PGDATA
5.) Unfreeze $PGDATA/pg_xlog/
6.) Mount and copy the snapshot
[7.) Unmount and remove the snaphsot]
In other words, if writes to pg_xlog/ drag on, will the postmaster just idly
wait, or will it decide to flush logs (checkpoint) while waiting for writes
to pg_xlog/ to complete.
Thanks,
Murthy
