Tom Lane wrote:
>Dmitry Tkach <dmitry@openratings.com> writes:
>
>
>>- The database files sit on a NetApp box (this is a very reliable disk array with parity control etc...
>>plus, it has a 'snapshot' capability, that allows you to take a live snapshot of any directory in real time - that
creates
>>a read-only copy of the content of that directory within a second or so).
>>- The snapshots are taken hourly by a cron job on the whole database (older snapshots get removed after a while), and
themost recent snapshot
>>gets backed up to the tape every night.
>>
>>
>
>Have you ever actually restored one of those snapshots? I don't believe
>this will work, unless you shut down the database while taking the
>snapshot. Or unless the snapshot is instantaneous across all files in
>the $PGDATA directory tree --- but your "within a second or so" doesn't
>sound real promising. If there's any time skew between copying
>different files, you've got problems.
>
>
>
Yes, it *is* instanteneous, that's the point!
The idea is, that it does not really copy anything at all - just freezes
all the stuff under that directory exactly the way it is, and then, if
somebody (like postmaster) is writing to that directory, it replaces the
particular block that's is neing modified, so that the snapshot contiues
to look at the old block, and the 'real thing' referes to the new one...
That 'under a second' time is what it needs to setup its bookeeping
recors, and it freezes all writes to the directory during that time...
Dima
