Обсуждение: Multiple clusters on one box for PITR

Hi all,

I run a PostgreSQL 8.1 server that hosts two main applications -- both
web-based OLTP-type work -- with a few databases for each. I'm currently
using pg_dump for backups, but the clients for one of the applications
we're hosting as asked to be moved to PITR because they like the idea of
being able to restore their databases to any point in time. The problem
is that while that client is only using about 1GB on the box, the other
is using about 25GB, and since PITR is cluster-wide, I'd be stuck
restoring 26GB worth of databases if I ever needed to go to the backups.

However, I'm running on the Debian Etch packages, which seem to make it
quite easy to run multiple clusters, enabling me to run PITR just on the
application that requires it. Would running separate clusters on the one
box for these two applications be a sensible thing to do? Are there
resource limits that I should be aware of? Any other issues?

Thanks
Leigh

Leigh Dyer wrote:
>
> However, I'm running on the Debian Etch packages, which seem to make it
> quite easy to run multiple clusters, enabling me to run PITR just on the
> application that requires it. Would running separate clusters on the one
> box for these two applications be a sensible thing to do? Are there
> resource limits that I should be aware of? Any other issues?

There's no problem running multiple copies of PG on the same box - lots
of the developers do it, and I do on my laptop here. Different data-dir
and port and you're away.

Efficiency can be an issue though - you'll have two sets of WAL files
being written to, and two separate sections of shared memory to balance.
If you're not pushing the hardware though, it should be straightforward
enough.

--
   Richard Huxton
   Archonet Ltd