Re: best config - Mailing list pgsql-general

On Feb 6, 2013, at 9:55 AM, Roberto Scattini =
<roberto.scattini@gmail.com> wrote:

>=20
> hi steven,
>=20
> > we have two new dell poweredge r720. based on recommendations from =
this list we have configued the five disks in raid10 + 1 hot spare.
>=20
> You might mention a bit more about how your drives are configured.  5 =
drives in a RAID1+0 sounds odd to me.
>=20
>=20
> i mean, 4 disks in raid10, plus one disk as hot spare.
>=20
> also, wasn't this list where recommended this setup, was in =
debian-user.
> =20

That makes a lot more sense.  Nothing wrong with that setup :-)

> >
> > now we are looking for advice in the postgres installation for our =
setup.
> >
> > we have two databases. one for a lot of small apps and one for one =
big app with a lot of data and a lot of usage.
> > we want to use streaming replication to have a functional copy of =
databases in a failure.
> >
> > one of the ideas is to have one database running on each server, and =
then have another instance of the other database running in streaming =
replication (i mean, crossed replications).
> >
> > the other idea is to have both databases running in one server and =
backup everything in the other with streaming replication.
> >
> > which alternative would you use?
>=20
> I would not introduce the complexity of having each server be master =
for half of the data unless you can show that this improves some metric =
you care a lot about.  Any failure or maintenance event will revert you =
back to the common configuration -- back to having both masters on one =
system -- until you do another promotion back to the "cross wired" =
setup.  Extra work without a proposed gain.
>=20
> Plus then you can get away with half as many Postgres installs to =
maintain.
>=20
> ok. we thought in this crossed-replication config because one heavy =
query in one of the databases wouldnt affect the performance of the =
other.=20

Both of your servers need to be powerful enough to handle the whole =
load, otherwise your replication setup will not continue to function =
acceptably when one of the servers is offline due to a crash or =
maintenance.

I don't think there is anything necessarily wrong with your proposal, I =
am just pointing out that simplicity is better than complexity unless =
you can prove (say, in a test environment) that your application =
actually performs "better enough to justify the administrative cost" =
with this cross-wired setup.