[Pgpool-general] Combining pgpool and pgbouncer

[Glyn Astill]

--- On Thu, 12/8/10, Josh Berkus <josh at agliodbs.com> wrote:

> Does anyone here combine pgpool and
> pgbouncer?
> 
> If so, what order do you chain them in ... pgpool, then
> multiple
> pgbouncer pools, or pgbouncer in front of pgpool?  And
> why?
> 
> I'm thinking that pgbouncer in front of pgpool makes sense
> just because
> it would reduce the number of connections which pgpool
> needs to deal with.
> 

Interesting, but how would it reduce the number of connections pgpool needs to deal with? Unless you can't get the pooling behaviour you want from pgpool? Is it not pooling the connections in the way you want?

In your previous message you stated you needed up to 600 concurrent connections, so if you also want up to 600 concurrent connections coming from pgbouncer you'd still need 600 pgpool backends no matter which way around you chain them wouldn't you?

Here we use them both separately, pgpool for our regular array of apps where we have lots of fairly persistent connections and login roles that change frequently.

Then we have pgbouncer for our web servers where there's only a few different login roles and whose connections are frequent and short lived. We have separate pgbouncer instances pointing to the replicas and the origin which the app chooses as appropriate.