Re: pgpool + BDR, is it possible?

On 17 March 2015 at 17:18, Ruth Melendo <rmelendo(at)teltronic(dot)es> wrote:

> Thanks for your help.
>
>
>
> My app is a GIS Server and we strongly need high availability. The reason
> to be master/master is to share load because we have a lot of users and as,
> this is an app for security sector, each node must be able to work alone to
> ensure availability all the time.
>

If that's a requirement, then you're going to need to write the application
to cope with the consequences.

What if, while there's replication lag due to network issues, someone
creates a row with id "42" in one node. Someone else creates a row with id
"42" in the same table in another node?

The app has to be able to deal with that, or use things (like BDR's global
sequences) to prevent it. And you can't always prevent it unless writeable
tables are basically append-only, since two UPDATEs that affect the same
row can also conflict.

So ... *if* you can review your application and change it where necessary
to cope with the anomalies that can arise in asynchronous multi-master
replication, BDR will be absolutely ideal for your needs. That's pretty
much what it's for. But you need to be able to do that review and
adjustment rather than just pointing your app at a couple of BDR nodes and
hoping for the best.

BDR has conflict logging and conflict statistics features that will help
you during testing, too; see the docs for details on them.

--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services