Re: postgresql scalability, active-active cluster

On Mon, 2007-01-22 at 17:37 -0300, Alvaro Herrera wrote:
> brian stone wrote:
> > I never considered MySQL because I really DO need transactions. MySQL
> > also lacks many enterprise features we need; well they say they have
> > them but from my testing they are a bit under-cooked.
> >
> > I need atomic actions across an N number of application servers. The
> > goal here is scalability, which is why I brought up clustering.
> > Unfortunately, we really do need those transactions :(
>
> Maybe two-phase commit can help you here? (see the COMMIT PREPARED
> docs)
>

How much of a concern are deadlocks?

Is the correct way to do this to serially execute all the statements on
Node1, then Node2, ... NodeN (in order) to avoid deadlocks?

If you used async execution on all nodes at once, and then re-issued any
deadlocked transactions, how rare would deadlocks need to be for that to
be faster?

Regards,
Jeff Davis