Thread: Active/Active clustering in postgres

I am looking for PostgreSQL active/active clustering and  whether PostgreSQL support any form of shared-storage clustering . Is there any methods or tools for implementing active/active clustering on Postgres supported by community or any third party tools.

-- 
john r pierce                                      37N 122W
somewhere on the middle of the left coast

(2014/11/27 2:20), John R Pierce wrote:
> On 11/26/2014 2:36 AM, Postgres India wrote:
>> I am looking for PostgreSQL active/active clustering and  whether
>> PostgreSQL support any form of shared-storage clustering . Is there
>> any methods or tools for implementing active/active clustering on
>> Postgres supported by community or any third party tools.
>
> you might look into PostgreSQL-XC and -XL (-XL is a fork of -XC). both
> of these are multi-master clusters based on postgres.


In addition, XC and XL are shared-nothing DB clusters.

I think that there is no multi-master shared-storage DB cluster based on
PostgreSQL.

* Suzuki Hironobu (hironobu@interdb.jp) wrote:
> (2014/11/27 2:20), John R Pierce wrote:
> >On 11/26/2014 2:36 AM, Postgres India wrote:
> >>I am looking for PostgreSQL active/active clustering and  whether
> >>PostgreSQL support any form of shared-storage clustering . Is there
> >>any methods or tools for implementing active/active clustering on
> >>Postgres supported by community or any third party tools.
> >
> >you might look into PostgreSQL-XC and -XL (-XL is a fork of -XC). both
> >of these are multi-master clusters based on postgres.
>
> In addition, XC and XL are shared-nothing DB clusters.
>
> I think that there is no multi-master shared-storage DB cluster
> based on PostgreSQL.

I'm not aware of any.  It would require the equivilant of the
distributed lock management which exists in RAC (and is also what makes
RAC limited in its ability to scale..).  Note that applications which
are written to use RAC are likely easier to migrate to shared-nothing
sharded database systems since they would have already had to deal with
the complications associated with determining which node to send writes
to.

If your application isn't doing that then you're very unlikely to be
using RAC in a performant manner (as it relates to write-load
distribution, specifically) and would operate just fine with a single
master and slave replicas, with appropriate systems in place to handle
failover.

    Thanks,

        Stephen