Re: Geographic High-Availability/Replication

Hi,

Matthew wrote:
> Hey all, new postgres user here. We are trying to setup/research an
> HA/Replicated solution with Postrgresql between a datacenter in LA and a
> d.c. in NY.
>
> We have a private LAN link between the two D.C.'s with a max round-trip
> of 150ms.
>
> We will have a web server at each d.c. (among other servers) that will
> write/read to/from the local LAN DB. On writes, that data should be
> xmited to the other data center so that if, for whatever reason, my
> website request was sent to LA instead of NY, all my session information
> etc will still exist.

This is commonly known as synchronous replication. As that involves
*at-least* one round-trip *before* committing, it's quite expensive. Can
you live with a delay of ~150ms before COMMIT confirmation?

Another issue is the reliability of your failure detectors. How does
server B know that server A is really down (and not only the link?).
Normally, that's solved with a quorum device. So that you have to have
at least three servers - preferably in different locations.

Regards

Markus

Disclaimer: I'm the developer behind Postgres-R (www.postgres-r.org),
have a look at it, it's designed to do what you are looking for.