Re: BDR - Failure of Primary Server - How to recover?

On 5 June 2015 at 23:32, cchee-ob <carter(dot)chee(at)objectbrains(dot)com> wrote:

> If my Primary Server in a BDR environment fails what is my recourse for
> recovery? My servers are in the cloud so I don't have control over IP
> address assignment either. This hasn't happen but I need to present a plan
> if our Production system has this occur.
>

You switch to writing to the secondary, presuming that you're not already
doing so (it's multi-master), stop writing to the old primary, and remove
the old primary using bdr.bdr_part_by_node_names(...).

Data that was committed to the old primary but not yet replicated is lost;
if you can't deal with that you'll need synchronous replication.

I would be more inclined to use normal streaming replication and hot
standby for a simple master/replica failover scenario, though UDR (or BDR
in subscribe mode) is becoming quite useful for this use case too. If you
don't need multi-master, don't use multi-master.

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