Re: Replication - Mailing list pgsql-general

From Craig Ringer
Subject Re: Replication
Date
Msg-id 4CD09BC5.6010809@postnewspapers.com.au
Whole thread Raw
In response to Re: Replication  (Vick Khera <vivek@khera.org>)
List pgsql-general
On 02/11/10 21:21, Vick Khera wrote:
> On Mon, Nov 1, 2010 at 6:39 PM, Jonathan Tripathy <jonnyt@abpni.co.uk> wrote:
>> So in the "Hot Standby" setup as described in
>> http://www.postgresql.org/docs/current/static/hot-standby.html , how would I
>> automatically make the slave a master?
>
> If you plan to make it automatic, be absolutely 1000000% sure that
> your automated criteria for doing the switchover is really 1000000%
> accurate, else you may end up switching when you didn't really want to
> for some temporary failure condition.  I've never been able to define
> something that perfect so we still only ever do manual switchovers.

From what I've seen, the only way automatic switchovers ever work sanely
is when the node that's promoting its self has a way to pull the plug on
the master it's taking over from. A USB-controlled power board seems to
be a popular cheap option, and isolation on a fibre-channel switch a
more expensive option.

Of course, even then you have to be sure your method for killing the old
master will always work when the slave promotes its self to master, and
will never trigger under any other circumstances. Good luck with that.

(Reading the above par, does anyone else find some IT terminology, when
read out of context, kind of creepy? Unix's killing of children in
particular.)

>
> Based on your questions, you perhaps should be seeking the advice of a
> paid consultant expert in such matters if you really value your data.

+1

There are many people on this list who do paid work. See the PostgreSQL
website for a list of companies that work with PostgreSQL.

--
Craig Ringer

Tech-related writing: http://soapyfrogs.blogspot.com/

pgsql-general by date:

Previous
From: Craig Ringer
Date:
Subject: Re: JDBC Transactions
Next
From: Tim Uckun
Date:
Subject: Re: Postgres connection errors