On Wed, 2008-12-17 at 17:54 +0100, Jonatan Evald Buus wrote:
> Greetings,
> We're currently developing an alternative to pg_standby, internally
> named pg_ha.
> pg_ha uses Record Based Log Shipping fetched by the warm standby
> server at millisecond intervals and will automatically restore
> completed WAL files on the standby server to minimize the time
> required when restoring it to production status after a failure on the
> primary server.
> Once completed we'd be happy to donate the code to the PostGreSQL
> community to assist in promoting the database server as a High
> Availability option.
We're reworking replication now for 8.4 and it sounds like we're already
some way ahead of you on that development.
Have a look at Skytools' walmgr if you want a stream using 8.3. It uses
production APIs built into the server in 8.2.
> During each restore cycle however we notice that PostGreSQL spawns a
> new timeline for the recovered archive as described in section 24.3.4
> at
> http://www.postgresql.org/docs/8.3/static/continuous-archiving.html#BACKUP-BASE-BACKUP.
> Is it possible to specify that no new timeline should be spawned
> during the archiver recovery process?
>
> The only workaround otherwise appears to be to detect the current
> timeline number using pg_controldata and rename copied WAL files
> accordingly prior to initiating the archive recovery process.
That sounds like a bad idea, for minor but annoying technical reasons.
Please check the archives if you're interested further.
--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support