Re: testing hot standby

Jaime Casanova wrote:
> On Mon, Apr 12, 2010 at 8:32 PM, Fujii Masao <masao(dot)fujii(at)gmail(dot)com> wrote:
>> But when I did one more restart of the primary and standby, I was
>> able to observe the problem. If this is the same as you encountered,
>> it would be the "can't start hot standby from a shutdown checkpoint"
>> issue that Heikki pointed out. So it's very helpful to check whether
>> the posted patch fixes your problem or not.
>> http://archives.postgresql.org/pgsql-hackers/2010-04/msg00407.php
>
> this patch seems to fix the problem for me...

Committed, thanks for the testing.

> i will read it on the morning and the thread where it is, something
> that seems strange to me is that the patch touch twophase.c and
> twophase.h, why?

When you start hot standby from an online checkpoint, the XIDs of any
two-phase transactions are included in the running-xacts record, just
like any other in-progress transactions. At a shutdown checkpoint, we
know that no regular transactions are in-progress, but there can be
transactions in prepared state, which need to be considered as
in-progress in the standby, but there's no information about them in the
shutdown record. So we scan pg_twophase to discover them.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com