On 12/11/2014 04:21 PM, Marco Nenciarini wrote:
> Il 11/12/14 12:38, Andres Freund ha scritto:
>> On December 11, 2014 9:56:09 AM CET, Heikki Linnakangas <hlinnakangas@vmware.com> wrote:
>>> On 12/11/2014 05:45 AM, Andres Freund wrote:
>>>
>>> Yeah. I was not able to reproduce this, but I'm clearly missing
>>> something, since both you and Sergey have seen this happening. Can you
>>> write a script to reproduce?
>>
>> Not right now, I only have my mobile... Its quite easy though. Create a pg-basebackup from a standby. Create a
recovery.confwith a broken primary conninfo. Start. Shutdown. Fix conninfo. Start.
>>
>
> Just tested it. There steps are not sufficient to reproduce the issue on
> a test installation. I suppose because, on small test datadir, the
> checkpoint location and the redo location on the pg_control are the same
> present in the backup_label.
>
> To trigger this bug you need to have at least a restartpoint happened on
> standby between the start and the end of the backup.
>
> you could simulate it issuing a checkpoint on master, a checkpoint on
> standby (to force a restartpoint), then copying the pg_control from the
> standby.
>
> This way I've been able to reproduce it.
Ok, got it. I was able to reproduce this by using pg_basebackup
--max-rate=1024, and issuing "CHECKPOINT" in the standby while the
backup was running.
Yeah, I agree with Andres' plan to relax the check, to also allow
DB_SHUTDOWNED_IN_RECOVERY state.
- Heikki