> On 26 Jan 2017, at 10:34, Michael Paquier <michael.paquier@gmail.com> wrote:
>
> On Thu, Jan 26, 2017 at 4:09 PM, Nikhil Sontakke
> <nikhils@2ndquadrant.com> wrote:
>>> I look at this patch from you and that's present for me:
>>> https://www.postgresql.org/message-id/CAMGcDxf8Bn9ZPBBJZba9wiyQq->Qk5uqq=VjoMnRnW5s+fKST3w@mail.gmail.com
>>
>>> --- a/src/backend/access/transam/xlog.c
>>> +++ b/src/backend/access/transam/xlog.c
>>> @@ -9573,6 +9573,7 @@ xlog_redo(XLogReaderState *record)
>>> (errmsg("unexpected timeline ID %u (should be %u)
>>> in checkpoint record",
>>> checkPoint.ThisTimeLineID, ThisTimeLineID)));
>>>
>>> + KnownPreparedRecreateFiles(checkPoint.redo);
>>> RecoveryRestartPoint(&checkPoint);
>>> }
>>
>> Oh, sorry. I was asking about CheckpointTwoPhase(). I don't see a
>> function by this name. And now I see, the name is CheckPointTwoPhase()
>> :-)
>
> My mistake then :D
>
>>> And actually, when a XLOG_CHECKPOINT_SHUTDOWN record is taken, 2PC
>>> files are not flushed to disk with this patch. This is a problem as a
>>> new restart point is created... Having the flush in CheckpointTwoPhase
>>> really makes the most sense.
>>
>> Umm, AFAICS, CheckPointTwoPhase() does not get called in the "standby
>> promote" code path.
>
> CreateRestartPoint() calls it via CheckPointGuts() while in recovery.
>
Huh, glad that this tread received a lot of attention.
> On 24 Jan 2017, at 17:26, Nikhil Sontakke <nikhils@2ndquadrant.com> wrote:
>
> We are talking about the recovery/promote code path. Specifically this
> call to KnownPreparedRecreateFiles() in PrescanPreparedTransactions().
>
> We write the files to disk and they get immediately read up in the
> following code. We could not write the files to disk and read
> KnownPreparedList in the code path that follows as well as elsewhere.
Thanks Nikhil, now I got that. Since we are talking about promotion we are on different timescale and 1-10 second
lag matters a lot.
I think I have in my mind realistic scenario when proposed recovery code path will hit the worst case: Google cloud.
They have quite fast storage, but fsync time is really big and can go up to 10-100ms (i suppose it is
network-attacheble).
Having say 300 prepared tx, we can delay promotion up to half minute.
So i think it worth of examination.
--
Stas Kelvich
Postgres Professional: http://www.postgrespro.com
The Russian Postgres Company