From: | Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> |
---|---|
To: | Kyotaro HORIGUCHI <horiguchi(dot)kyotaro(at)lab(dot)ntt(dot)co(dot)jp> |
Cc: | Fujii Masao <masao(dot)fujii(at)gmail(dot)com>, PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: some review comments on logical rep code |
Date: | 2017-04-25 01:11:06 |
Message-ID: | CAD21AoBAXRMBBaH5-mYXxksPaTis0xt_-yvvb2Y+jG2m-EWAoQ@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On Mon, Apr 24, 2017 at 7:57 PM, Kyotaro HORIGUCHI
<horiguchi(dot)kyotaro(at)lab(dot)ntt(dot)co(dot)jp> wrote:
> Hello,
>
> At Mon, 24 Apr 2017 11:18:32 +0900, Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote in <CAD21AoBU8mZdGx-sTJ3u+qkaej5RPnOuotW4kfeXc4XdDNF8cw(at)mail(dot)gmail(dot)com>
>> >> BEGIN;
>> >> ALTER SUBSCRIPTION hoge_sub ENABLE;
>> >> PREPARE TRANSACTION 'g';
>> >> BEGIN;
>> >> SELECT 1;
>> >> COMMIT; -- wake up the launcher at this point.
>> >> COMMIT PREPARED 'g';
>> >>
>> >> Even if we wake up the launcher even when a ROLLBACK PREPARED, it's
>> >> not a big deal to the launcher process actually. Compared with the
>> >> complexly of changing the logic so that on_commit_launcher_wakeup
>> >> corresponds to prepared transaction, we might want to accept this
>> >> behavior.
>> >
>> > on_commit_launcher_wakeup needs to be recoreded in 2PC state
>> > file to handle this issue properly. But I agree with you, that would
>> > be overkill for small gain. So I'm thinking to add the following
>> > comment into your patch and push it. Thought?
>> >
>> > -------------------------
>> > Note that on_commit_launcher_wakeup flag doesn't work as expected in 2PC case.
>> > For example, COMMIT PREPARED on the transaction enabling the flag
>> > doesn't wake up
>> > the logical replication launcher if ROLLBACK on another transaction
>> > runs before it.
>> > To handle this case properly, the flag needs to be recorded in 2PC
>> > state file so that
>> > COMMIT PREPARED and ROLLBACK PREPARED can determine whether to wake up
>> > the launcher. However this is overkill for small gain and false wakeup
>> > of the launcher
>> > is not so harmful (probably we can live with that), so we do nothing
>> > here for this issue.
>> > -------------------------
>> >
>>
>> Agreed.
>>
>> I added this comment to the patch and attached it.
>
> The following "However" may need a follwoing comma.
>
>> However this is overkill for small gain and false wakeup of the
>> launcher is not so harmful (probably we can live with that), so
>> we do nothing here for this issue.
>
> I agree this as a whole. But I think that the main issue here is
> not false wakeups, but 'possible delay of launching new workers
> by 3 minutes at most'
In what case does launching of the apply worker delay 3 minutes at most?
Regards,
--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center
From | Date | Subject | |
---|---|---|---|
Next Message | Amit Langote | 2017-04-25 01:22:23 | Re: pgrowlocks relkind check |
Previous Message | Noah Misch | 2017-04-25 01:10:52 | Re: PG_GETARG_GISTENTRY? |