Re: Allow interrupts on waiting standby

From: Michael Paquier <michael(dot)paquier(at)gmail(dot)com>
To: Simon Riggs <simon(at)2ndquadrant(dot)com>
Cc: Andres Freund <andres(at)anarazel(dot)de>, Robert Haas <robertmhaas(at)gmail(dot)com>, PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: Allow interrupts on waiting standby
Date: 2017-01-27 01:35:17
Message-ID: CAB7nPqRNcd+FR=8_Lyb65jBcRPX+59hXobi5G-wf4fxRak_6gw@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-hackers

On Fri, Jan 27, 2017 at 4:36 AM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:
> On 26 January 2017 at 19:20, Andres Freund <andres(at)anarazel(dot)de> wrote:
>> On 2017-01-26 12:24:44 -0500, Robert Haas wrote:
>>> On Thu, Jan 26, 2017 at 7:18 AM, Simon Riggs <simon(at)2ndquadrant(dot)com> wrote:
>>> > Currently a waiting standby doesn't allow interrupts.
>>> >
>>> > Patch implements that.
>>> >
>>> > Barring objection, patching today with backpatches.
>>>
>>> "today" is a little quick, but the patch looks fine. I doubt anyone's
>>> going to screech too loud about adding a CHECK_FOR_INTERRUPTS() call.
>>
>> I don't quite get asking for agreement, and then not waiting as
>> suggested. I'm personally fine with going with a CHECK_FOR_INTERRUPTS
>> for now, but I think it'd better to replace it with a latch.
>
> I have waited, so not sure what you mean. Tomorrow is too late.

This gives really little time for any feedback :(

> Replacing with a latch wouldn't be backpatchable, IMHO.
> I've no problem if you want to work on a deeper fix for future versions.

A deeper fix for HEAD proves to not be that complicated. Please see
the attached. The other two calls of pg_usleep() in standby.c are
waiting with 5ms and 10ms, it is not worth switching them to a latch.
--
Michael

Attachment Content-Type Size
standby-delay-latch.patch application/octet-stream 2.1 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Michael Paquier 2017-01-27 01:53:44 Re: increasing the default WAL segment size
Previous Message Andres Freund 2017-01-27 01:34:26 Re: safer node casting