Re: Separating bgwriter and checkpointer

On Tue, Oct 18, 2011 at 12:53 PM, Simon Riggs <simon@2ndquadrant.com> wrote:
> On Tue, Oct 18, 2011 at 5:39 PM, Robert Haas <robertmhaas@gmail.com> wrote:
>> On Tue, Oct 18, 2011 at 9:18 AM, Simon Riggs <simon@2ndquadrant.com> wrote:
>>> Any reason or objection to committing this patch?
>>
>> Not on my end, though I haven't reviewed it in detail.  One minor note
>> - I was mildly surprised to see that you moved this to the
>> checkpointer rather than leaving it in the bgwriter:
>>
>> +       /* Do this once before starting the loop, then just at SIGHUP time. */
>> +       SyncRepUpdateSyncStandbysDefined();
>>
>> My preference would probably have been to leave that in the background
>> writer, on the theory that the checkpointer's work is likely to be
>> more bursty and therefore it might be less responsive.
>
> That needs to be in the checkpointer because that is the process that
> shuts down last.
>
> The bgwriter is now more like the walwriter. It shuts down early in
> the shutdown process, while the checkpointer is last out.
>
> So it wasn't preference, it was a requirement of the new role definitions.

Oh, I see.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company