Bruce Momjian <bruce@momjian.us> writes:
> On Fri, Aug 3, 2018 at 04:56:32PM -0400, Bruce Momjian wrote:
>> On Fri, Aug 3, 2018 at 01:55:04PM -0400, Tom Lane wrote:
>>> Right now is probably not a good time to fix this, but it seems like
>>> something that could be improved. I'd be kind of inclined to remove
>>> the pidfile checking business altogether in favor of inspecting the
>>> state in pg_control; or at least do them both in the same place with
>>> the same recovery attempt if we don't like what we see.
>> Yes, I realize this was inconsistent. It was done this way purely based
>> on how easy it would be to check each item in each place. I am fine
>> with removing the #3 cleanup so we are consistent. We can also add docs
>> to say it should be a "clean" shutdown.
> How do you want me to handle this, considering it has to be backpatched?
Well, removing the check entirely is certainly not a good idea.
I looked at the code briefly and concur that making it work "nicely" isn't
as simple as one could wish; the code you added has some dependencies on
the context it's in, so that moving it elsewhere would require code
duplication or refactoring. Maybe it's not something to try to shoehorn
into v11. We can always improve it later.
(I'd also say that the weekend before a release wrap is no time
to be committing anything noncritical, so even if you're feeling
motivated to fix it in v11, best to hold off till after the wrap.)
regards, tom lane