Re: recovery modules - Mailing list pgsql-hackers

From Nathan Bossart
Subject Re: recovery modules
Date
Msg-id 20230214005558.GA1145703@nathanxps13
Whole thread Raw
In response to Re: recovery modules  (Andres Freund <andres@anarazel.de>)
Responses Re: recovery modules
List pgsql-hackers
On Mon, Feb 13, 2023 at 03:37:33PM -0800, Andres Freund wrote:
> On 2023-02-13 14:56:47 -0800, Nathan Bossart wrote:
>> On Mon, Feb 13, 2023 at 04:37:10PM +0900, Michael Paquier wrote:
>> > +   basic_archive_context = data->context;
>> > +   if (CurrentMemoryContext == basic_archive_context)
>> > +       MemoryContextSwitchTo(TopMemoryContext);
>> > +
>> > +   if (MemoryContextIsValid(basic_archive_context))
>> > +       MemoryContextDelete(basic_archive_context);
>> > 
>> > This is a bit confusing, because it means that we enter in the
>> > shutdown callback with one context, but exit it under
>> > TopMemoryContext.  Are you sure that this will be OK when there could
>> > be multiple callbacks piled up with before_shmem_exit()?  shmem_exit()
>> > has nothing specific to memory contexts.
>> 
>> Well, we can't free the memory context while we are in it, so we have to
>> switch to another one.  I agree that this is a bit confusing, though.
> 
> Why would we be in that memory context? I'd just add an assert documenting
> we're not.
> 
> 
>> On second thought, I'm not sure it's important to make sure the state is
>> freed in the shutdown callback.  It's only called just before the archiver
>> process exits, so we're not really at risk of leaking anything.  I suppose
>> we might not always restart the archiver in this case, but I also don't
>> anticipate that behavior changing in the near future.  I think this
>> callback is more useful for things like shutting down background workers.
> 
> I think it's crucial. Otherwise we're just ossifying the design that there's
> just one archive module active at a time.
> 
> 
>> I went ahead and removed the shutdown callback from basic_archive and the
>> note about leaking from the documentation.
> 
> -1

Okay.  I've added it back in v12 with the suggested adjustment for the
memory context stuff.

-- 
Nathan Bossart
Amazon Web Services: https://aws.amazon.com

Attachment

pgsql-hackers by date:

Previous
From: Andres Freund
Date:
Subject: Re: pg_walinspect memory leaks
Next
From: Nathan Bossart
Date:
Subject: Re: recovery modules