Re: Operation log for major operations

Justin Pryzby <pryzby(at)telsasoft(dot)com> writes:
> On Thu, Mar 02, 2023 at 08:57:43PM +0300, Dmitry Koval wrote:
>> These changes did not interest the community. It was expected (topic is very
>> specifiс: vendor's technical support). So no sense to distract developers

> Actually, I think there is interest, but it has to be phrased in a
> limited sense to go into the control file.

> In November, I referenced 2 threads, but I think you misunderstood one
> of them. If you skim the first couple mails, you'll find a discussion
> about recording crash information in the control file.

> https://www.postgresql.org/message-id/666c2599a07addea00ae2d0af96192def8441974.camel%40j-davis.com

> It's come up several times now, and there seems to be ample support for
> adding some limited information.

> But a "log" which might exceed a few dozen bytes (now or later), that's
> inconsistent with the pre-existing purpose served by pg_control.

I'm pretty dubious about recording *anything* in the control file.
Every time we write to that, we risk the entire database on completing
the write successfully. I don't want to do that more often than once
per checkpoint. If you want to put crash info in some less-critical
place, maybe we could talk.

regards, tom lane