Re: pg_upgrade should truncate/remove its logs before running

On Wed, Dec 22, 2021 at 09:52:26AM -0500, Tom Lane wrote:
> Michael Paquier <michael(at)paquier(dot)xyz> writes:
> > On Mon, Dec 20, 2021 at 09:39:26PM -0600, Justin Pryzby wrote:
> >> Are you suggesting to remove these ?
> >> -/pg_upgrade_internal.log
> >> -/loadable_libraries.txt
>
> > Yep, it looks so as these are part of the logs, the second one being a
> > failure state.
>
> >> -/reindex_hash.sql
>
> > But this one is not, no?
>
> I'd like to get to a state where there's just one thing to "rm -rf"
> to clean up after any pg_upgrade run. If we continue to leave the
> we-suggest-you-run-these scripts loose in $CWD then we've not really
> improved things much.

My patch moves reindex_hash.sql, and I'm having trouble seeing why it shouldn't
be handled in .gitignore the same way as other stuff that's moved.

But delete-old-cluster.sh is not moved, and I'm not sure how to improve on
that.

> Perhaps there'd be merit in putting log files into an additional
> subdirectory of that output directory, like
> pg_upgrade_output.d/logs/foo.log, so that the more-ignorable
> output files would be separated from the less-ignorable ones.
> Or perhaps that's just gilding the lily.

In the case it's successful, everything is removed - except for the delete
script. I can see the case for separating the dumps (which are essentially
internal and of which there may be many) and the logs (same), from
the .txt error files like loadable_libraries.txt (which are user-facing).

It could also be divided with each DB having its own subdir, with a dumpfile
and a logfile.

Should the unix socket be created underneath the "output dir" ?

Should it be possible to set the output dir to "." ? That would give the
pre-existing behavior, but only if we don't use subdirs for log/ and dump/.