Re: port conflicts when running tests concurrently on windows.

Hi,

On 2021-12-10 10:22:13 +0100, Peter Eisentraut wrote:
> On 09.12.21 19:41, Andres Freund wrote:
> > Withhttps://github.com/anarazel/postgres/commit/046203741803da863f6129739fd215f8a32ec357
> > all tests pass. pg_regress requires PG_REGRESS_SOCK_DIR because it checks for
> > TMPDIR, but windows only has TMP and TEMP.
>
> This looks reasonable so far.

I pushed that part, since we clearly need something like them.

> The commit messages 8f3ec75de4060d86176ad4ac998eeb87a39748c2 and
> 1d53432ff940b789c2431ba476a2a6e2db3edf84 contain some notes about what I
> thought at the time didn't work yet.

> In particular, the pg_upgrade tests
> don't support the use of Unix sockets on Windows. (Those tests have been
> rewritten since, so I don't know what the status is.)

ISTM we still use two different implementations of the pg_upgrade tests :(. I
recall there being some recent-ish work on moving it to be a tap test, but
apparently not yet committed.

It doesn't look like the vcregress.pl implementation respects
PG_TEST_USE_UNIX_SOCKETS right now.

> pg_regress.c at remove_temp() is still there. These things should probably
> be addressed before we can consider making this the default.

Hm, not immediately obvious what to do about this. Do you know if windows has
restrictions around the length of unix domain sockets? If not, I wonder if it
could be worth using the data directory as the socket path on windows instead
of the separate temp directory?

Greetings,

Andres Freund