Wouldn't it be better to make the openssl and mysqli test parallel, rather than faffing with tests that might need to be slow by necessity? (it might even be possible to just remove the conflict file for ext/openssl now that ephemeral ports are used)

The second one is actually slow because the tested behavior is to break the program when memory isn't enough. We could just lower the memory amount to break it way earlier though. The first one is similar (it's testing max execution time. We don't need to wait for the default max execution time)

However, I do agree we can just remove the conflict file for ext/openssl now. Let's do it.

(33.675 s) mysqli_fetch_array() - large packages (to test compression) [D:\a\php-src\php-src\ext\mysqli\tests\mysqli_fetch_array_large.phpt]
(30.397 s) Bug #81145 (copy() and stream_copy_to_stream() fail for +4GB files) [D:\a\php-src\php-src\ext\standard\tests\file\bug81145.phpt]

It is faster for a couple of seconds for every OpenSSL and mysql tests as observed.

Probably makes sense to split the openssl test changes into its own PR.

I split it. See #22499

See e1ec67a. This was a conscious decision, as these tests caused more failures with parallelization. We do have the retry mechanism now, I don't know if that's enough.

@iluuu1994 I was aware. My point is to change the conflict key from all to mysql. Since ext/mysqli/tests/CONFLICTS and ext/pdo_mysql/tests/CONFLICTS both use mysql, these tests remain serialized with the MySQL test group and should still avoid shared MySQL server contention, and also allow unrelated tests to run in parallel.

wdyt? This is also the same in #22499

See https://github.com/php/php-src/blob/master/ext/mysqli/tests/CONFLICTS. This has no effect, the conflict key is already there. The CONFLICTS all was added after, because running it in parallel with other tests still caused failures. I don't remember the details, but could have been network overload.

I see. I thought this is solved because we now use ephemeral ports.

Given the uncertainty of the result of this PR. I think we can just return to the original idea to lower the ini setting to avoid unnecessary waiting. I am closing #22499

Ephemeral ports solve a different issue (different servers being spun up in parallel). There were different efforts to parallelize tests (avoid conflicting table names), but it looks like the CI change was never merged (i.e. removing the CONCLIFCTS file). Maybe @Girgias remembers why that is.