Issues with tests after upgrading to v1.19.0-rc.1

[amatalai]

Elixir and Erlang/OTP versions

Erlang/OTP 28 [erts-16.1] [source] [64-bit] [smp:12:12] [ds:12:12:10] [async-threads:1] [jit]

Elixir 1.19.0-rc.1 (f642b30) (compiled with Erlang/OTP 28)

Operating system

MacOS 15.6 and docker erlang:28.1

Current behavior

After upgrading, mix test for our umbrella app become very slow.

1.19.0-rc.0 (otp 28.1): 1m 23s — all tests passed
1.19.0-rc.1 (otp 28.1): 14m 58s — multiple tests failed with process-related errors

It looks like it's running tests for an app and then waiting for something to happen before running tests for the next one.

Expected behavior

Tests should pass in a similar time as in earlier versions of Elixir

[amatalai]

I checked cd apps/my_app_web && mix test. All tests passed there, but the overall run still appears to stall at certain points.

I also tried rm -rf deps _build but the issue persists.

[josevalim]

Can you post the failures for the process tests? Try writing the output to a file and paste the first failing tests.

Or any chance you can provide a minimum way to reproduce it?

[amatalai]

The first error appears from the logger

20:43:32.926 [error] GenServer MyApp.SomethingWorker terminating
** (DBConnection.OwnershipError) cannot find ownership process for #PID<0.6023.0>.

The next ones look like they are caused by the supervision tree being dead.

I also noticed that tests are slowed down only for those umbrella apps that use ecto. I will try to create a small project to reproduce this.

[josevalim]

it seems something is causing your application supervisor tree to shut down, it may not be related to tests necessarily.

You can try this elixir --logger-sasl-reports true -S mix test and it should log supervision tree reports.

[amatalai]

You're right. It was some of the dependencies in the supervision tree that started to run slowly, not the tests themselves.
Unfortunately, copied to a smaller project they do not allow us to reproduce the problem.

I tried #14813 and the problem disappeared. After this change, tests seem to be even much faster than before.

I'm closing this because the core issue seems to be the same as in #14810

Thank you for your help