Pipeline timeouts caused by high concurrency under low CPU core count

[Timovzl]

Greetings,

Has anyone else run into this: Pipeline agents generally have few CPU cores, and combined with high concurrency this sometimes causes tests to run for more than 30 seconds and thus time out.

For me, this reared its head as soon as I started passing cancellation tokens. It seems to occur particularly with async integration tests. That makes sense - these actually benefit from concurrency rather than just from parallelism. If the cores are too busy handling various pending requests, some may run for longer.

I've worked around the issue with a 120s timeout, but the 30s default seems perfectly reasonable to me. Any ideas for a more robust solution?

[thomhurst]

You can use the CLI flag --maximum-parallel-tests # or you can use a [ParallelLimiter<T>] attribute in the code and apply it to your assembly.

[thomhurst]

More info on the latter here: https://tunit.dev/docs/execution/parallelism/#parallellimitert--limiting-concurrent-test-count

[Timovzl]

Especially the command line argument and environment variable seem interesting for pipelines specifically.

[Timovzl]

@thomhurst Does placing the [ParallelLimiter] attribute on a base class affect all of its subclasses?

[thomhurst]

@Timovzl it should do, yep!