Do not run gilknocker in testsuite

[fjetter]

I noticed that stopping the gilknocker requires us to wait for a couple hundred ms here

distributed/distributed/system_monitor.py

Line 240 in 0ce1ee5

self._gilknocker.stop()

cc @milesgranger

This isn't great for gen_cluster since the test logic within a unit test is typically much faster such that this waiting on the gilknocker is the most expensive part of a test startup, execution and teardown.

At least locally, this makes a big difference for me. When writing dask/dask#10722 this made a difference of 0.4s vs 0.1s per test iteration, i.e. the tests are four times faster without gilknocker.

[milesgranger]

Good for me, assuming tests are happy.

Makes sense, could have done a .stop(fast=True) or something in gilknocker that would flag it to not wait to join the underlying thread. But I think I like this better. 👍

[milesgranger]

Oop, will have to enable it for the test_prometheus tests.
https://github.com/dask/distributed/actions/runs/7260897311/job/19781001789?pr=8423#step:19:2921

[github-actions]

Unit Test Results

See test report for an extended history of previous test failures. This is useful for diagnosing flaky tests.

       27 files  ±0         27 suites  ±0   9h 39m 24s ⏱️ - 2h 14m 37s
  3 950 tests ±0    3 839 ✔️ +1     109 💤 ±0  2 ❌ ±0 
49 683 runs   - 1  47 392 ✔️ ±0  2 289 💤 ±0  2 ❌ ±0 

For more details on these failures, see this check.

Results for commit 2f5bf41. ± Comparison against base commit 5af14d6.

♻️ This comment has been updated with latest results.

[fjetter]

Test runtimes of CI went down significantly. Just cherry picking a couple of runtimes, it looks like we could reduce the duration of individual jobs easily by 10-20%