I'm running a lot of UI tests in a pipeline via a GitLab Runner. The test are run on VMs in the GCP in docker containers.
Most of the time this works without problems, but sporadically a job running the ui tests will time out, because one test is stuck until the global job timeout is reached

Context:

• This happens to random tests, not always the same
• Test timeout is set via BrowserContext...setDefaultTimeout() but these tests will never timeout
  • However, timeout does work in general for test failures/flickers
  • setDefaultTimeout() will also set setDefaultNavigationTimeout
• VMs and docker container are reachable and there is seemingly enough RAM and CPU
• Thread dump just shows that the test is at PlaywrightAssertions.assertThat(locator).isVisible(); and waiting
• This also happens when given copious amount of RAM and CPU
• In the trace viewer it seems to be stuck on an "about:blank" page and the error "Timeout 10000ms exceeded." is shown
  • This always seem to happen very early in the test, basically after the first Page.navigate and then Frame.click
  • The browser (in the trace viewer) is always blank for every step
• Actively killing the running chrome instance fails the tests and then it succeeds in subsequent retries
• Only way to reproduce this behavior somewhat locally is to send a SIGSTOP signal to the chrome instance
  • Sending SIGCONT will just resume the test, it will not immediately time out
• Setting DEBUG=pw:api does unfortunately not output anything for the stuck test execution

Already done:

• More memory for the server
• Updated to newest Playwright version 1.49.0
• Updated to Chrome channel new BrowserType...setChannel("chrome")

Questions:

• Are there more possibilities to set timeouts besides setting a JUnit timeout for the UI tests?
• How could I further debug this?