Chrome canary and stable runs are intermittently hanging

[DanielRyanSmith]

Initially identified by @jgraham:

https://wpt.fyi/runs?label=master&label=stable&from=2025-12-08T00%3A00 looks like Chrome stable runs are hanging: https://github.com/web-platform-tests/wpt/runs/58035842405 Maybe also in nightly(!) https://github.com/web-platform-tests/wpt/runs/58035840099

Based on my initial investigation, it looks like the job shard named wpt-chrome-{canary|stable}-testharness-7 takes a significantly longer time to complete than the other shards. On a sucessful run, it tends to finish in around 170 minutes, which is 10 minutes away from the 180 minute timeout. At first glance, I am not seeing any specific tests in the logs that are taking substantially longer to complete. It could be that these failures are just a coincidental side effect of the run time getting too close to the timeout.

Firefox runs of this shard take a significantly shorter time to complete; around 60 to 90 minutes (example), so I am assuming there is something Chrome-specific about this increased run time.

[DanielRyanSmith]

After digging a little deeper, it seems like the problem is simply that there are too many tests timing out with a 1-minute timeout window. Specifically, speculation-rules/prerender tests (see logs). These tests are taking well over an hour to all timeout. These tests do not affect Firefox runs because they error out rather than timing out (see logs).

The quickest fix here is to just increase the timeout, or increase the number of chunks for the test harness tests (both here), but this does not address the issue of long runs due to consecutive test timeouts.

[jgraham]

This is a problem that other browser have often faced when some feature that's not widely implemented has tests that timeout. So far we don't have a better general purpose solution[*] than just fixing the tests to fail more gracefully (although in this case the tests timing out in Chrome suggests some other issue, since presumably the feature works there)

[*] Although if you look in wptrunner you notice there's some hacky code in the testloader added by the people running Firefox CI so that we don't run tests that are known to timeout on every run.