Nice to have. Currently, this is non-trivial since the multiprocess worker pull the tests from a queue. We don't know what's pulled from the main thread. We'd need to either implement some feedback mechanism, or indeed handle SIGTERM better on the worker side and print what's currently running. Though, I tried the latter before, it somehow doesn't work for a particular class of hanging tests that can't be killed. In particular on Mac.

Michael, do you have a draft CL for your failed attempt?

I think I only tried locally. But on Mac, on failed death-tests we can see in some builds that the text we usually print on caught SIGTERM is not printed. So we couldn't print anything else either...

The it seems we need to go with the watchdog thread approach that will have a way to communicate with the worker thread to get the list of currently running tests. One disadvantage with this is that we'll like have to duplicate the information about the timeout in cr-buildbucket.cfg and in logic for this watchdog thread, unless we find a way to request the current timeout from parent swarming process on the bot somehow or by reading cr-buildbucket.cfg via Gitiles APIs.