On CI I see a lot of flakes on Windows:
The test is not written terribly well, but basically this means that the batch script running ping every second in a loop was sent a termination signal yet continued running for at least five seconds after the interrupt. Did the signal get lost? Sent to the wrong subprocess without breaking the loop? Was it going to get handled but the system was just too heavily loaded? Could probably improve test to:
- Use a single process for the batch script, like ping -n 99999 127.0.0.1 >tmp.
- Wait indefinitely (up to global test timeout) for the file to not have been touched in the last few seconds.