The "up and running" file was written before the signal handler
was installed, so it was possible for the test to kill the
subprocess before, leading to a missed "exiting" line in the
output.

I hope this fixes the sporadic failures we're seeing (#8080).