See T13320. Here's roughly what I observed:
- Process 12345 was a daemon process.
- The host underwent a ninja reboot.
- After the host came back up, process 12345 was a system process.
- bin/phd start still attempted to kill it, based on the out-of-date PID log on disk.
The easiest fix is probably to make the "stop existing processes" check the name of the processes it is stopping, although I suspect this whole flow could use updates in general.