With no events waiting, bin/process-repository-events.php exited after three empty batches a quarter of a second apart. Supervisor's program (deployment/repository-events-supervisor.conf, startsecs=2) counts a program that exits within two seconds as failing to start, so it stayed in STARTING and BACKOFF and would give up as FATAL.
The worker now looks for events every two seconds while there are none and exits after five minutes, so Supervisor starts it again on the current release. A stop signal ends a wait at once. A second copy that finds the lock held waits three seconds before exiting.