Supervisor fails to restart half of the time

This is not necessarily an error from supervisor. I see from your systemctl status output that supervisor is started through the sysv-init compatibility layer, so the failure could be in the /etc/init.d/supervisor script. It would explain the absence of errors in the supervisord logs.

To debug the init script, the easiest way is to add a set -x as first non-comment instruction in that file, and look in the journalctl output the trace of the script execution.

EDIT:

I've reproduced and debugged it on a test system with Debian Sid.

The issue is that the stop target of the supervisor init-script does not check if the daemon has been really terminated but only send a signal if the process exists. If the daemon process takes a while to shutdown, the subsequent start action will fail due to the dying daemon process, which is counted as already running.

I've opened a bug on Debian Bug Tracker: http://bugs.debian.org/805920

WORKAROUND:

You can workaround the issue with:

/etc/init.d/supervisor force-stop && \
/etc/init.d/supervisor stop && \
/etc/init.d/supervisor start

• force-stop will ensure the supervisord has been terminated (outside systemd).
• stop make sure systemd know it's terminated
• start starts it again

The stop after the force-stop is required otherwise systemd will ignore any subsequent start request. stop and start can be combined using restart, but here I've put both of them to show how it works.