Supervisord slow reload/restart demo.

Due to supervisor signalling behavior separate programms get signalled one at a time. Groups on the other hand share signals between all workers simultaniuosly. Supervisord will wait for script to exit before starting another task, signalling to other programm or starting another one.

supervisorctls controller plugins do_signal separate process names and group names and signall them differenly.

RPC on the other side on receiving signal proceeds incoming signal to group processes via make_all_func SignalProcessGroup.

As of supervisor 4.2.4 supervisorctl reload does not accept arguments. So we cannot use all, like with supervisor restart.

But we can leverage group notification mechanics.

Simply creating group (not called all for unexpected conflicts with supervisorctl hardcoded all) containing all of the programms controlled by supervisord. This insures that all processes will recieve SIGTERM simultaiously upon reload/restart.

This repository contains:

• simple python script imitating some real service behavior, upon receiving SIGTERM script will sleep for 3 seconds then exit with rc0.
• dockerized supervisord/supervisorctl.
• sample config imitating 30 different realworldish commands

I've included some supervisorctl reload logs with group enabled and without as a worst case scenario.

docker compose up -d
docker compose exec supervisor supervisorctl reload

upon completion you can examine logs via:

docker compose exec supervisor cat /var/log/supervisord.log

Now can comment group, restart supervisor container, rerun benchmark and see results for yourself.