Solution to clean up after a killed evobench-run daemon

[pflanze]

Currently, when killing the evobench-run daemon (currently via ctl-c in the screen session, in the future by way of restarting it via daemon infrastructure), it can leave behind:

• temporary files in /dev/shm

• in the night, when it ran "sudo /opt/silo-benchmark-ci/root/other-activity stop", nothing will run "sudo /opt/silo-benchmark-ci/root/other-activity start" again, meaning the staging services will stay down if someone, e.g. me, forcibly stops or restarts the daemon at night during some testing (this happened Mon 2025-08-18 to Felix, Fabian had to contact me to learn what happened)

Nothing high priority.

But then I also realized that currently, when the whole server restarts, e.g. due to security updates, the same thing will happen, and the hack to disable the daemon restart hack is just a file /opt/lapis/container-restart-hack.disabled that will linger across the restarts. This is perhaps more important.

Have to think through what the best solution will be.

[pflanze]

The solution should probably be:

For the activity stop: instead of stop & start actions, open a pipe then fork a daemon (i.e. separate session id via setsid) that exec's the user script, which then can do the start and stop, and in our case take a flock based lock (is let go across a reboot) instead of the existence of the "..disabled" file (that stays around); the user script reads&blocks on its end of the pipe; when evobench-run is done it closes the pipe, making the user script execute the end action. The same happens automatically when evobench-run is killed.

The tempfile cleaning is another forked process, same principle, that deletes the file when the pipe is closed.