We want to call master_restart callback on start too - this callback
is used for cleaning up prometheus files. It has been added to
Omnibus already:
https://gitlab.com/gitlab-org/omnibus-gitlab/merge_requests/3350
So make sure that Unicorn example configs are up-to-date and that
the same callback is called for Puma too.
There is no reason to monitor transition states so we ignore
ready and active states. We can get ratio of completed vs failed
requests from final states.
PumaWorkerKiller is used for periodically checking and killing
workers (the biggest one) if overall memory reaches specified
limit. This metric allows us to watch number of killed workers.
The Puma Worker Killer checks the total RAM used by both the master
and worker processes. Bump the limits to N+1 instead of N workers
to account for this.
This allows us (and others) to test drive Puma without it affecting all
users. Puma can be enabled by setting the environment variable
"EXPERIMENTAL_PUMA" to a non empty value.
