Track termination of Unicorn workers in Prometheus
Being able to track (in Prometheus) whenever a Unicorn worker is terminated would help us determine the impact of memory usage and any memory limits. For this we'd probably have to hijack the Unicorn::WorkerKiller code in some shape or form.
The data would be stored in a separate series (e.g. "rails_worker_restarts") and would contain the following:
- hostname
- PID
- number of requests since the process started
- The process' lifetime (in seconds) since it started
Edited by Yorick Peterse