### Summary Stale Prometheus metrics are not removed on Unicorn start-up. Only on restarts. Stale Prometheus metrics are not removed if envvar `prometheus_multiproc_dir` is missing which is true for e.g. `development` environment. ### Steps to reproduce * Remove stale metrics (`rm -fr tmp/prometheus_multiproc_dir/*.db`) * Start Unicorn (e.g via GDK `gdk run`) * Visit http://127.0.0.1/-/metrics. `http_requests_total{method="get"}` should read `2` (actually why not `1` :question:) * Reload http://127.0.0.1/-/metrics. `http_requests_total` increases * Stop Unicorn * Start Unicorn * Visit http://127.0.0.1/-/metrics. `http_requests_total` is not reset ### What is the current *bug* behavior? Stale Prometheus metrics are not removed * after starting Unicorn * in `development` environment ### What is the expected *correct* behavior? Stale Prometheus metrics are removed * after Unicorn has been started * in `development` environment ### Possible fixes * Delete stale metrics also in the [`on_worker_start`](https://gitlab.com/gitlab-org/gitlab-ce/blob/fac725c9e54bead036ca1ad5e2da0b3441aecbdf/config/initializers/7_prometheus_metrics.rb#L29) callback. * Use `Prometheus::Client.configuration.multiprocess_files_dir` instead of `ENV['prometheus_multiproc_dir']` [when deleting stale metrics](https://gitlab.com/gitlab-org/gitlab-ce/blob/fac725c9e54bead036ca1ad5e2da0b3441aecbdf/config/initializers/7_prometheus_metrics.rb#L45) ```ruby unless Rails.env.test? prometheus_multiproc_dir = Prometheus::Client.configuration.multiprocess_files_dir old_metrics = Dir[File.join(prometheus_multiproc_dir, '*.db')] FileUtils.rm_rf(old_metrics) end ``` /cc @bjk\-gitlab