2016-11-21: 11.0.3 exception request for gitlab-org/omnibus-gitlab!2582

Exception request

Why it needs to be picked

We lose Prometheus Unicorn metrics on certain nodes whenever we send a HUP signal to it because of the way we delete the Prometheus memory-mapped files. This MR changes how and when these files get deleted.

This is currently blocking @yorickpeterse from being able to deprecate more InfluxDB dashboards in favor of Prometheus.

Potential negative impact of picking

This change adds file removal code before_exec block to the unicorn configuration, which is a non-trivial change. Before this file removal was happening in the runit script, which runs as root. The user may not have permission to remove these files, in which case it's possible that the files remain indefinitely. Corrupted files owned by root, for example, may cause the worker to fail.

Release manager sign-off

(Assign this issue to RMs managing this release. See https://about.gitlab.com/release-managers/)

The release manager needs to provide initial approval for this exception request.

Sign-off for patch release (delete this section if RC)

Upon initial approval, the release manager will then ping and assign one more role for final approval:

After approval

Check that the following is accurate before closing this issue:

Edited by Filipa Lacerda