[DRE-maint] Bug#1029623: gitlab: Services are not restarted in case of a missing database server during startup

[Lars Kruse]

Package: gitlab
Version: 15.2.3+ds1-2~fto11+1
Severity: normal

Dear Maintainer,

recently I noticed, that the following gitlab-related services remained
in the "failed" state after booting the VM:
* gitlab-puma
* gitlab-sidekiq
* gitlab-workhorse

The problem was caused by the absence of the database server (it was
just booting up in a separate VM).

The same applies probably to `gitlab-mailroom`, but that service was
probably just lucky or arrived late.

Currently the `Restart` directive in the service files of the above
services is set to `Restart=on-abnormal`.
Thus, non-zero exit codes do not trigger a restart.

But the missing database connection caused exactly such a non-zero
exitcode.

Thus I would suggest to use `Restart=on-failure` for these (all?)
services, instead.

Thank you for maintaining the gitlab package!

Cheers,
Lars