Changes to Gitlab Operation

[Ben Cooksley]

Hi all,

This is a notice about a mostly "under the hood" change to how we are
running Gitlab which will hopefully improve service going forward.

Currently for those awake ih the very early European mornings you will have
noticed a period where Gitlab becomes temporarily unavailable while it is
restarted.

This restart was necessary for two things:
(a) To enable logs to be rotated; and
(b) To restart Sidekiq, which over it's runtime leaks memory

I have now switched us over to using systemd user units, which support the
proper reloading of the corresponding Gitlab processes (unlike the init
scripts) and also provide process supervision capability which has allowed
for Gitlab's Sidekiq memory killer to be enabled.

This means Gitlab itself should no longer become unavailable for 30-60
seconds every evening now, and overall performance of the system should be
improved by having several gigabytes of system memory available to cache
frequently accessed resources instead of containing Sidekiq memory leaks.

If you notice any issues, please let us know.

Thanks,
Ben
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kde-devel/attachments/20221109/feddbdcb/attachment-0001.htm>