Lower `UpdateAllMirrorsWorker` rescheduling threshold to ensure queue is always as full as can be
Spun off of gitlab-com/gl-infra/production#848 (comment 175335813):
Right now, UpdateAllMirrorsWorker
will only reschedule itself if there are enough "mirrors ready to sync" to fill up all free slots in the Sidekiq queue based on the configured max capacity, which means we are not actually keeping the queue as full as possible all the time, and are not fully utilizing the Sidekiq threads dedicated to mirroring, resulting in more "overdue" mirrors than necessary.
We can change the logic in Gitlab::Mirror.reschedule_immediately?
to only require, say, 10% of the queue capacity, to be able to be filled up, or we could replace the existing "Capacity threshold" setting, described as ”Minimum capacity to be available before we schedule more mirrors preemptively.“, with a new one described as ”Minimum number of mirrors to sync to be available before we schedule more mirrors preemptively.”, since this setting is currently set to 1 on GitLab.com anyway.