Broken storage used to be used to test situations where the Git storage
wasn't being reached. These days we can just mock the Gitaly response.
But given the broken storage is removed now, Gitaly can take over
control of the storage being reachable. If it's not, Gitaly won't boot.
That's nice for situations where a disk wasn't mounted for instance.
Gitaly MR: https://gitlab.com/gitlab-org/gitaly/merge_requests/675
We saw in production that DispatchWorker was running about twice an hour,
which would schedule twice as many jobs as it should.
For some reason, BatchWorker was running 1000 times per hour, possibly
due to Sidekiq RSS kills that caused these jobs to restart.
Adding an ExclusiveLease prevents these jobs from running more
than they should.
Relates to https://gitlab.com/gitlab-com/infrastructure/issues/4526