This adds counters for build artifacts and LFS objects, and moves
the preexisting repository_size and commit_count from the projects
table into a new project_statistics table.
The counters are displayed in the administration area for projects
and groups, and also available through the API for admins (on */all)
and normal users (on */owned)
The statistics are updated through ProjectCacheWorker, which can now
do more granular updates with the new :statistics argument.
This refactors repository caching so it's possible to selectively
refresh certain caches, instead of just expiring and refreshing
everything.
To allow this the various methods that were cached (e.g. "tag_count" and
"readme") use a similar pattern that makes expiring and refreshing
their data much easier.
In this new setup caches are refreshed as follows:
1. After a commit (but before running ProjectCacheWorker) we expire some
basic caches such as the commit count and repository size.
2. ProjectCacheWorker will recalculate the commit count, repository
size, then refresh a specific set of caches based on the list of
files changed in a push payload.
This requires a bunch of changes to the various methods that may be
cached. For one, data should not be cached if a branch used or the
entire repository does not exist. To prevent all these methods from
handling this manually this is taken care of in
Repository#cache_method_output. Some methods still manually check for
the existence of a repository but this result is also cached.
With selective flushing implemented ProjectCacheWorker no longer uses an
exclusive lease for all of its work. Instead this worker only uses a
lease to limit the number of times the repository size is updated as
this is a fairly expensive operation.
This changes ProjectCacheWorker.perform_async so it only schedules a job
when no lease for the given project is present. This ensures we don't
end up scheduling hundreds of jobs when they won't be executed anyway.
Dumping too many jobs in the same queue (e.g. the "default" queue) is a
dangerous setup. Jobs that take a long time to process can effectively
block any other work from being performed given there are enough of
these jobs.
Furthermore it becomes harder to monitor the jobs as a single queue
could contain jobs for different workers. In such a setup the only
reliable way of getting counts per job is to iterate over all jobs in a
queue, which is a rather time consuming process.
By using separate queues for various workers we have better control over
throughput, we can add weight to queues, and we can monitor queues
better. Some workers still use the same queue whenever their work is
related. For example, the various CI pipeline workers use the same
"pipeline" queue.
This commit includes a Rails migration that moves Sidekiq jobs from the
old queues to the new ones. This migration also takes care of doing the
inverse if ever needed. This does require downtime as otherwise new jobs
could be scheduled in the old queues after this migration completes.
This commit also includes an RSpec test that blacklists the use of the
"default" queue and ensures cron workers use the "cronjob" queue.
Fixesgitlab-org/gitlab-ce#23370
This ensures ProjectCacheWorker jobs for a given project are performed
at most once per 15 minutes. This should reduce disk load a bit in cases
where there are multiple pushes happening (which should schedule
multiple ProjectCacheWorker jobs).
Yorick Peterse