Clever Runner: problems Cellar downloads ( large job caches, clone-bundles)
The clever-cloud
runner (see #196 (closed)) has a shared (S3) cache. See GitLab docs about that. This should end up in some user documentation about foss.h.n.
Projects with lots of parallel jobs will probably need to make sure that their jobs don't race to update the cache. Upstream GitLab general advice is to have a separate stage for cache update (with no concurrency) and the parallel jobs only pulling the cache in a later stage.
Also we currently experience problems with very large caches on the S3 backend, such as in line 66 of this job. This is currently under investigation by network administrators. The same problem also occurred at least once with the clone-bundle for Tryton
Edited by Georges Racinet