Looking at https://gitlab.com/gitlab-org/gitlab-ce/issues/62876#note_183734170, and connected with our performance testing, I think it would be really helpful if we could change dynamically a number of threads and workers of Puma running in Cluster mode. The execution flow could be: 1. Change number of threads, 2. Get `/metrics` to read the latest stats to see that change was applied, as this would also force to reboot all workers, 3. Run tests, 4. Gather `/metrics` to get memory usage after the test execution.