Upgrading machine type to C2 for Redis CPU bound workloads
In https://gitlab.com/gitlab-com/gl-infra/scalability/-/issues/230 we rolled out the change to our Redis Sidekiq cluster to switch the machine type to C2 with some dramatic (positive) results as noted in! https://gitlab.com/gitlab-com/gl-infra/scalability/-/issues/230#note_312403063
@davis_townsend mentioned in the Infrastructure call that the time to change machine types is now, and on a fairly short timeline.
I propose that we roll out C2 machine type to all CPU bound workloads, but at minimum all of our Redis fleet (with an exception of sentinels maybe):
redis-03-db-gprd Intel Haswell n1-standard-8
redis-cache-03-db-gprd Intel Haswell n1-highmem-16
redis-01-db-gprd Intel Haswell n1-standard-8
redis-cache-01-db-gprd Intel Haswell n1-highmem-16
redis-02-db-gprd Intel Haswell n1-standard-8
redis-cache-02-db-gprd Intel Haswell n1-highmem-16
redis-cache-sentinel-03-db-gprd Intel Haswell n1-standard-1
redis-cache-sentinel-01-db-gprd Intel Haswell n1-standard-1
pubsub-duplicate-redis-inf-gprd Intel Haswell n1-standard-1
redis-cache-sentinel-02-db-gprd Intel Haswell n1-standard-1
Cost comparison has been done in https://gitlab.com/gitlab-com/gl-infra/scalability/-/issues/230#cost and concluded that cost difference was insignificant.
Edited by Rachel Nienaber