More CPU for prometheus gprd/regional
There is a correlation between CPU throttling of these pods, and slow rule evaluation peaks. See gitlab-com/gl-infra/production#3853 (closed) for context.
Note that as per gitlab-com/gl-infra/production#3853 (comment 522203297), these pods are and will continue to be scheduled on the sidekiq-urgent-other pool, not the default pool as their preferred (but not required) affinity would suggest. The nodes in this pool are the only ones that can accomodate such large pods: https://console.cloud.google.com/kubernetes/clusters/details/us-east1/gprd-gitlab-gke/nodes?project=gitlab-production
We should address this problem separately.
Edited by Craig Furman