reduce the size of the hot nodes fleet
https://gitlab.com/gitlab-com/gl-infra/infrastructure/-/issues/10227 should bring in memory optimizations. Once that is in production, we can then reduce the size of the hot nodes fleet.
The performance indicators that we discussed that should indicate saturation include:
- cpu utilization on the hot nodes
- memory related metrics on the hot nodes (removing nodes means more shards will be scheduled per node)
- indexing latency
Edited by Michal Wasilewski