Skip to content

Stop recording connection pool metrics for load balancing hosts

Sean McGivern requested to merge stop-recording-load-balancing-metrics into master

What does this MR do?

The cardinality of these is simply too high: for a single (Rails) host on GitLab.com, we have 384 elements due to a combination of:

  1. Puma threads.
  2. Load balancing hosts.
  3. Load balancing ports.

Realistically, the saturation for a load balancing connection pool should never be higher than for ActiveRecord::Base anyway.

See https://thanos-query.ops.gitlab.net/graph?g0.range_input=2w&g0.max_source_resolution=0s&g0.expr=gitlab_database_connection_pool_size%7Bclass%3D%22Gitlab%3A%3ADatabase%3A%3ALoadBalancing%3A%3AHost%22%2C%20fqdn%3D%22web-cny-06-sv-gprd.c.gitlab-production.internal%22%7D&g0.tab=1 for an example of that. Also see gitlab-com/gl-infra/scalability#153 (comment 354487739).

Edited by Sean McGivern

Merge request reports