Skip to content

fix: pgbouncer_client_conn saturation flapping

Steve Xuereb requested to merge fix/pgbouncer-primary-saturation into master

What

Increase the saturation soft and hard SLO for the pgbouncer_client_conn from 80% (soft) and 85% (hard) to 90% (soft) to 95% (hard)

Why

In https://gitlab.com/gitlab-com/gl-infra/reliability/-/issues/16121 we are working on increasing the max_client_conn for the primary pgbouncers, however, this requires testing and a better understanding if we saturating CPU or if there would be a knock-on effect on the Primary CPU usage. In the mean time, the EoC is getting paged 2-3 times a day with no action to take.

We don't want to create silence for this alert because this will have negative effect to our users if we are 100% saturated so the EoC should still, be aware that we are saturating if this happens.

Reference: https://gitlab.com/gitlab-com/gl-infra/reliability/-/issues/16121

Merge request reports