Increase the pgbouncer async connection pool
The connection pool for Sidekiq to the primary database has been close to saturation for a while back (https://gitlab.com/gitlab-com/gl-infra/capacity-planning/-/issues/32).
Since then, we've decreased the connection pool twice:
- https://gitlab.com/gitlab-com/gl-infra/chef-repo/-/merge_requests/1679
- https://gitlab.com/gitlab-com/gl-infra/chef-repo/-/merge_requests/1617
This was done as part of production#6440 (comment 917448371) when we were redirecting some traffic from the main-primary to the ci-primary.
This has caused some apdex degradations in shards with lower concurrency: production#6880 (comment 919554551). Those are more affected by not getting a connection, because there's less work ongoing that the GVL can be given too, stalling the jobs running on those shards.
In the short term, I think we should increase the connection pool for Sidekiq again.