Increased DB Connections after Upgrade to 3.14.1


#1

Hi, we run Concourse with the Helm chart on GCP with Cloud Postgres.

After the upgrade to 3.1.4.1 the workers started behaving strange. We removed the existing Helm deployment including the pvc’s and installed from scratch. Since then we have a 100% CPU utilization on the cloud postgres and we see almost tribble db connections. (from ~ 20 -> 50)


#2

We also see same the same behavior.

We thought this was somehow related to the secret parameter store we use (AWS SSM), but it seems it’s more general. We could work around this by reducing the resource check frequency from every minute to every five minutes.

We specified --resource-checking-interval=5m on the ATC command line.

It seems like that the DB connections leak in the ATC when an error occurs in the resource check logic


#3

Associated ticket https://github.com/concourse/concourse/issues/2346