You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@emilianomoscato thanks so much for raising this, it will really help others!
I believe the issue is:
the readinessProbe is timing out (or otherwise failing)
the pgbouncer pod, becomes "unready"
airflow-pgbouncer.{NAMESPACE}.svc.cluster.local not stops resolving to an IP (because there are no ready pods)
the livenessProbe having "connection refused", restarting the pod
(possibly because so many connections are created after the restart, it overloads the pod, causing the readinessProbe to timeout, starting the whole process over again)
I have raised issue #526 to propose replacing the readinessProbe with a startupProbe, which should hopefully address this issue.
Checks
Chart Version
8.5.3
Kubernetes Version
Helm Version
Description
When deploying chart to our AWS EKS cluster, deploy does not end successfully and keeps in CrashLoopBackOff state.
Disabling Pgbouncer solves the issue.
Relevant Logs
Custom Helm Values
The text was updated successfully, but these errors were encountered: