-
Bug
-
Resolution: Done
-
Critical
-
None
-
None
-
3
-
False
-
None
-
False
-
This fix removes the liveness probe in the application controller statefulset. This solves the problems of multiple restarts of application controllers in a large set of applications caused due to unresponsiveness of liveness probes.
-
GITOPS Sprint 222, GITOPS Sprint 223, GITOPS Sprint 224
Upstream has removed the liveness probe from the application controller's statefulset manifest with https://github.com/argoproj/argo-cd/pull/9557
Reasoning behind this is:
In cases when the controller is overloaded reconciling large queues restarting the Pod is more harmful than letting it running. In discussion with @alexmt we understand that liveness probe should be removed from the controller to prevent it from being restarted.
We should remove the liveness probe as well, the reasoning makes sense, and I have observed it multiple times now with a large set of apps that the application controller will be constantly restarted due to overload/unresponsiveness of the liveness probe.
This should also be backported to 1.4, 1.5 and 1.6 patch releases.
- links to
- mentioned on