I'm running a rails container in a digital ocean kubernetes cluster with a horizontal pod autoscaler. Sometimes when our nodes are scaled, customers encounter 503s.
- When the replica count is increased but the node count stays the same this does not happen.
- I have set the readiness probe to a quite high value (30s) and as I said the scheduling of new pods in general seems to work without errors.
Has anyone seen this behaviour or maybe solved it?
Any help or hints would be appreciated.