Hi team. We are using Flink Kubernetes Operator to deploy our Flink jobs, but sometimes we are acing issues that task managers are NOT being created when the “SpecChanged”. The deployment stays at the

Reconciling

status. Most of the time we would have to

kubectl delete FlinkDeployment

to destroy the job and redeploy it, but the consequence of doing that is the state and checkpoints are lost. Does anyone know how to avoid/resolve the issue?

I’ve encountered this also. What happens when you do a

kubectl delete deployment …

instead of

FlinkDeployment

? With our offset initializer config this did honour the existing checkpoints. This worked in some cases for us.

oooh… that might work. Gonna try that out. Thanks for sharing!