https://flink.apache.org/ logo
Join Slack
Powered by
Is there a way for me to instruct `flink-k8s-opera...
# random
z
Is there a way for me to instruct 
flink-k8s-operator
to use the latest deployment template to reconcile a flink deployment when that deployment is not in a stable state? Here is the problem I am struggling. I have a flink deployment yaml X and the flink operator deploys it without any issues. Then I made some edits to X called it X1 and deployed it again. However, there are some errors in X1, which caused the deployment to fail to start. As a result, the deployment is in the forever 
reconciling
state. Then I submit template X2, which contains all the fixes for X1, to the flink operator. Because the flink deployment is still in the 
reconciling
state, it won’t handle the new request.
g
Which operator version are you using?
It depends a lot on the specific error and upgradeMode you have set. If you have a stateful upgradeMode then the operator needs to be able to figure out the latest checkpoint / create a savepoint to perform the deployment
You always have the option to delete the current FlinkDeployment and create a new one while setting the initialSavepointPath
However we are always working on improving the behavior to avoid cases where the operator cannot reconcile a resource. You could also build the operator version from the current main branch that contains a very nice improvement for this with startup probes for the jobmanager
z
release-1.4
is the version we are using.
The issue is that a misconfigured 
flink deployment
can get stuck in the 
reconciling
state forever.
The k8s operator doesn’t seem to be able to handle new updates for a flink deployment that is not in the 
stable
state.
I don’t know whether the above statement is true or not, but it looks like that from my observations.
g
Most of these problems are fully eliminated in the latest main branch . You can also upgrade to 1.5 that also fixes some cases you might be hitting
👍 1
Open in Slack
PreviousNext
Powered by