Prerequisite: I use the s3 file system, k8s HA, application mode.

I run flink clusters in kubernetes that run in application mode, I observed flink re-upload the user jar to s3 each time when I delete the jobmanager deployment in kubernetes(operator will recreate jobmanager deployment), and even though the image is not changed.

It seems like flink must upload user jar after the flink cluster recreated.

And I am confused why flink would upload the jar to s3 when the user jar is embedded in the image in application mode.

This resulted in a lot of s3 storage being wasted because I was maintaining thousands of flink clusters, all using the same user jar/image(It is a real scene)。

How do I prevent flink from this behavior, flink operator in the scale-down mechanism often causes flink clusters to receate.