Hi. I just wonder Flink deployment both between Application mode VS Session mode. Given that i use K8s as resource provider for running Flink. If I just use Flink Operator to run one job per Flink session cluster, it would be the similar as using Flink Application in regards to resource isolation, Excepts, one point 1. where main() is executed, which is on Jobmanager for Application mode and on clients for Session mode

except that session + job is a much more fragile setup than an application cluster with regards to upgrades, failure handling etc

Could you explain more a bit? You mean, to upgrade Flink in session mode i need to teardown session cluster, and then create a new one while and is done automatically for application mode? What about failure what the situations are and how it is more fragile

In application mode the lifecycle of the job is closely tied to the cluster . So failure scenarios are much more tractable. You can't lose a job on an application cluster or you can't lose a cluster without direct feedback to the job