Hey folks, got a general question about Flink memory configuration: It seems that it is recommended in containerized deployment,

taskmanager.memory.process.size

is set equal to container memory. However, in Kubernetes, it seems we should always leave some container memory for system-level processes? Is this part of memory accounted in

process.size

, or we might be expecting in some cases, system memory runs out and K8s has to kill this container?

Yes you are correct, there is a risk for container level OOM. You can increase the available system memory by increasing JVM overhead. But reducing the process size would also do the trick.

Thanks Danny for the suggestion! We did observed after setting taskmanager process size equal to container memory size, some of our Flink jobs encountered some memory issues - after increasing the JVM overhead, they seem to run stable now.