Hello, We are using Flink 1.13.6 with Azure Kubernetes Service (AKS). Job worked fine for many months, but right after recent upgrade of AKS to 1.25.5, taskmanager started to get OOM killed every day. I suspected that this is maybe because AKS 1.25.x uses cgroups v2, which is not supported by JDK 8 < 372, but on other hand Flink explicitly sets -Xmx, so looks like taskmanager should not depend on cgroup v2 support by JDK. Does RocksDB use cgroups to calculate limits? Any ideas what else it could be? Is anyone using AKS 1.25.x or any K8s with enabled cgroups v2 with Flink, how is your experience?
Thanks, Alexey