Hi, We deploy Flink in Session Cluster mode with ...
# troubleshooting
a
Hi, We deploy Flink in Session Cluster mode with the FlinkDeployment CRD and need guidance on two issues: a. Separate resource requests vs. limits Can we specify different values for requests and limits? At the moment, the CPU- and memory-values defined for the JobManager or TaskManager are applied to both; I don’t see a separate setting. a. TaskManager out-of-memory (Metaspace) errors Some TaskManager pods crash with Metaspace OOM. We could raise the cluster-level memory, but a single session cluster may run many jobs, and only a few need the extra memory. For example, in a cluster running 20 jobs, 5 of them trigger OOM, the affected pod is killed, a new one is created, and the job restarts. How can we isolate or control these heavy jobs so they don’t fail with out of memory issue? Currently we don’t want to move them to a new session cluster. Thank You