Hi All, My flink app (k8s) receives ~1000TPS from kafka source and it just map and sink into another kafka topic. Its capacity is 1 taskmanager, 2cpu and 2G memory for both JM and TM. When I see its memory usage, it gradually gets increased over time but the cpu load is in idle state after some time of job startup. I am wondering why the memory usage keeps on growing when there is no special operations. Any input about this? Thanks in advance.

It is very common for JVM process memory to keep growing since there are continuous new object generated in young generation and promoted to old generation.

Sorry, I forgot to add one important point. To debug this case, I disabled the checkpoint and retried the same use case. Afterwards, the memory usage seems pretty good with same ~1000TPS. I am using Rocksdb as a state backend with incremental checkpointing and use to persist checkpoints and save points in k8s pvc.

Have u dumped the TM’s heap for analysis?