Here’s a queston on state management and memory setup. According to the documentation taskmanagers shoudl be set up differently for RocksDB and HapsMap state backends. Does this mean it’s suboptimal to run a mix of jobs using the two backends on the same flink instance? Also, any thoughts on the feasibility of having different operators use different state backend (I don’t think it’s supported right now, but I can image it being beneficial in certain scenarios)?