What is the recommended strategy to scale Flink jobs with async operators? I need to do multiple asynchronous Redis calls for data enrichment and my operator throughput is only at ~3k RPS. Do I add more nodes? Do I increase CPU? I've deployed a job with parallelism of 35 to a 10-node cluster and it quickly saturated the CPU. Thank you!

You probably should start profiling it, e.g. with flame graphs