I am using flink 1.4.0 in standalone cluster mode I have a job with a graph like a Source(parallelism 3) -> Filter(parallelism 3) -> Map(parallelism 1) -> sink(parallelism 3)
As per the understanding, max number of tasks slots should be bounded by the max parallelism, which in this case is 3. The behavior i observe is that, when more than required number of slots(say 10) are available in cluster, this job uses 5 slots. I attempted setting the same slotsharinggroup to all streamnodes, but didn't help. Is this a change in behavior in 1.4.0?. If so are there any workarounds to force the old behavior -- Thanks, -Vijay