@pnowojski , I am supposed to cover the broadcast scenario benchmark later
because of #6417, thanks for adding this timely. :)
The codes look good to me, and I only have two additional concerns:
1. Actually there are two ways to realize broadcast semantics. The first way is
via `RecordWriter#broadcastEmit()` as you realized in this PR. The other way is
via `RecordWriter#emit(BroadcastChannelSelector)`, and in current
`StreamNetworkBenchmarkEnvironment#createRecordWriter` we only define the
`RoundRobinChannelSelector` for the writer. If we make the `ChannelSelector` as
a parameter, we can also support other partitioner modes, such as `rebalance`,
`forward`, etc. How do you consider these two ways?
2. Do we need make the var `broadcastMode` as `@Param({"true","false"})` for
covering the existing tests in `StreamNetworkThroughputBenchmarkTest`? Because
currently there are no tests in the
`StreamNetworkBroadcastThroughputBenchmarkTest`.
[ Full content available at: https://github.com/apache/flink/pull/6697 ]
This message was relayed via gitbox.apache.org for [email protected]
