[ https://issues.apache.org/jira/browse/FLINK-5059?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15666474#comment-15666474 ]
ASF GitHub Bot commented on FLINK-5059: --------------------------------------- Github user NicoK commented on a diff in the pull request: https://github.com/apache/flink/pull/2805#discussion_r87965585 --- Diff: flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/tasks/StreamTask.java --- @@ -586,7 +588,18 @@ private boolean performCheckpoint(CheckpointMetaData checkpointMetaData) throws // yet be created final CancelCheckpointMarker message = new CancelCheckpointMarker(checkpointMetaData.getCheckpointId()); for (ResultPartitionWriter output : getEnvironment().getAllWriters()) { - output.writeEventToAllChannels(message); + final Buffer eventBuffer = EventSerializer.toBuffer(message); --- End diff -- That was the case before and I could have adapted `ResultPartitionWriter#writeEventToAllChannels()` accordingly. The question is, however, whether we want `ResultPartitionWriter` to be aware of the difference between events and buffers or offer a cleaner API that is based on buffers only... > only serialise events once in RecordWriter#broadcastEvent > --------------------------------------------------------- > > Key: FLINK-5059 > URL: https://issues.apache.org/jira/browse/FLINK-5059 > Project: Flink > Issue Type: Improvement > Components: Network > Reporter: Nico Kruber > Assignee: Nico Kruber > > Currently, > org.apache.flink.runtime.io.network.api.writer.RecordWriter#broadcastEvent > serialises the event once per target channel. Instead, it could serialise the > event only once and use the serialised form for every channel and thus save > resources. -- This message was sent by Atlassian JIRA (v6.3.4#6332)