[ https://issues.apache.org/jira/browse/FLINK-4975?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15647868#comment-15647868 ]
ASF GitHub Bot commented on FLINK-4975: --------------------------------------- Github user tillrohrmann commented on a diff in the pull request: https://github.com/apache/flink/pull/2754#discussion_r87013382 --- Diff: flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/io/BarrierBuffer.java --- @@ -135,19 +138,20 @@ else if (next.getEvent().getClass() == CheckpointBarrier.class) { processBarrier((CheckpointBarrier) next.getEvent(), next.getChannelIndex()); } } + else if (next.getEvent().getClass() == CancelCheckpointMarker.class) { + processCancellationBarrier((CancelCheckpointMarker) next.getEvent()); + } else { if (next.getEvent().getClass() == EndOfPartitionEvent.class) { - numClosedChannels++; - // no chance to complete this checkpoint - releaseBlocks(); + processEndOfPartition(next.getChannelIndex()); } return next; } } else if (!endOfStream) { // end of input stream. stream continues with the buffered data endOfStream = true; - releaseBlocks(); + releaseBlocksAndResetBarriers(); return getNextNonBlocked(); --- End diff -- Can't we replace the tail recursive call by staying in the `while` loop? The same could then apply to the other tail recursive call after calling `completeBufferedSequence` (replacing the recursive call with a `continue`). > Add a limit for how much data may be buffered during checkpoint alignment > ------------------------------------------------------------------------- > > Key: FLINK-4975 > URL: https://issues.apache.org/jira/browse/FLINK-4975 > Project: Flink > Issue Type: Improvement > Components: State Backends, Checkpointing > Affects Versions: 1.1.3 > Reporter: Stephan Ewen > Assignee: Stephan Ewen > Fix For: 1.2.0, 1.1.4 > > > During checkpoint alignment, data may be buffered/spilled. > We should introduce an upper limit for the spilled data volume. After > exceeding that limit, the checkpoint alignment should abort and the > checkpoint be canceled. -- This message was sent by Atlassian JIRA (v6.3.4#6332)