rkhachatryan commented on a change in pull request #8693:
URL: https://github.com/apache/flink/pull/8693#discussion_r425893378



##########
File path: 
flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/tasks/SubtaskCheckpointCoordinatorImpl.java
##########
@@ -183,6 +235,42 @@ public void notifyCheckpointComplete(long checkpointId, 
OperatorChain<?, ?> oper
                
env.getTaskStateManager().notifyCheckpointComplete(checkpointId);
        }
 
+       @Override
+       public void notifyCheckpointAborted(long checkpointId, OperatorChain<?, 
?> operatorChain, Supplier<Boolean> isRunning) throws Exception {
+
+               if (isRunning.get()) {
+                       LOG.debug("Notification of aborted checkpoint for task 
{}", taskName);
+                       // only happens when the task always received 
checkpoints to abort but never trigger or executing.
+                       if (abortedCheckpointIds.size() >= 
maxRecordAbortedCheckpoints) {
+                               abortedCheckpointIds.pollFirst();
+                       }
+
+                       channelStateWriter.abort(checkpointId, new 
NotifiedCheckpointAbortedException(checkpointId));
+                       boolean canceled = 
asyncCheckpointRunnableRegistry.cancelAsyncCheckpointRunnable(checkpointId);
+
+                       if (!canceled) {
+                               if (checkpointId > lastCheckpointId) {
+                                       // only record checkpoints that have 
not triggered on task side.
+                                       abortedCheckpointIds.add(checkpointId);
+                               }
+                       }
+
+                       for (StreamOperatorWrapper<?, ?> operatorWrapper : 
operatorChain.getAllOperators(true)) {
+                               
operatorWrapper.getStreamOperator().notifyCheckpointAborted(checkpointId);

Review comment:
       For `notifyCheckpointComplete` it's less important because operators are 
expected to work correctly even without the notification (and 
`channelStateWriter` has only in-memory resources).
   
   ~~Another question: this is a new method and we call it only on RPC 
notification, right? Should we also call it in `abortCheckpointOnBarrier`?~~
   (in line with `notifyCheckpointComplete` we shouldn't)




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


Reply via email to