taegeonum commented on a change in pull request #153: [NEMO-245,247] Handle
watermark in OutputWriter and Implement unbounded word count example
URL: https://github.com/apache/incubator-nemo/pull/153#discussion_r232482195
##########
File path:
runtime/executor/src/main/java/org/apache/nemo/runtime/executor/task/MultiThreadParentTaskDataFetcher.java
##########
@@ -96,14 +100,37 @@ private void fetchDataLazily() {
final List<CompletableFuture<DataUtil.IteratorWithNumBytes>> futures =
readersForParentTask.read();
numOfIterators = futures.size();
+ if (numOfIterators > 1) {
+ inputWatermarkManager = new MultiInputWatermarkManager(numOfIterators,
new WatermarkCollector());
+ } else {
+ inputWatermarkManager = new SingleInputWatermarkManager(new
WatermarkCollector());
+ }
+
futures.forEach(compFuture -> compFuture.whenComplete((iterator,
exception) -> {
// A thread for each iterator
queueInsertionThreads.submit(() -> {
if (exception == null) {
// Consume this iterator to the end.
while (iterator.hasNext()) { // blocked on the iterator.
final Object element = iterator.next();
- elementQueue.offer(element);
+
+ if (LOG.isDebugEnabled()) {
+ LOG.debug("Receive data : {}", element);
+ }
+
+ if (element instanceof WatermarkWithIndex) {
+ // watermark element
+ // the input watermark manager is accessed by multiple threads
+ // so we should synchronize it
+ synchronized (inputWatermarkManager) {
+ final WatermarkWithIndex watermarkWithIndex =
(WatermarkWithIndex) element;
+ inputWatermarkManager.trackAndEmitWatermarks(
Review comment:
The InputWatermarkManager is also used for multiple edges in intra task but
we don't need synchronized in the intra task because one thread accesses the
input watermark manager. So I think it would be good to put the synchronized
here.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
