Copilot commented on code in PR #3829:
URL: https://github.com/apache/celeborn/pull/3829#discussion_r3877086917
##########
client/src/main/scala/org/apache/celeborn/client/ChangePartitionManager.scala:
##########
@@ -90,26 +90,38 @@ class ChangePartitionManager(
batchHandleChangePartitionExecutors.submit {
new Runnable {
override def run(): Unit = {
- val distinctPartitions = {
- val requestSet = inBatchPartitions.get(shuffleId)
- val locksForShuffle = locks.computeIfAbsent(shuffleId,
locksRegisterFunc)
- requests.asScala.map { case (partitionId, request) =>
- locksForShuffle(partitionId %
locksForShuffle.length).synchronized {
- if (!requestSet.contains(partitionId) &&
requests.containsKey(
- partitionId)) {
- requestSet.add(partitionId)
- Some(request.asScala.toArray.maxBy(_.epoch))
- } else {
- None
+ try {
+ val distinctPartitions = {
+ val requestSet = inBatchPartitions.get(shuffleId)
+ val locksForShuffle =
locks.computeIfAbsent(shuffleId, locksRegisterFunc)
+ requests.asScala.map { case (partitionId, request) =>
+ locksForShuffle(partitionId %
locksForShuffle.length).synchronized {
+ if (!requestSet.contains(partitionId) &&
requests.containsKey(
+ partitionId)) {
+ requestSet.add(partitionId)
+ Some(request.asScala.toArray.maxBy(_.epoch))
+ } else {
+ None
+ }
}
- }
- }.filter(_.isDefined).map(_.get).toArray
- }
- if (distinctPartitions.nonEmpty) {
- handleRequestPartitions(
- shuffleId,
- distinctPartitions,
-
lifecycleManager.commitManager.isSegmentGranularityVisible(shuffleId))
+ }.filter(_.isDefined).map(_.get).toArray
+ }
+ if (distinctPartitions.nonEmpty) {
+ handleRequestPartitions(
+ shuffleId,
+ distinctPartitions,
+
lifecycleManager.commitManager.isSegmentGranularityVisible(shuffleId))
+ }
+ } catch {
+ case e: InterruptedException =>
+ logError(
+ s"Batch handle change partition for shuffle
$shuffleId interrupted.",
+ e)
+ throw e
Review Comment:
This catch rethrows `InterruptedException`, but the interrupt status is
cleared when the exception is thrown. Re-interrupting the thread before rethrow
helps preserve cancellation/shutdown semantics for executor threads (see
similar handling in `ApplicationHeartbeater.scala`).
##########
client/src/main/java/org/apache/celeborn/client/read/DfsPartitionReader.java:
##########
@@ -283,10 +284,16 @@ public ByteBuf next() throws Exception {
results.put(Pair.of(currentChunkIndex,
Unpooled.wrappedBuffer(buffer)));
logger.debug("add index {} to results", currentChunkIndex++);
}
- } catch (Exception e) {
- logger.warn("Fetch thread is cancelled.", e);
- exception.set(e);
+ } catch (InterruptedException e) {
// cancel a task for speculative, ignore this exception
+ logger.warn("Read thread is interrupted.", e);
+ } catch (Throwable t) {
Review Comment:
The DFS fetch background thread now swallows InterruptedException without
setting `exception`. If `close()` (or shutdownNow) interrupts the fetch thread
while another thread is blocked in `next()`, the fetch thread exits and
`next()` can wait indefinitely because `checkException()` never trips and no
more results will be produced.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
