turboFei commented on code in PR #2921:
URL: https://github.com/apache/celeborn/pull/2921#discussion_r1893641020
##########
client-spark/spark-3/src/main/java/org/apache/spark/shuffle/celeborn/SparkUtils.java:
##########
@@ -386,9 +402,17 @@ public static synchronized boolean
taskAnotherAttemptRunningOrSuccessful(long ta
int taskIndex = taskInfo.index();
for (TaskInfo ti : taskAttempts) {
if (ti.taskId() != taskId) {
- if (ti.successful()) {
+ if (reportedStageTaskIds.contains(ti.taskId())) {
+ LOG.info(
+ "StageId={} index={} taskId={} attempt={} another attempt {}
has reported shuffle fetch failure, ignore it.",
+ stageId,
+ taskIndex,
+ taskId,
+ taskInfo.attemptNumber(),
+ ti.attemptNumber());
Review Comment:
I have a offline discussion with Shuang.
To prevent the race condition that the task two attempts report the shuffle
fetch failure concurrently and both true returned due these attempts are both
alive at that time.
We will record the reported shuffle fetch failure taskIds in the stage.
Then exclude the TaskInfos that already reported shuffle fetch failure.
https://github.com/apache/celeborn/pull/2921/commits/549f11cacd281409bc63023ac556e45c48d4d682
Thanks @RexXiong , also cc @pan3793
--
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]
