Ngone51 commented on a change in pull request #34834:
URL: https://github.com/apache/spark/pull/34834#discussion_r769237528
##########
File path: core/src/main/scala/org/apache/spark/scheduler/TaskSetManager.scala
##########
@@ -960,19 +960,27 @@ private[spark] class TaskSetManager(
assert (null != failureReason)
taskSetExcludelistHelperOpt.foreach(_.updateExcludedForFailedTask(
info.host, info.executorId, index, failureReason))
- numFailures(index) += 1
- if (numFailures(index) >= maxTaskFailures) {
- logError("Task %d in stage %s failed %d times; aborting job".format(
- index, taskSet.id, maxTaskFailures))
- abort("Task %d in stage %s failed %d times, most recent failure:
%s\nDriver stacktrace:"
- .format(index, taskSet.id, maxTaskFailures, failureReason),
failureException)
- return
+ if (!successful(index)) {
Review comment:
In this way, we'd miss the task failure info once, though `numFailures`
isn't used anywhere else. It's more straightforward to me that we record the
correct num failures first and then check whether to abort depending on
`successful(index)`.
BTW, in the case of executor loss, the succeeded task can be re-added to the
pending list. If we reset `numFailures(index)` before, the task could have
another `maxTaskFailures` times to fail. But if not reset, the task only has
`maxTaskFailures` - `numFailures(index)` times to fail. This depends on how we
define the `maxTaskFailures`, especially whether it's for the continuous count.
However, the current doc doesn't give a clear description of this.
cc @jiangxb1987 @mridulm
--
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]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
