Github user davies commented on a diff in the pull request:
https://github.com/apache/spark/pull/12899#discussion_r62708707
--- Diff: core/src/main/scala/org/apache/spark/scheduler/Task.scala ---
@@ -155,7 +155,13 @@ private[spark] abstract class Task[T](
*/
def collectAccumulatorUpdates(taskFailed: Boolean = false):
Seq[AccumulatorV2[_, _]] = {
if (context != null) {
- context.taskMetrics.accumulators().filter { a => !taskFailed ||
a.countFailedValues }
+ context.taskMetrics.internalAccums.filter { a =>
+ // RESULT_SIZE accumulator is always zero at executor, we need to
send it back as its
+ // value will be updated at driver side.
+ !a.isZero || a.name == Some(InternalAccumulator.RESULT_SIZE)
+ // zero value external accumulators may still be useful, e.g.
SQLMetrics, we should not filter
--- End diff --
If there is no spilling, we could say the size of spilling is not define
(null for unknown). We also have total value, we could know that how many task
had spilled. Right now, we can't know how many tasks had spilled, actually it's
worse.
I don't think it's wrong.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
