AngersZhuuuu commented on a change in pull request #28541:
URL: https://github.com/apache/spark/pull/28541#discussion_r426988749
##########
File path: core/src/main/scala/org/apache/spark/memory/ExecutionMemoryPool.scala
##########
@@ -138,6 +140,11 @@ private[memory] class ExecutionMemoryPool(
if (toGrant < numBytes && curMem + toGrant < minMemoryPerTask) {
logInfo(s"TID $taskAttemptId waiting for at least 1/2N of $poolName
pool to be free")
lock.wait()
+ } else if (toGrant == 0 && memoryFree > 0) {
Review comment:
> That's exactly what I mean, the executor is oversubscribed by tasks.
The task OOM because the executor memory is not enough, the choice here is
either to launch worker node with more memory, or change the value of
`spark.task.cpus` to allow less tasks run in parallel.
In our case, spark.task.cpus is default value, but core per executor is 4.
Active Task not use too much memory decrease the limit of task which use more
memory. So want heavy task wait a little for other task release. A little
corner case
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
