Ngone51 commented on a change in pull request #29468:
URL: https://github.com/apache/spark/pull/29468#discussion_r472645226
##########
File path:
core/src/main/scala/org/apache/spark/scheduler/TaskSchedulerImpl.scala
##########
@@ -1062,25 +1062,36 @@ private[spark] class TaskSchedulerImpl(
}
def getExecutorsAliveOnHost(host: String): Option[Set[String]] =
synchronized {
- hostToExecutors.get(host).map(_.toSet)
+
hostToExecutors.get(host).map(_.filterNot(isExecutorDecommissioned)).map(_.toSet)
}
def hasExecutorsAliveOnHost(host: String): Boolean = synchronized {
- hostToExecutors.contains(host)
+ hostToExecutors.get(host)
+ .exists(executors => executors.exists(e =>
!isExecutorDecommissioned(e)))
}
def hasHostAliveOnRack(rack: String): Boolean = synchronized {
- hostsByRack.contains(rack)
+ hostsByRack.get(rack)
Review comment:
`hostToExecutors` will also be used to handle executor removal and
blacklisting stuff. In other words, a decommissioned executor could also be
removed or blacklisted at the same time. So, remove the executor from
`hostToExecutors` might also affect other places. I don't want to mix them
together to make things complex.
Besides, similar to `executorIdToRunningTaskIds`, a decommissioned executor
could also have running tasks on it. So we don't eagerly remove it from
`executorIdToRunningTaskIds` too.
----------------------------------------------------------------
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:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
