sandeep-katta edited a comment on issue #23697: [SPARK-26758][core]Idle Executors are not getting killed after spark.dynamiAllocation.executorIdleTimeout value URL: https://github.com/apache/spark/pull/23697#issuecomment-459021616 > Just to confirm, they won't be removed at 60s but will be removed at 120s, but of course we expect them to be removed at 60s, right? > > Yeah that was kind of my point, can we just set initializing = true somewhere else to fix this? That's not hard either, but I kind of think this change is an OK way to do it for other reasons (i.e. I don't quite get why you'd remove idle executors right _after_ checking if the number of executors is right). No they wont be removed at 120s also, `removeTimes` is the one which holds the executors Id with timestamp, at 60s all the executors are expired and it is cleared from `removeTimes` map.So at 120s this Map is empty,so executors will not be removed May be if you look at the driver logs below you can get the point why I am checking target executors before removing idle executors 2019-01-30 14:16:39,134 | INFO | spark-dynamic-executor-allocation | Request to remove **executorIds: 2, 1** | org.apache.spark.internal.Logging$class.logInfo(Logging.scala:54) 2019-01-30 14:16:39,135 | DEBUG | spark-dynamic-executor-allocation | **Not removing idle executor 2 because there are only 3 executor(s) left (number of executor target 3)** | org.apache.spark.internal.Logging$class.logDebug(Logging.scala:58) Since `initializing` flag was still true `updateAndSyncNumExecutorsTarget` is not resetting the numExecutorsTarget to the actual executors needed **Or I can do 1 more fix.** call `scheduleTask` with some delay like below executor.scheduleWithFixedDelay(scheduleTask, intervalMillis, intervalMillis, TimeUnit.MILLISECONDS) So the 1st call to `schedule` comes at 60s, then set `initializing` to false at the start of this function,then this problem will be solved
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
