You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "gaoyajun02 (Jira)" <ji...@apache.org> on 2021/10/15 03:26:00 UTC
[jira] [Updated] (SPARK-36964) Reuse CachedDNSToSwitchMapping for
yarn container requests
[ https://issues.apache.org/jira/browse/SPARK-36964?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
gaoyajun02 updated SPARK-36964:
-------------------------------
Description:
Similar to SPARK-13704, In some cases, YarnAllocator add or remove container requests can be expensive, it may call the topology script for rack awareness.
When submit a very large job in a very large Yarn cluster, the topology script may take signifiant time to run. And this blocks receiving YarnSchedulerBackend's RequestExecutors rpc calls, This request comes from spark dynamic executor allocation thread, which may blocks the ExecutorAllocationListener,
{code:text}
21/09/29 12:04:35 INFO spark-dynamic-executor-allocation ExecutorAllocationManager: Error reaching cluster manager.21/09/29 12:04:35 INFO spark-dynamic-executor-allocation ExecutorAllocationManager: Error reaching cluster manager.org.apache.spark.rpc.RpcTimeoutException: Futures timed out after [120 seconds]. This timeout is controlled by spark.rpc.askTimeout at org.apache.spark.rpc.RpcTimeout.org$apache$spark$rpc$RpcTimeout$$createRpcTimeoutException(RpcTimeout.scala:47) at org.apache.spark.rpc.RpcTimeout$$anonfun$addMessageIfTimeout$1.applyOrElse(RpcTimeout.scala:62) at org.apache.spark.rpc.RpcTimeout$$anonfun$addMessageIfTimeout$1.applyOrElse(RpcTimeout.scala:58) at scala.runtime.AbstractPartialFunction.apply(AbstractPartialFunction.scala:38) at org.apache.spark.rpc.RpcTimeout.awaitResult(RpcTimeout.scala:76) at org.apache.spark.scheduler.cluster.CoarseGrainedSchedulerBackend.requestTotalExecutors(CoarseGrainedSchedulerBackend.scala:839) at org.apache.spark.ExecutorAllocationManager.addExecutors(ExecutorAllocationManager.scala:411) at org.apache.spark.ExecutorAllocationManager.updateAndSyncNumExecutorsTarget(ExecutorAllocationManager.scala:361) at org.apache.spark.ExecutorAllocationManager.org$apache$spark$ExecutorAllocationManager$$schedule(ExecutorAllocationManager.scala:316) at org.apache.spark.ExecutorAllocationManager$$anon$1.run(ExecutorAllocationManager.scala:227) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) at java.lang.Thread.run(Thread.java:745)Caused by: java.util.concurrent.TimeoutException: Futures timed out after [120 seconds] at scala.concurrent.impl.Promise$DefaultPromise.ready(Promise.scala:259) at scala.concurrent.impl.Promise$DefaultPromise.result(Promise.scala:263) at org.apache.spark.util.ThreadUtils$.awaitResult(ThreadUtils.scala:294) at org.apache.spark.rpc.RpcTimeout.awaitResult(RpcTimeout.scala:75) ... 12 more21/09/29 12:04:35 WARN spark-dynamic-executor-allocation ExecutorAllocationManager: Unable to reach the cluster manager to request 1922 total executors!{code}
and then result in executorManagement queue backlog. e.g. some log:
{code:text}
21/09/29 12:02:49 ERROR dag-scheduler-event-loop AsyncEventQueue: Dropping event from queue executorManagement. This likely means one of the listeners is too slow and cannot keep up with the rate at which tasks are being started by the scheduler.
21/09/29 12:02:49 WARN dag-scheduler-event-loop AsyncEventQueue: Dropped 1 events from executorManagement since the application started.
21/09/29 12:02:55 INFO spark-listener-group-eventLog AsyncEventQueue: Process of event SparkListenerExecutorAdded(1632888172920,543,org.apache.spark.scheduler.cluster.ExecutorData@8cfab8f5,None) by listener EventLoggingListener took 3.037686034s.
21/09/29 12:03:03 INFO spark-listener-group-eventLog AsyncEventQueue: Process of event SparkListenerBlockManagerAdded(1632888181779,BlockManagerId(1359, --, 57233, None),2704696934,Some(2704696934),Some(0)) by listener EventLoggingListener took 1.462598355s.
21/09/29 12:03:49 WARN dispatcher-BlockManagerMaster AsyncEventQueue: Dropped 74388 events from executorManagement since Wed Sep 29 12:02:49 CST 2021.
21/09/29 12:04:35 INFO spark-listener-group-executorManagement AsyncEventQueue: Process of event SparkListenerStageSubmitted(org.apache.spark.scheduler.StageInfo@52f810ad,{...}) by listener ExecutorAllocationListener took 116.526408932s.
21/09/29 12:04:49 WARN heartbeat-receiver-event-loop-thread AsyncEventQueue: Dropped 18892 events from executorManagement since Wed Sep 29 12:03:49 CST 2021.
21/09/29 12:05:49 WARN dispatcher-BlockManagerMaster AsyncEventQueue: Dropped 19397 events from executorManagement since Wed Sep 29 12:04:49 CST 2021.
{code}
was:
Similar to SPARK-13704, In some cases, YarnAllocator add or remove container requests can be expensive, it may call the topology script for rack awareness.
When submit a very large job in a very large Yarn cluster, the topology script may take signifiant time to run. And this blocks receiving YarnSchedulerBackend's RequestExecutors rpc calls, This request comes from spark dynamic executor allocation thread, which may blocks the ExecutorAllocationListener, and then result in executorManagement queue backlog.
> Reuse CachedDNSToSwitchMapping for yarn container requests
> -----------------------------------------------------------
>
> Key: SPARK-36964
> URL: https://issues.apache.org/jira/browse/SPARK-36964
> Project: Spark
> Issue Type: Improvement
> Components: Spark Core, YARN
> Affects Versions: 3.0.3, 3.1.2
> Reporter: gaoyajun02
> Priority: Major
>
> Similar to SPARK-13704, In some cases, YarnAllocator add or remove container requests can be expensive, it may call the topology script for rack awareness.
> When submit a very large job in a very large Yarn cluster, the topology script may take signifiant time to run. And this blocks receiving YarnSchedulerBackend's RequestExecutors rpc calls, This request comes from spark dynamic executor allocation thread, which may blocks the ExecutorAllocationListener,
> {code:text}
> 21/09/29 12:04:35 INFO spark-dynamic-executor-allocation ExecutorAllocationManager: Error reaching cluster manager.21/09/29 12:04:35 INFO spark-dynamic-executor-allocation ExecutorAllocationManager: Error reaching cluster manager.org.apache.spark.rpc.RpcTimeoutException: Futures timed out after [120 seconds]. This timeout is controlled by spark.rpc.askTimeout at org.apache.spark.rpc.RpcTimeout.org$apache$spark$rpc$RpcTimeout$$createRpcTimeoutException(RpcTimeout.scala:47) at org.apache.spark.rpc.RpcTimeout$$anonfun$addMessageIfTimeout$1.applyOrElse(RpcTimeout.scala:62) at org.apache.spark.rpc.RpcTimeout$$anonfun$addMessageIfTimeout$1.applyOrElse(RpcTimeout.scala:58) at scala.runtime.AbstractPartialFunction.apply(AbstractPartialFunction.scala:38) at org.apache.spark.rpc.RpcTimeout.awaitResult(RpcTimeout.scala:76) at org.apache.spark.scheduler.cluster.CoarseGrainedSchedulerBackend.requestTotalExecutors(CoarseGrainedSchedulerBackend.scala:839) at org.apache.spark.ExecutorAllocationManager.addExecutors(ExecutorAllocationManager.scala:411) at org.apache.spark.ExecutorAllocationManager.updateAndSyncNumExecutorsTarget(ExecutorAllocationManager.scala:361) at org.apache.spark.ExecutorAllocationManager.org$apache$spark$ExecutorAllocationManager$$schedule(ExecutorAllocationManager.scala:316) at org.apache.spark.ExecutorAllocationManager$$anon$1.run(ExecutorAllocationManager.scala:227) at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511) at java.util.concurrent.FutureTask.runAndReset(FutureTask.java:308) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.access$301(ScheduledThreadPoolExecutor.java:180) at java.util.concurrent.ScheduledThreadPoolExecutor$ScheduledFutureTask.run(ScheduledThreadPoolExecutor.java:294) at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142) at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617) at java.lang.Thread.run(Thread.java:745)Caused by: java.util.concurrent.TimeoutException: Futures timed out after [120 seconds] at scala.concurrent.impl.Promise$DefaultPromise.ready(Promise.scala:259) at scala.concurrent.impl.Promise$DefaultPromise.result(Promise.scala:263) at org.apache.spark.util.ThreadUtils$.awaitResult(ThreadUtils.scala:294) at org.apache.spark.rpc.RpcTimeout.awaitResult(RpcTimeout.scala:75) ... 12 more21/09/29 12:04:35 WARN spark-dynamic-executor-allocation ExecutorAllocationManager: Unable to reach the cluster manager to request 1922 total executors!{code}
> and then result in executorManagement queue backlog. e.g. some log:
> {code:text}
> 21/09/29 12:02:49 ERROR dag-scheduler-event-loop AsyncEventQueue: Dropping event from queue executorManagement. This likely means one of the listeners is too slow and cannot keep up with the rate at which tasks are being started by the scheduler.
> 21/09/29 12:02:49 WARN dag-scheduler-event-loop AsyncEventQueue: Dropped 1 events from executorManagement since the application started.
> 21/09/29 12:02:55 INFO spark-listener-group-eventLog AsyncEventQueue: Process of event SparkListenerExecutorAdded(1632888172920,543,org.apache.spark.scheduler.cluster.ExecutorData@8cfab8f5,None) by listener EventLoggingListener took 3.037686034s.
> 21/09/29 12:03:03 INFO spark-listener-group-eventLog AsyncEventQueue: Process of event SparkListenerBlockManagerAdded(1632888181779,BlockManagerId(1359, --, 57233, None),2704696934,Some(2704696934),Some(0)) by listener EventLoggingListener took 1.462598355s.
> 21/09/29 12:03:49 WARN dispatcher-BlockManagerMaster AsyncEventQueue: Dropped 74388 events from executorManagement since Wed Sep 29 12:02:49 CST 2021.
> 21/09/29 12:04:35 INFO spark-listener-group-executorManagement AsyncEventQueue: Process of event SparkListenerStageSubmitted(org.apache.spark.scheduler.StageInfo@52f810ad,{...}) by listener ExecutorAllocationListener took 116.526408932s.
> 21/09/29 12:04:49 WARN heartbeat-receiver-event-loop-thread AsyncEventQueue: Dropped 18892 events from executorManagement since Wed Sep 29 12:03:49 CST 2021.
> 21/09/29 12:05:49 WARN dispatcher-BlockManagerMaster AsyncEventQueue: Dropped 19397 events from executorManagement since Wed Sep 29 12:04:49 CST 2021.
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org