Github user brad-kaiser commented on a diff in the pull request: https://github.com/apache/spark/pull/19041#discussion_r179551630 --- Diff: core/src/main/scala/org/apache/spark/CacheRecoveryManager.scala --- @@ -0,0 +1,201 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark + +import java.util.concurrent.TimeUnit + +import scala.collection.JavaConverters._ +import scala.concurrent.{ExecutionContext, Future, Promise} +import scala.util.Failure + +import com.google.common.cache.CacheBuilder + +import org.apache.spark.CacheRecoveryManager.{DoneRecovering, KillReason, Timeout} +import org.apache.spark.internal.Logging +import org.apache.spark.internal.config.DYN_ALLOCATION_CACHE_RECOVERY_TIMEOUT +import org.apache.spark.rpc.RpcEndpointRef +import org.apache.spark.storage.BlockManagerId +import org.apache.spark.storage.BlockManagerMessages._ +import org.apache.spark.util.ThreadUtils + +/** + * Responsible for asynchronously replicating all of an executor's cached blocks, and then shutting + * it down. + */ +private class CacheRecoveryManager( + blockManagerMasterEndpoint: RpcEndpointRef, + executorAllocationManager: ExecutorAllocationManager, + conf: SparkConf) + extends Logging { + + private val forceKillAfterS = conf.get(DYN_ALLOCATION_CACHE_RECOVERY_TIMEOUT) + private val threadPool = ThreadUtils.newDaemonCachedThreadPool("cache-recovery-manager-pool") + private implicit val asyncExecutionContext = ExecutionContext.fromExecutorService(threadPool) + private val scheduler = + ThreadUtils.newDaemonSingleThreadScheduledExecutor("cache-recovery-shutdown-timers") + private val recoveringExecutors = CacheBuilder.newBuilder() + .expireAfterWrite(forceKillAfterS * 2, TimeUnit.SECONDS) + .build[String, String]() + + /** + * Start the recover cache shutdown process for these executors + * + * @param execIds the executors to start shutting down + * @return a sequence of futures of Unit that will complete once the executor has been killed. + */ + def startCacheRecovery(execIds: Seq[String]): Future[Seq[KillReason]] = { + logDebug(s"Recover cached data before shutting down executors ${execIds.mkString(", ")}.") + val canBeRecovered: Future[Seq[String]] = checkMem(execIds) + + canBeRecovered.flatMap { execIds => + execIds.foreach { execId => recoveringExecutors.put(execId, execId) } + Future.sequence(execIds.map { replicateUntilTimeoutThenKill }) + } + } + + def replicateUntilTimeoutThenKill(execId: String): Future[KillReason] = { + val timeoutFuture = returnAfterTimeout(Timeout, forceKillAfterS) + val replicationFuture = replicateUntilDone(execId) + + Future.firstCompletedOf(List(timeoutFuture, replicationFuture)).andThen { --- End diff -- Ah, so my original thought was that whatever future finished first, the replication future or the timeout future, the other would complete harmlessly in the background. I didn't realize that the future returned by firstCompletedOf wouldn't complete until both were done. I will fix this. Thanks
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org