Github user squito commented on a diff in the pull request: https://github.com/apache/spark/pull/8760#discussion_r45791721 --- Diff: core/src/main/scala/org/apache/spark/scheduler/BlacklistStrategy.scala --- @@ -0,0 +1,146 @@ +/* + * 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.scheduler + +import scala.collection.mutable +import org.apache.spark.Logging +import org.apache.spark.SparkConf +import org.apache.spark.util.SystemClock +import org.apache.spark.util.Clock + +/** + * The interface to determine executor blacklist and node blacklist. + */ +private [scheduler] trait BlacklistStrategy { + /** Defined a time interval to expire failure information of executors */ + val expireTimeInMilliseconds: Long + + /** Return executors in blacklist which are related to given taskIndex */ + def getExecutorBlacklist( + executorIdToFailureStatus: mutable.HashMap[String, FailureStatus], + atomTask: BlacklistAtomTask, clock: Clock): Set[String] + + /** Return all nodes in blacklist */ + def getNodeBlacklist( + executorIdToFailureStatus: mutable.HashMap[String, FailureStatus]): Set[String] + + /** + * Default implementation to remove failure executors from HashMap based on given time period. + * The return value identity whether or not it updated anything + */ + def expireExecutorsInBlackList( + executorIdToFailureStatus: mutable.HashMap[String, FailureStatus], clock: Clock): Boolean = { + val now = clock.getTimeMillis() + val expiredKey = executorIdToFailureStatus.filter { + case (executorid, failureStatus) => { + (now - failureStatus.updatedTime) >= expireTimeInMilliseconds + } + }.keySet + + if (expiredKey.isEmpty) { + false + } else { + executorIdToFailureStatus --= expiredKey + true + } + } +} + +/** + * This strategy is simply based on given threshold and is taskId unrelated. An executor will be + * in blacklist, if it failed more than "maxFailureTaskNumber" times. A node will be in blacklist, + * if there are more than "maxBlackExecutorNumber" executors on it in executor blacklist. + * + * In this case, provided taskId will be ignored. The benefit for taskId unrelated strategy is that + * different taskSets can learn experience from other taskSet to avoid allocating tasks on + * problematic executors. + */ +private[scheduler] class SimpleStrategy( + maxFailureTaskNumber: Int, + maxBlackExecutorNumber: Int, --- End diff -- can you change these params to `maxFailedTasks` and `maxBlacklistedExecutors`, and change the comment to: This strategy adds an executor to the blacklist for *all* tasks when the executor has too many task failures. An executor is placed in the blacklist when there are more than [[maxFailedTasks]] failed tasks. Furthermore, all executors in one node are put into the blacklist if there are more than [[maxBlacklistedExecutors]] blacklisted executors on one node. The benefit of this strategy is that different taskSets can learn experience from other taskSet to avoid allocating tasks on problematic executors.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org