spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From sryza <...@git.apache.org>
Subject [GitHub] spark pull request: [SPARK-4352][YARN][WIP] Incorporate locality p...
Date Thu, 25 Jun 2015 23:24:42 GMT
Github user sryza commented on a diff in the pull request:

    https://github.com/apache/spark/pull/6394#discussion_r33315886
  
    --- Diff: yarn/src/main/scala/org/apache/spark/deploy/yarn/ContainerPlacementStrategy.scala
---
    @@ -0,0 +1,206 @@
    +/*
    + * 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.deploy.yarn
    +
    +import scala.collection.mutable.ArrayBuffer
    +
    +import org.apache.hadoop.conf.Configuration
    +import org.apache.hadoop.yarn.util.RackResolver
    +
    +import org.apache.spark.SparkConf
    +
    +private[yarn] case class ContainerLocalityPreferences(nodes: Array[String], racks: Array[String])
    +
    +private[yarn] trait ContainerPlacementStrategy {
    +
    +  /**
    +   * Calculate each container's node locality and rack locality
    +   * @param numContainer number of containers to calculate
    +   * @param numLocalityAwarePendingTasks number of locality required pending tasks
    +   * @param preferredLocalityToCount a map to store the preferred host name and its required
count
    +   *
    +   * @return node localities and rack localities, each locality is an array of string,
    +   *         the length of localities is the same as number of containers
    +   */
    +  def localityOfRequestedContainers(
    +      numContainer: Int,
    +      numLocalityAwarePendingTasks: Int,
    +      preferredLocalityToCount: Map[String, Int]
    +    ): Array[ContainerLocalityPreferences]
    +}
    +
    +/**
    + * This strategy is calculating the optimal locality preferences of YARN containers by
considering
    + * the node ratio of pending tasks, number of required cores/containers and and locality
of current
    + * existed containers. The target of this algorithm is to maximize the number of tasks
that
    + * would run locally.
    + *
    + * The details of this algorithm is described as below, if we have 20 tasks which
    + * require (host1, host2, host3) and 10 tasks which require (host1, host2, host4),
    + * besides each container has 2 cores and cpus per task is 1, so the required container
number is
    + * 15, and host ratio is (host1: 30, host2: 30, host3: 20, host4: 10).
    + *
    + * 1. If requested container number (18) is more than the required container number (15):
    + *
    + * requests for 5 containers with nodes: (host1, host2, host3, host4)
    + * requests for 5 containers with nodes: (host1, host2, host3)
    + * requests for 5 containers with nodes: (host1, host2)
    + * requests for 3 containers with no locality preferences.
    + *
    + * The placement ratio is 3 : 3 : 2 : 1, and set the additional containers with no locality
    + * preferences.
    + *
    + * 2. If requested container number (10) is less than or equal to the required container
number
    + * (15):
    + *
    + * requests for 4 containers with nodes: (host1, host2, host3, host4)
    + * requests for 3 containers with nodes: (host1, host2, host3)
    + * requests for 3 containers with nodes: (host1, host2)
    + *
    + * The placement ratio is 10 : 10 : 7 : 4, close to expected ratio (3 : 3 : 2 : 1)
    + *
    + * 3. If containers are existed but no matching localities, follow the method of 1 and
2.
    + *
    + * 4. If containers are existed and some localities are matched. For example if we have
1
    + * containers on each node (host1: 1, host2: 1: host3: 1, host4: 1), and the expected
containers
    + * on each node would be (host1: 5, host2: 5, host3: 4, host4: 2),
    + * so the newly requested containers on each node would be updated to (host1: 4, host2:
4,
    + * host3: 3, host4: 1), 12 containers by total.
    + *
    + *   4.1 If requested container number (18) is more than newly required containers (12).
Follow
    + *   method 1 with updated ratio 4 : 4 : 3 : 1.
    + *
    + *   4.2 If request container number (10) is more than newly required containers (12).
Follow
    + *   method 2 with updated ratio 4 : 4 : 3 : 1.
    + *
    + * 5. If containers are existed and existing localities can fully cover the requested
localities.
    + * For example if we have 5 containers on each node (host1: 5, host2: 5, host3: 5, host4:
5),
    + * which could cover the current requested localities. This algorithm will allocate all
the
    + * requested containers with no localities.
    + */
    +private[yarn] class LocalityPreferredContainerPlacementStrategy(
    +    val sparkConf: SparkConf,
    +    val yarnConf: Configuration,
    +    val yarnAllocator: YarnAllocator) extends ContainerPlacementStrategy {
    +
    +  // Number of CPUs per task
    +  private val CPUS_PER_TASK = sparkConf.getInt("spark.task.cpus", 1)
    +
    +  // Get the required cores of locality aware task
    +  private def localityAwareTaskCores(localityAwarePendingTasks: Int): Int = {
    +    localityAwarePendingTasks * CPUS_PER_TASK
    +  }
    +
    +  // Get the expected number of locality aware containers
    +  private def expectedLocalityAwareContainerNum(localityAwarePendingTasks: Int): Int
= {
    +    (localityAwareTaskCores(localityAwarePendingTasks) +
    +      yarnAllocator.resource.getVirtualCores - 1) /
    +        yarnAllocator.resource.getVirtualCores
    +  }
    +
    +  // Update the expected locality distribution by considering the existing allocated
container
    --- End diff --
    
    Use javadoc of the form `/** */` for method headers.
    
    Also, in the doc, provide an explanation of what each entry in the returned map represents.



---
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 infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org


Mime
View raw message