[SPARK-15606] Driver hang in o.a.s.DistributedSuite on 2 core machine

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.6.2, 2.0.0
• Fix Version/s: 1.6.3, 2.0.0
• Component/s: Spark Core

Environment

AMD64 box with only 2 cores

• Assignee: Peter George Robbins
• Reporter: Peter George Robbins
• Created: 27/May/16 08:39
• Updated: 23/Jun/16 21:43
• Resolved: 02/Jun/16 17:15

Description

repeatedly failing task that crashes JVM *** FAILED ***
The code passed to failAfter did not complete within 100000 milliseconds. (DistributedSuite.scala:128)

This test started failing and DistrbutedSuite hanging following https://github.com/apache/spark/pull/13055

It looks like the extra message to remove the BlockManager deadlocks as there are only 2 message processing loop threads. Related to https://issues.apache.org/jira/browse/SPARK-13906

  /** Thread pool used for dispatching messages. */
  private val threadpool: ThreadPoolExecutor = {
    val numThreads = nettyEnv.conf.getInt("spark.rpc.netty.dispatcher.numThreads",
      math.max(2, Runtime.getRuntime.availableProcessors()))
    val pool = ThreadUtils.newDaemonFixedThreadPool(numThreads, "dispatcher-event-loop")
    for (i <- 0 until numThreads) {
      pool.execute(new MessageLoop)
    }
    pool
  }

Setting a minimum of 3 threads alleviates this issue but I'm not sure there isn't another underlying problem.

Issue Links

• links to: [Github] Pull Request #13355 (robbinspg)