[SPARK-21408] Default RPC dispatcher thread pool size too large for small executors

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.3.0
• Fix Version/s: 2.3.0
• Component/s: Spark Core
• Assignee: Marcelo Masiero Vanzin
• Reporter: Marcelo Masiero Vanzin
• Created: 13/Jul/17 23:00
• Updated: 18/Jul/17 20:36
• Resolved: 18/Jul/17 20:36

Description

This is the code that sizes the RPC dispatcher thread pool:

  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")

That is based on the number of available cores on the host, instead of the number of cores the executor was told to use. Meaning if you start an executor with a single "core" on a host with 64 CPUs, you'll get 64 threads, which is kinda overkill.

Using the allocated cores + a lower bound is probably a better approach.

Issue Links

• links to: [Github] Pull Request #18639 (vanzin)