Github user tgravescs commented on a diff in the pull request:
https://github.com/apache/spark/pull/12113#discussion_r62327844
--- Diff: core/src/main/scala/org/apache/spark/MapOutputTracker.scala ---
@@ -296,10 +290,86 @@ private[spark] class MapOutputTrackerMaster(conf:
SparkConf)
protected val mapStatuses = new ConcurrentHashMap[Int,
Array[MapStatus]]().asScala
private val cachedSerializedStatuses = new ConcurrentHashMap[Int,
Array[Byte]]().asScala
+ private val maxRpcMessageSize = RpcUtils.maxMessageSizeBytes(conf)
+
+ // Kept in sync with cachedSerializedStatuses explicitly
+ // This is required so that the Broadcast variable remains in scope
until we remove
+ // the shuffleId explicitly or implicitly.
+ private val cachedSerializedBroadcast = new HashMap[Int,
Broadcast[Array[Byte]]]()
+
+ // This is to prevent multiple serializations of the same shuffle -
which happens when
+ // there is a request storm when shuffle start.
+ private val shuffleIdLocks = new ConcurrentHashMap[Int, AnyRef]()
+
+ // requests for map output statuses
+ private val mapOutputRequests = new
LinkedBlockingQueue[GetMapOutputMessage]
+
+ // Thread pool used for handling map output status requests. This is a
separate thread pool
+ // to ensure we don't block the normal dispatcher threads.
+ private val threadpool: ThreadPoolExecutor = {
+ val numThreads =
conf.getInt("spark.shuffle.mapOutput.dispatcher.numThreads", 8)
+ val pool = ThreadUtils.newDaemonFixedThreadPool(numThreads,
"map-output-dispatcher")
+ for (i <- 0 until numThreads) {
+ pool.execute(new MessageLoop)
+ }
+ pool
+ }
+
+ // Make sure that that we aren't going to exceed the max RPC message
size by making sure
+ // we use broadcast to send large map output statuses.
+ if (minSizeForBroadcast > maxRpcMessageSize) {
+ val msg = s"spark.shuffle.mapOutput.minSizeForBroadcast
($minSizeForBroadcast bytes) must " +
+ s"be <= spark.rpc.message.maxSize ($maxRpcMessageSize bytes) to
prevent sending an rpc " +
+ "message that is to large."
+ logError(msg)
+ throw new IllegalArgumentException(msg)
--- End diff --
Personally I preferring erroring out in this case. It fails very quickly
so you won't lose any work and if the user sets an undocumented config wrong I
think we should tell them so they can fix and don't get unexpected behavior.
Now its possible they set spark.rpc.message.maxSize really small which I think
would be a bad thing also so again erroring feels like the better thing here.
If there are other reasons we should just warn let me know
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
