Github user pwendell commented on a diff in the pull request:
https://github.com/apache/spark/pull/126#discussion_r11374568
--- Diff:
core/src/main/scala/org/apache/spark/broadcast/HttpBroadcast.scala ---
@@ -17,34 +17,65 @@
package org.apache.spark.broadcast
-import java.io.{File, FileOutputStream, ObjectInputStream, OutputStream}
-import java.net.{URL, URLConnection, URI}
+import java.io.{File, FileOutputStream, ObjectInputStream,
ObjectOutputStream, OutputStream}
+import java.net.{URI, URL, URLConnection}
import java.util.concurrent.TimeUnit
-import it.unimi.dsi.fastutil.io.FastBufferedInputStream
-import it.unimi.dsi.fastutil.io.FastBufferedOutputStream
+import it.unimi.dsi.fastutil.io.{FastBufferedInputStream,
FastBufferedOutputStream}
-import org.apache.spark.{SparkConf, HttpServer, Logging, SecurityManager,
SparkEnv}
+import org.apache.spark.{HttpServer, Logging, SecurityManager, SparkConf,
SparkEnv}
import org.apache.spark.io.CompressionCodec
import org.apache.spark.storage.{BroadcastBlockId, StorageLevel}
import org.apache.spark.util.{MetadataCleaner, MetadataCleanerType,
TimeStampedHashSet, Utils}
+/**
+ * A [[org.apache.spark.broadcast.Broadcast]] implementation that uses
HTTP server
+ * as a broadcast mechanism. The first time a HTTP broadcast variable
(sent as part of a
+ * task) is deserialized in the executor, the broadcasted data is fetched
from the driver
+ * (through a HTTP server running at the driver) and stored in the
BlockManager of the
+ * executor to speed up future accesses.
+ */
private[spark] class HttpBroadcast[T](@transient var value_ : T, isLocal:
Boolean, id: Long)
extends Broadcast[T](id) with Logging with Serializable {
- def value = value_
+ def getValue = value_
- def blockId = BroadcastBlockId(id)
+ val blockId = BroadcastBlockId(id)
+ /*
+ * Broadcasted data is also stored in the BlockManager of the driver.
The BlockManagerMaster
+ * does not need to be told about this block as not only need to know
about this data block.
+ */
HttpBroadcast.synchronized {
- SparkEnv.get.blockManager.putSingle(blockId, value_,
StorageLevel.MEMORY_AND_DISK, false)
+ SparkEnv.get.blockManager.putSingle(
+ blockId, value_, StorageLevel.MEMORY_AND_DISK, tellMaster = false)
}
if (!isLocal) {
HttpBroadcast.write(id, value_)
}
- // Called by JVM when deserializing an object
+ /**
+ * Remove all persisted state associated with this HTTP broadcast on the
executors.
+ */
+ def doUnpersist(blocking: Boolean) {
--- End diff --
should these be `override`? Not sure if this applies when the function is
not concrete in the trait (but I think it does).
---
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.
---
