maheshk114 commented on code in PR #45228:
URL: https://github.com/apache/spark/pull/45228#discussion_r1535382480
##########
core/src/main/scala/org/apache/spark/internal/config/package.scala:
##########
@@ -551,6 +551,22 @@ package object config {
.checkValue(_.endsWith(java.io.File.separator), "Path should end with
separator.")
.createOptional
+ private[spark] val STORAGE_FALLBACK_STORAGE_NUM_THREADS_FOR_SHUFFLE_READ =
+ ConfigBuilder("spark.storage.fallbackStorage.num.threads.for.shuffle.read")
Review Comment:
done
##########
core/src/main/scala/org/apache/spark/storage/FallbackStorage.scala:
##########
@@ -147,6 +222,14 @@ private[spark] object FallbackStorage extends Logging {
}
}
+ def stopThreadPool(conf: SparkConf): Unit = {
+ logInfo(s" Stopping thread pool")
+ if (getFallbackStorage(conf).isDefined &&
+ getFallbackStorage(conf).get.fetchThreadPool.isDefined) {
Review Comment:
done
##########
core/src/main/scala/org/apache/spark/storage/FallbackStorage.scala:
##########
@@ -92,6 +95,57 @@ private[storage] class FallbackStorage(conf: SparkConf)
extends Logging {
val hash = JavaUtils.nonNegativeHash(filename)
fallbackFileSystem.exists(new Path(fallbackPath,
s"$appId/$shuffleId/$hash/$filename"))
}
+
+ private val fetchThreadPool: Option[ThreadPoolExecutor] = {
+ val numShuffleThreads = FallbackStorage.getNumReadThreads(conf)
+ if (numShuffleThreads > 0) {
+ logInfo(s"FallbackStorage created thread pool using
${numShuffleThreads} thread(s)")
+ Some(ThreadUtils.newDaemonCachedThreadPool(
+ "FetchFromFallbackStorage-threadPool", numShuffleThreads))
+ } else {
+ logInfo("FallbackStorage thread pool not created")
+ None
+ }
+ }
+
+ def fetchBlocks(
+ blockManager: BlockManager,
+ blocks: collection.Seq[FetchBlockInfo],
+ address: BlockManagerId,
+ listener: BlockFetchingListener): Unit = {
+ fetchThreadPool match {
+ case Some(p) if !p.isShutdown =>
+ blocks.foreach(block =>
+ p.submit(new Runnable {
+ override def run(): Unit = {
+ fetchShuffleBlocks(block, blockManager, listener)
+ }
+ })
+ )
+ case _ =>
+ logInfo(s" fetchThreadPool does not exists for $address or shutdown")
+ blocks.foreach(block => fetchShuffleBlocks(block, blockManager,
listener))
+ }
+ }
+
+ private def fetchShuffleBlocks(
+ block: FetchBlockInfo,
Review Comment:
done
##########
core/src/test/scala/org/apache/spark/storage/FallbackStorageSuite.scala:
##########
@@ -172,61 +172,64 @@ class FallbackStorageSuite extends SparkFunSuite with
LocalSparkContext {
}
test("migrate shuffle data to fallback storage") {
- val conf = new SparkConf(false)
Review Comment:
The test is not removed. I have just added a new conf to avoid duplicating
the same test case.
##########
core/src/main/scala/org/apache/spark/executor/CoarseGrainedExecutorBackend.scala:
##########
@@ -354,7 +355,8 @@ private[spark] class CoarseGrainedExecutorBackend(
// We can only trust allBlocksMigrated boolean value if there
were no tasks running
// since the start of computing it.
if (allBlocksMigrated && (migrationTime >
lastTaskFinishTime.get())) {
- logInfo("No running tasks, all blocks migrated, stopping.")
+ val timeTakenMs = (System.nanoTime() - startTime) / (1000 *
1000)
+ logInfo(s"No running tasks, all blocks migrated in
$timeTakenMs ms, stopping.")
Review Comment:
This log line can be used to compare the time taken to migrate to external
storage and remote node. This may be helpful to user to check the increase in
migration time.
##########
core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala:
##########
@@ -1558,6 +1587,7 @@ object ShuffleBlockFetcherIterator {
mapIndex: Int,
address: BlockManagerId,
size: Long,
+ timeTaken: Long,
Review Comment:
This is not just for logging but to capture the time taken by read
operations. This was helpful in comparing the time diff between read from
external storage and remote node.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
