You are viewing a plain text version of this content. The canonical link for it is here.
Posted to reviews@spark.apache.org by GitBox <gi...@apache.org> on 2021/11/09 03:51:00 UTC

[GitHub] [spark] Ngone51 commented on a change in pull request #33896: [SPARK-33701][SHUFFLE] Adaptive shuffle merge finalization for push-based shuffle

Ngone51 commented on a change in pull request #33896:
URL: https://github.com/apache/spark/pull/33896#discussion_r745261645



##########
File path: core/src/main/scala/org/apache/spark/shuffle/ShuffleBlockPusher.scala
##########
@@ -327,10 +338,32 @@ private[spark] class ShuffleBlockPusher(conf: SparkConf) extends Logging {
         s"stop.")
       return false
     } else {
+      if (remainingBlocks.isEmpty && pushRequests.isEmpty && deferredPushRequests.isEmpty) {

Review comment:
       It looks like the `remainingBlocks` is only for the current request. So it's possible that `remainingBlocks` could be non-empty for other ongoing requests (which are not either in `pushRequests` or `deferredPushRequests` ). 

##########
File path: core/src/main/scala/org/apache/spark/internal/config/package.scala
##########
@@ -2193,6 +2193,30 @@ package object config {
       // with small MB sized chunk of data.
       .createWithDefaultString("3m")
 
+  private[spark] val PUSH_BASED_SHUFFLE_MERGE_FINALIZE_THREADS =
+    ConfigBuilder("spark.shuffle.push.merge.finalizeThreads")
+      .doc("Number of threads used by driver to finalize shuffle merge. Since it could" +
+        " potentially take seconds for a large shuffle to finalize, having multiple threads helps" +
+        " driver to handle concurrent shuffle merge finalize requests when push-based" +
+        " shuffle is enabled.")
+      .intConf
+      .createWithDefault(3)
+
+  private[spark] val PUSH_BASED_SHUFFLE_SIZE_MIN_SHUFFLE_SIZE_TO_WAIT =
+    ConfigBuilder("spark.shuffle.push.minShuffleSizeToWait")
+      .doc("Driver will wait for merge finalization to complete only if total shuffle size is" +
+        " more than this threshold. If total shuffle size is less, driver will immediately" +
+        " finalize the shuffle output")
+      .bytesConf(ByteUnit.BYTE)
+      .createWithDefaultString("500m")
+
+  private[spark] val PUSH_BASED_SHUFFLE_MIN_PUSH_RATIO =
+    ConfigBuilder("spark.shuffle.push.minPushRatio")

Review comment:
       How about `minCompletedPushRatio`?




-- 
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: reviews-unsubscribe@spark.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org