andygrove commented on code in PR #6607:
URL: https://github.com/apache/datafusion-comet/pull/6607#discussion_r4197877768


##########
spark/src/test/scala/org/apache/spark/sql/comet/execution/shuffle/CometCelebornShufflePlanningSuite.scala:
##########
@@ -265,6 +265,21 @@ class CometCelebornShufflePlanningSuite extends 
CometTestBase {
     }
   }
 
+  test("the shuffle input conversion leaves a Spark child to Spark's shuffle") 
{
+    // The conversion only takes over from the JVM columnar shuffle, which 
Celeborn does not
+    // have, and Celeborn's native shuffle needs a native child.
+    withSQLConf(
+      CometConf.COMET_SHUFFLE_MODE.key -> "native",

Review Comment:
   Confirmed: with the Celeborn block in `shuffleSupported` removed, the test 
still passed in `native`. Running it in `auto` alone wouldn't catch that 
either, though. Under Celeborn, `shuffleSupported` first calls 
`cometCelebornShuffleFallbackReason` through `isCometShuffleEnabledReason`, and 
that already turns down `auto` and `jvm` with "Celeborn-backed Comet shuffle 
requires spark.comet.shuffle.mode=native". The `auto` version passed with the 
block removed, and failed only once both checks were gone.
   
   c90dcfca5 runs the test in `auto`, `jvm` and `native`, and checks that the 
fallback reason names Celeborn, which in `native` only the block gives; without 
it the reason is "Comet columnar shuffle not enabled". The test now fails if 
the block is removed, if both Celeborn checks are removed, since `auto` then 
converts, or if the gate stops requiring `CometColumnarShuffle`. Removing only 
the mode checks in `cometCelebornShuffleFallbackReason` fails eight other tests 
in the suite.
   



-- 
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]

Reply via email to