Kalvin2077 commented on code in PR #3834:
URL: https://github.com/apache/celeborn/pull/3834#discussion_r4004264954


##########
client-spark/spark-3/src/test/scala/org/apache/spark/shuffle/celeborn/CelebornShuffleManagerSuite.scala:
##########
@@ -92,6 +92,20 @@ class SparkShuffleManagerSuite extends Logging {
     sc.stop()
   }
 
+  @junit.Test
+  def testReliableStorageFollowsFallbackPolicy(): Unit = {
+    // supportsReliableStorage must be true only for NEVER; AUTO/ALWAYS may 
fall back to local-disk
+    // shuffle, so reporting reliable storage would let DRA reclaim executors 
holding fallback data.
+    def reliable(policy: String): Boolean = {
+      val conf = new SparkConf()
+        .set(s"spark.${CelebornConf.SPARK_SHUFFLE_FALLBACK_POLICY.key}", 
policy)
+      new CelebornShuffleDataIO(conf).driver().supportsReliableStorage()

Review Comment:
   **[P1] New test calls the Spark 3.5 interface, causing older versions of 
Spark to fail compilation**
   
   The declared return type is Spark’s `ShuffleDriverComponents`, and this 
interface does not have `supportsReliableStorage()` in Spark 3.0–3.4. This 
shared test therefore fails during the test compilation phase for older 
profiles, and this cannot be avoided even when configured to use 
`LocalDiskShuffleDataIO`. When performing a minimal Scala compilation using the 
actual JARs for Spark 3.0.3 and 3.4.4, the error `value supportsReliableStorage 
is not a member of ... ShuffleDriverComponents`; the corresponding test for 
3.5.1 passed. It is recommended to cast the return value to 
`CelebornShuffleDriverComponents` (from the same package) and call its existing 
methods, or use a compatible reflection-based approach, and then re-verify the 
test compilation for Spark 3.0–3.4.



-- 
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]

Reply via email to