Some benchmarks fail with jdk26 after JDK-8361426: (ref) Remove jdk.internal.ref.Cleaner

[MBaesken]

9 benchmarks fail after the removal of Cleaner, for example als, chi-square.

als :

java.lang.ExceptionInInitializerError
	at org.apache.spark.unsafe.array.ByteArrayMethods.<clinit>(ByteArrayMethods.java:47)
	at org.apache.spark.memory.MemoryManager.defaultPageSizeBytes$lzycompute(MemoryManager.scala:261)
	at org.apache.spark.memory.MemoryManager.defaultPageSizeBytes(MemoryManager.scala:251)
	at org.apache.spark.memory.MemoryManager.$anonfun$pageSizeBytes$1(MemoryManager.scala:270)
	at scala.runtime.java8.JFunction0$mcJ$sp.apply(JFunction0$mcJ$sp.scala:17)
	at scala.Option.getOrElse(Option.scala:201)
	at org.apache.spark.memory.MemoryManager.<init>(MemoryManager.scala:270)
	at org.apache.spark.memory.UnifiedMemoryManager.<init>(UnifiedMemoryManager.scala:58)
	at org.apache.spark.memory.UnifiedMemoryManager$.apply(UnifiedMemoryManager.scala:207)
	at org.apache.spark.SparkEnv$.create(SparkEnv.scala:320)
	at org.apache.spark.SparkEnv$.createDriverEnv(SparkEnv.scala:194)
	at org.apache.spark.SparkContext.createSparkEnv(SparkContext.scala:284)
	at org.apache.spark.SparkContext.<init>(SparkContext.scala:478)
	at org.apache.spark.SparkContext$.getOrCreate(SparkContext.scala:2883)
	at org.apache.spark.sql.SparkSession$Builder.$anonfun$getOrCreate$2(SparkSession.scala:1099)
	at scala.Option.getOrElse(Option.scala:201)
	at org.apache.spark.sql.SparkSession$Builder.getOrCreate(SparkSession.scala:1093)
	at org.renaissance.apache.spark.SparkUtil.setUpSparkContext(SparkUtil.scala:81)
	at org.renaissance.apache.spark.SparkUtil.setUpSparkContext$(SparkUtil.scala:54)
	at org.renaissance.apache.spark.Als.setUpSparkContext(Als.scala:56)
	at org.renaissance.apache.spark.Als.setUpBeforeAll(Als.scala:127)
	at org.renaissance.harness.ExecutionDriver.executeBenchmark(ExecutionDriver.java:82)
	at org.renaissance.harness.RenaissanceSuite$.runBenchmarks$$anonfun$1(RenaissanceSuite.scala:172)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:15)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:10)
	at scala.collection.immutable.List.foreach(List.scala:334)
	at org.renaissance.harness.RenaissanceSuite$.runBenchmarks(RenaissanceSuite.scala:161)
	at org.renaissance.harness.RenaissanceSuite$.main(RenaissanceSuite.scala:130)
	at org.renaissance.harness.RenaissanceSuite.main(RenaissanceSuite.scala)
	at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:104)
	at java.base/java.lang.reflect.Method.invoke(Method.java:565)
	at org.renaissance.core.Launcher.loadAndInvokeHarnessClass(Launcher.java:129)
	at org.renaissance.core.Launcher.launchHarnessClass(Launcher.java:78)
	at org.renaissance.core.Launcher.main(Launcher.java:43)
Caused by: java.lang.IllegalStateException: java.lang.ClassNotFoundException: jdk.internal.ref.Cleaner
	at org.apache.spark.unsafe.Platform.<clinit>(Platform.java:115)
	... 34 more
Caused by: java.lang.ClassNotFoundException: jdk.internal.ref.Cleaner
	at java.base/java.net.URLClassLoader.findClass(URLClassLoader.java:377)
	at java.base/java.lang.ClassLoader.loadClass(ClassLoader.java:557)
	at java.base/java.lang.ClassLoader.loadClass(ClassLoader.java:490)
	at java.base/java.lang.Class.forName0(Native Method)
	at java.base/java.lang.Class.forName(Class.java:467)
	at java.base/java.lang.Class.forName(Class.java:458)
	at org.apache.spark.unsafe.Platform.<clinit>(Platform.java:95)
	... 34 more
]

Benchmark 'chi-square' failed with exception:
java.lang.ExceptionInInitializerError
	at org.apache.spark.unsafe.array.ByteArrayMethods.<clinit>(ByteArrayMethods.java:47)
	at org.apache.spark.memory.MemoryManager.defaultPageSizeBytes$lzycompute(MemoryManager.scala:261)
	at org.apache.spark.memory.MemoryManager.defaultPageSizeBytes(MemoryManager.scala:251)
	at org.apache.spark.memory.MemoryManager.$anonfun$pageSizeBytes$1(MemoryManager.scala:270)
	at scala.runtime.java8.JFunction0$mcJ$sp.apply(JFunction0$mcJ$sp.scala:17)
	at scala.Option.getOrElse(Option.scala:201)
	at org.apache.spark.memory.MemoryManager.<init>(MemoryManager.scala:270)
	at org.apache.spark.memory.UnifiedMemoryManager.<init>(UnifiedMemoryManager.scala:58)
	at org.apache.spark.memory.UnifiedMemoryManager$.apply(UnifiedMemoryManager.scala:207)
	at org.apache.spark.SparkEnv$.create(SparkEnv.scala:320)
	at org.apache.spark.SparkEnv$.createDriverEnv(SparkEnv.scala:194)
	at org.apache.spark.SparkContext.createSparkEnv(SparkContext.scala:284)
	at org.apache.spark.SparkContext.<init>(SparkContext.scala:478)
	at org.apache.spark.SparkContext$.getOrCreate(SparkContext.scala:2883)
	at org.apache.spark.sql.SparkSession$Builder.$anonfun$getOrCreate$2(SparkSession.scala:1099)
	at scala.Option.getOrElse(Option.scala:201)
	at org.apache.spark.sql.SparkSession$Builder.getOrCreate(SparkSession.scala:1093)
	at org.renaissance.apache.spark.SparkUtil.setUpSparkContext(SparkUtil.scala:81)
	at org.renaissance.apache.spark.SparkUtil.setUpSparkContext$(SparkUtil.scala:54)
	at org.renaissance.apache.spark.ChiSquare.setUpSparkContext(ChiSquare.scala:41)
	at org.renaissance.apache.spark.ChiSquare.setUpBeforeAll(ChiSquare.scala:86)
	at org.renaissance.harness.ExecutionDriver.executeBenchmark(ExecutionDriver.java:82)
	at org.renaissance.harness.RenaissanceSuite$.runBenchmarks$$anonfun$1(RenaissanceSuite.scala:172)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:15)
	at scala.runtime.function.JProcedure1.apply(JProcedure1.java:10)
	at scala.collection.immutable.List.foreach(List.scala:334)
	at org.renaissance.harness.RenaissanceSuite$.runBenchmarks(RenaissanceSuite.scala:161)
	at org.renaissance.harness.RenaissanceSuite$.main(RenaissanceSuite.scala:130)
	at org.renaissance.harness.RenaissanceSuite.main(RenaissanceSuite.scala)
	at java.base/jdk.internal.reflect.DirectMethodHandleAccessor.invoke(DirectMethodHandleAccessor.java:104)
	at java.base/java.lang.reflect.Method.invoke(Method.java:565)
	at org.renaissance.core.Launcher.loadAndInvokeHarnessClass(Launcher.java:129)
	at org.renaissance.core.Launcher.launchHarnessClass(Launcher.java:78)
	at org.renaissance.core.Launcher.main(Launcher.java:43)
Caused by: java.lang.IllegalStateException: java.lang.ClassNotFoundException: jdk.internal.ref.Cleaner
	at org.apache.spark.unsafe.Platform.<clinit>(Platform.java:115)
	... 34 more
Caused by: java.lang.ClassNotFoundException: jdk.internal.ref.Cleaner
	at java.base/java.net.URLClassLoader.findClass(URLClassLoader.java:377)
	at java.base/java.lang.ClassLoader.loadClass(ClassLoader.java:557)
	at java.base/java.lang.ClassLoader.loadClass(ClassLoader.java:490)
	at java.base/java.lang.Class.forName0(Native Method)
	at java.base/java.lang.Class.forName(Class.java:467)
	at java.base/java.lang.Class.forName(Class.java:458)
	at org.apache.spark.unsafe.Platform.<clinit>(Platform.java:95)
	... 34 more

Maybe you can update those benchmarks, or flag them as not compatible with jdk26 ?

[MBaesken]

See https://bugs.openjdk.org/browse/JDK-8361426 .

[farquet]

Thanks for opening the issue! We've noticed the problem too.
For Spark benchmarks, it's likely that a version bump would fix the issue since it has already been addressed there.

However, db-shootout depends on chronicle-map which still has the problematic code in its code base. An issue has been opened there: OpenHFT/Chronicle-Map#575

[farquet]

Note that bumping the Spark version to 4.0+ would have the drawback of making those benchmarks require JDK17+. I think this is one of the reason this has not been made earlier.
The fact that we would be stuck at max JDK25 is probably a good reason to pay the price of dropping earlier versions support now. Previous renaissance release would still offer compatibility with older JDKs if need be.

[lbulej]

Do you think it would be worthwhile to make a quick release with the affected benchmarks limited to JDK 25 (so that the suite refuses to run the benchmarks on JDK 26 instead of crashing)?

Regardless of that, I'll check where Spark version updates (and potentially other changes in db-shootout) take us. Also, we could make it an explicit policy/goal to always support the latest JDK regardless of the cost of dropping older LTS JDKs — it could untie our hands a little bit (provided we are able to keep a reasonable set of benchmarks :-).

[MBaesken]

Note that bumping the Spark version to 4.0+ would have the drawback of making those benchmarks require JDK17+

So Spark 4.0 has it fixed? Do you know the exact version (or even change?) of Spark fixing this?

[lbulej]

I have tested a local build of Renaissance using Spark 4.0 with jdk-26+6 (which exposes the problem — db-shootout crashes), but upgrading to Spark 4.0 does not fix the problem — I still get the same exceptions as in the original post.

[MBaesken]

I have tested a local build of Renaissance using Spark 4.0 with jdk-26+6 (which exposes the problem — db-shootout crashes), but upgrading to Spark 4.0 does not fix the problem — I still get the same exceptions as in the original post.

Hi Lubomír , so do you think Spark needs more fixes / adjustments ? Should we open an issue at Spark (or are you aware that there is already an existing one) ?

[lbulej]

Hi Lubomír , so do you think Spark needs more fixes / adjustments ? Should we open an issue at Spark (or are you aware that there is already an existing one) ?

The code of the org.apache.spark.unsafe.Platform class in the Spark (both in branch-4.0 and master branches) still looks up jdk.internal.ref.Cleaner using Class.forName().

I haven't found any recent open issue related to that in Spark's JIRA. There was some related discussion around accessing sun.misc.Cleaner in Spark 3.0 era (SPARK-24421), including Alan Bateman's warning about not relying on jdk.internal.ref that were apparently not heeded, but that's about it.

I noted elsewhere in the JIRA that moving to java.lang.ref.Cleaner is not entirely trivial, because it cannot be triggered explicitly by calling the clean() method (which is public in the jdk.internal.ref.Cleaner but not even private in java.lang.ref.Cleaner), which is why I don't think a simple patch on our side (like we did with the Hadoop Client API in #453 to fix #439) would cut it.

Anyway, so far this does not look like a problem being fixed at the moment.

[lbulej]

It seems we can run again on Java 26+ (see #481), so I would like to release 0.16.1 soonish to improve compatibility before starting another round of library updates and possibly other changes.