CompilationException on count(*) in co-related join since Trino 378

[takezoe]

SQL:

-- Table c1000 has 1000 columns
-- Table c10 has 10 columns
SELECT * FROM c1000 t1 WHERE (SELECT COUNT(*) FROM c10 t2 WHERE t1.col0 = t2.col0) > 2;

Stacktrace:

com.google.common.util.concurrent.UncheckedExecutionException: io.airlift.bytecode.CompilationException: Error compiling class: io/trino/$gen/PagesHashStrategy_20241103_105832_49
	at com.google.common.cache.LocalCache$Segment.get(LocalCache.java:2085)
	at com.google.common.cache.LocalCache.get(LocalCache.java:4017)
	at com.google.common.cache.LocalCache.getOrLoad(LocalCache.java:4040)
	at com.google.common.cache.LocalCache$LocalLoadingCache.get(LocalCache.java:4989)
	at com.google.common.cache.LocalCache$LocalLoadingCache.getUnchecked(LocalCache.java:4996)
	at com.google.common.cache.ForwardingLoadingCache.getUnchecked(ForwardingLoadingCache.java:54)
	at io.trino.sql.gen.JoinCompiler.compilePagesHashStrategyFactory(JoinCompiler.java:170)
	at io.trino.operator.StreamingAggregationOperator$StreamingAggregation.<init>(StreamingAggregationOperator.java:198)
	at io.trino.operator.StreamingAggregationOperator.<init>(StreamingAggregationOperator.java:147)
	at io.trino.operator.StreamingAggregationOperator$Factory.create(StreamingAggregationOperator.java:120)
	at io.trino.operator.WorkProcessorOperatorAdapter.<init>(WorkProcessorOperatorAdapter.java:108)
	at io.trino.operator.WorkProcessorOperatorAdapter$Factory.createOperator(WorkProcessorOperatorAdapter.java:64)
	at io.trino.operator.DriverFactory.createDriver(DriverFactory.java:107)
	at io.trino.execution.SqlTaskExecution$DriverSplitRunnerFactory.createDriver(SqlTaskExecution.java:667)
	at io.trino.execution.SqlTaskExecution$DriverSplitRunner.processFor(SqlTaskExecution.java:879)
	at io.trino.execution.executor.dedicated.SplitProcessor.run(SplitProcessor.java:77)
	at io.trino.execution.executor.dedicated.TaskEntry$VersionEmbedderBridge.lambda$run$0(TaskEntry.java:201)
	at io.trino.$gen.Trino_dev____20241103_105232_2.run(Unknown Source)
	at io.trino.execution.executor.dedicated.TaskEntry$VersionEmbedderBridge.run(TaskEntry.java:202)
	at io.trino.execution.executor.scheduler.FairScheduler.runTask(FairScheduler.java:172)
	at io.trino.execution.executor.scheduler.FairScheduler.lambda$submit$0(FairScheduler.java:159)
	at java.base/java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:572)
	at com.google.common.util.concurrent.TrustedListenableFutureTask$TrustedFutureInterruptibleTask.runInterruptibly(TrustedListenableFutureTask.java:131)
	at com.google.common.util.concurrent.InterruptibleTask.run(InterruptibleTask.java:76)
	at com.google.common.util.concurrent.TrustedListenableFutureTask.run(TrustedListenableFutureTask.java:82)
	at java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1144)
	at java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642)
	at java.base/java.lang.Thread.run(Thread.java:1575)
Caused by: io.airlift.bytecode.CompilationException: Error compiling class: io/trino/$gen/PagesHashStrategy_20241103_105832_49
	at io.airlift.bytecode.ByteCodeGenerator.generateByteCode(ByteCodeGenerator.java:106)
	at io.airlift.bytecode.ClassGenerator.defineClasses(ClassGenerator.java:100)
	at io.airlift.bytecode.ClassGenerator.defineClass(ClassGenerator.java:90)
	at io.trino.util.CompilerUtils.defineClass(CompilerUtils.java:81)
	at io.trino.util.CompilerUtils.defineClass(CompilerUtils.java:75)
	at io.trino.sql.gen.JoinCompiler.internalCompileHashStrategy(JoinCompiler.java:260)
	at io.trino.sql.gen.JoinCompiler.lambda$new$1(JoinCompiler.java:121)
	at com.google.common.cache.CacheLoader$FunctionToCacheLoader.load(CacheLoader.java:169)
	at com.google.common.cache.LocalCache$LoadingValueReference.loadFuture(LocalCache.java:3574)
	at com.google.common.cache.LocalCache$Segment.loadSync(LocalCache.java:2316)
	at com.google.common.cache.LocalCache$Segment.lockedGetOrLoad(LocalCache.java:2189)
	at com.google.common.cache.LocalCache$Segment.get(LocalCache.java:2079)
	... 27 more
Caused by: org.objectweb.asm.MethodTooLargeException: Method too large: io/trino/$gen/PagesHashStrategy_20241103_105832_49.rowEqualsRow (ILio/trino/spi/Page;ILio/trino/spi/Page;)Z
	at org.objectweb.asm.MethodWriter.computeMethodInfoSize(MethodWriter.java:2088)
	at org.objectweb.asm.ClassWriter.toByteArray(ClassWriter.java:506)
	at io.airlift.bytecode.ByteCodeGenerator.generateByteCode(ByteCodeGenerator.java:103)
	... 38 more

The root cause of the compilation error is too many grouping keys came from c1000 on decorelated count(*).

This query had worked until Trino 377 thanks to PushAggregationThroughOuterJoin which pushes aggregation below Left Join and reduced the number of grouping keys. However, PushFilterThroughCountAggregation was added in Trino 378 (#12047). It blocks PushAggregationThroughOuterJoin when applied.

I came up with some ideas to avoid this error but not sure if they are any good.

• Shorten auto generated byte-code for rowEqualsRow (split to multiple methods?)
• Skip PushFilterThroughCountAggregation if grouping keys are too many which has no chance to work

[wendigo]

Have you tried newer Trino? Does it still occur?

[martint]

Shorten auto generated byte-code for hashPosition (split to multiple methods?)

We should do this. Take a look at #23721 for a similar situation.

[takezoe]

Have you tried newer Trino? Does it still occur?

Yes, this happens even on the latest master.

We should do this. Take a look at #23721 for a similar situation.

Thanks. Let me try.

[takezoe]

The root cause of the compilation error is too many grouping keys came from c1000 on decorelated count(*).

Actually this happens in JoinCompiler due to too many joinChannels (as stacktrace says). The compilation error happens for many auto-generated methods which loops on joinChannelTypes.

If I remember correctly, it happened for hashPosition when we faced this error but Trino version was not the latest at that time. I will look into a bit more closely to see whether this difference is caused by the difference in Trino version.

[takezoe]

Where the compilation error happening depends on the number of columns. All patterns work without error if PushFilterThroughCountAggregation is disabled.

With 1000 columns:

Caused by: org.objectweb.asm.MethodTooLargeException: Method too large: io/trino/$gen/PagesHashStrategy_20241107_230454_49.rowEqualsRow (ILio/trino/spi/Page;ILio/trino/spi/Page;)Z

With 2000 columns:

Caused by: org.objectweb.asm.MethodTooLargeException: Method too large: io/trino/$gen/PagesHashStrategy_20241107_230550_59.appendTo (IILio/trino/spi/PageBuilder;I)V

With 3000 columns:

Caused by: org.objectweb.asm.MethodTooLargeException: Method too large: io/trino/$gen/PagesHashStrategy_20241107_230642_69.<init> (Ljava/util/List;Ljava/util/OptionalInt;)V

[takezoe]

Tried to split methods auto-generated in JoinCompiler:
master...takezoe:split-auto-generated-methods-in-join-compiler

This worked with 1000 columns but failed with 2000 columns with the following error. Splitting methods also has an upper limit.

Caused by: org.objectweb.asm.ClassTooLargeException: Class too large: io/trino/$gen/PagesHashStrategy_20241108_040102_49
	at org.objectweb.asm.ClassWriter.toByteArray(ClassWriter.java:616)
	at io.airlift.bytecode.ByteCodeGenerator.generateByteCode(ByteCodeGenerator.java:103)
	... 38 more