[FLINK-2076] Bug in re-openable hash join

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.9
• Fix Version/s: 0.9
• Component/s: Runtime / Task
• Assignee: Chiwan Park
• Reporter: Stephan Ewen
• Created: 21/May/15 19:13
• Updated: 04/Jun/15 11:29
• Resolved: 04/Jun/15 11:29

Description

It happens deterministically in my machine with the following setup:

TaskManager:

• heap size: 512m
• network buffers: 4096
• slots: 32

Job:

• ConnectedComponents
• 100k vertices
• 1.2m edges

--> this gives around 260 m Flink managed memory, across 32 slots is 8MB per slot, with several mem consumers in the job, makes the iterative hash join out-of-core

java.lang.RuntimeException: Hash Join bug in memory management: 
Memory buffers leaked.
	at org.apache.flink.runtime.operators.hash.MutableHashTable.buildTableFromSpilledPartition(MutableHashTable.java:733)
	at org.apache.flink.runtime.operators.hash.MutableHashTable.prepareNextPartition(MutableHashTable.java:508)
	at org.apache.flink.runtime.operators.hash.ReOpenableMutableHashTable.prepareNextPartition(ReOpenableMutableHashTable.java:167)
	at org.apache.flink.runtime.operators.hash.MutableHashTable.nextRecord(MutableHashTable.java:541)
	at org.apache.flink.runtime.operators.hash.NonReusingBuildSecondHashMatchIterator.callWithNextKey(NonReusingBuildSecondHashMatchIterator.java:102)
	at org.apache.flink.runtime.operators.AbstractCachedBuildSideMatchDriver.run(AbstractCachedBuildSideMatchDriver.java:155)
	at org.apache.flink.runtime.operators.RegularPactTask.run(RegularPactTask.java:496)
	at org.apache.flink.runtime.iterative.task.AbstractIterativePactTask.run(AbstractIterativePactTask.java:139)
	at org.apache.flink.runtime.iterative.task.IterationIntermediatePactTask.run(IterationIntermediatePactTask.java:92)
	at org.apache.flink.runtime.operators.RegularPactTask.invoke(RegularPactTask.java:362)
	at org.apache.flink.runtime.taskmanager.Task.run(Task.java:560)
	at java.lang.Thread.run(Thread.java:745)