[jira] [Created] (FLINK-12544) Deadlock during releasing memory in SpillableSubpartition

["zhijiang (JIRA)" <ji...@apache.org>]

zhijiang created FLINK-12544:
--------------------------------

             Summary: Deadlock during releasing memory in SpillableSubpartition
                 Key: FLINK-12544
                 URL: https://issues.apache.org/jira/browse/FLINK-12544
             Project: Flink
          Issue Type: Bug
          Components: Runtime / Network
            Reporter: zhijiang
            Assignee: zhijiang


It is reported by flink user, and the original jstack is as following:

 
{code:java}
// "CoGroup (2/2)":
                at org.apache.flink.runtime.io.network.NetworkEnvironment.registerTask(NetworkEnvironment.java:213)
                - waiting to lock <0x000000062bf859b8> (a java.lang.Object)
                at org.apache.flink.runtime.taskmanager.Task.run(Task.java:614)
                at java.lang.Thread.run(Thread.java:745)
"CoGroup (1/2)":
                at org.apache.flink.runtime.io.network.buffer.LocalBufferPool.recycle(LocalBufferPool.java:277)
                - waiting to lock <0x000000063fdf4888> (a java.util.ArrayDeque)
                at org.apache.flink.runtime.io.network.buffer.NetworkBuffer.deallocate(NetworkBuffer.java:172)
                at org.apache.flink.shaded.netty4.io.netty.buffer.AbstractReferenceCountedByteBuf.release0(AbstractReferenceCountedByteBuf.java:95)
                at org.apache.flink.shaded.netty4.io.netty.buffer.AbstractReferenceCountedByteBuf.release(AbstractReferenceCountedByteBuf.java:84)
                at org.apache.flink.runtime.io.network.buffer.NetworkBuffer.recycleBuffer(NetworkBuffer.java:147)
                at org.apache.flink.runtime.io.network.buffer.BufferConsumer.close(BufferConsumer.java:121)
                at org.apache.flink.runtime.io.network.partition.SpillableSubpartition.spillFinishedBufferConsumers(SpillableSubpartition.java:274)
                at org.apache.flink.runtime.io.network.partition.SpillableSubpartition.releaseMemory(SpillableSubpartition.java:239)
                - locked <0x000000063fdf4ac8> (a java.util.ArrayDeque)
                at org.apache.flink.runtime.io.network.partition.ResultPartition.releaseMemory(ResultPartition.java:371)
                at org.apache.flink.runtime.io.network.buffer.LocalBufferPool.setNumBuffers(LocalBufferPool.java:375)
                at org.apache.flink.runtime.io.network.buffer.NetworkBufferPool.redistributeBuffers(NetworkBufferPool.java:408)
                at org.apache.flink.runtime.io.network.buffer.NetworkBufferPool.createBufferPool(NetworkBufferPool.java:297)
                - locked <0x000000063c785350> (a java.lang.Object)
                at org.apache.flink.runtime.io.network.buffer.NetworkBufferPool.createBufferPool(NetworkBufferPool.java:259)
                at org.apache.flink.runtime.io.network.NetworkEnvironment.setupInputGate(NetworkEnvironment.java:272)
                at org.apache.flink.runtime.io.network.NetworkEnvironment.registerTask(NetworkEnvironment.java:224)
                - locked <0x000000062bf859b8> (a java.lang.Object)
                at org.apache.flink.runtime.taskmanager.Task.run(Task.java:614)
                at java.lang.Thread.run(Thread.java:745)
"DataSource  (1/1)":
                at org.apache.flink.runtime.io.network.partition.SpillableSubpartition.releaseMemory(SpillableSubpartition.java:227)
                - waiting to lock <0x000000063fdf4ac8> (a java.util.ArrayDeque)
                at org.apache.flink.runtime.io.network.partition.ResultPartition.releaseMemory(ResultPartition.java:371)
                at org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestMemorySegment(LocalBufferPool.java:257)
                - locked <0x000000063fdf4888> (a java.util.ArrayDeque)
                at org.apache.flink.runtime.io.network.buffer.LocalBufferPool.requestBufferBuilderBlocking(LocalBufferPool.java:218)
                at org.apache.flink.runtime.io.network.api.writer.RecordWriter.requestNewBufferBuilder(RecordWriter.java:213)
                at org.apache.flink.runtime.io.network.api.writer.RecordWriter.sendToTarget(RecordWriter.java:144)
                at org.apache.flink.runtime.io.network.api.writer.RecordWriter.emit(RecordWriter.java:107)
                at org.apache.flink.runtime.operators.shipping.OutputCollector.collect(OutputCollector.java:65)
                at org.apache.flink.runtime.operators.util.metrics.CountingCollector.collect(CountingCollector.java:35)
                at org.apache.flink.runtime.operators.DataSourceTask.invoke(DataSourceTask.java:193)
                at org.apache.flink.runtime.taskmanager.Task.run(Task.java:711)
                at java.lang.Thread.run(Thread.java:745)
{code}
Based on the analysis, it happens when one task A is trying to release sub partition memory, then it occupies the lock in `LocalBufferPool`, and trying to get the lock in `SpillSubpartition`. Meanwhile, another task B is submitted to TM to trigger previous task to release memory, then it would already occupy the lock of SpillableSubpartition in A, and trying to get the lock in `LocalBufferPool` during recycling the buffer. So it causes the deadlock here.

 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)