You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Aleksey Yeschenko (Jira)" <ji...@apache.org> on 2020/04/15 13:44:00 UTC
[jira] [Commented] (CASSANDRA-15726) buffer pool may throw NPE with
concurrent release due to in-progress tiny pool eviction
[ https://issues.apache.org/jira/browse/CASSANDRA-15726?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17084099#comment-17084099 ]
Aleksey Yeschenko commented on CASSANDRA-15726:
-----------------------------------------------
Sure, I'll take a look once I've gotten through some items in my queue.
> buffer pool may throw NPE with concurrent release due to in-progress tiny pool eviction
> ---------------------------------------------------------------------------------------
>
> Key: CASSANDRA-15726
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15726
> Project: Cassandra
> Issue Type: Bug
> Components: Legacy/Core
> Reporter: ZhaoYang
> Assignee: ZhaoYang
> Priority: Normal
> Fix For: 4.0-alpha
>
>
> This can be reproduced by running {{LongBufferPoolTest}}, 1 out 5 runs..
> {code:java}
> java.lang.NullPointerException
> at org.apache.cassandra.utils.memory.BufferPool$Chunk.access$1300(BufferPool.java:836)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.lambda$remove$1(BufferPool.java:716)
> at org.apache.cassandra.utils.memory.BufferPool$MicroQueueOfChunks.removeIf(BufferPool.java:460)
> at org.apache.cassandra.utils.memory.BufferPool$MicroQueueOfChunks.access$1500(BufferPool.java:304)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.remove(BufferPool.java:716)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.put(BufferPool.java:590)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.recycle(BufferPool.java:709)
> at org.apache.cassandra.utils.memory.BufferPool$Chunk.recycle(BufferPool.java:909)
> at org.apache.cassandra.utils.memory.BufferPool$Chunk.tryRecycle(BufferPool.java:903)
> at org.apache.cassandra.utils.memory.BufferPool$Chunk.release(BufferPool.java:896)
> at org.apache.cassandra.utils.memory.BufferPool$MicroQueueOfChunks.removeIf(BufferPool.java:465)
> at org.apache.cassandra.utils.memory.BufferPool$MicroQueueOfChunks.access$1500(BufferPool.java:304)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.addChunk(BufferPool.java:736)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.addChunkFromParent(BufferPool.java:725)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.tryGetInternal(BufferPool.java:691)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.tryGet(BufferPool.java:679)
> at org.apache.cassandra.utils.memory.BufferPool$LocalPool.access$000(BufferPool.java:518)
> at org.apache.cassandra.utils.memory.BufferPool.tryGet(BufferPool.java:120)
> at org.apache.cassandra.utils.memory.LongBufferPoolTest$2.testOne(LongBufferPoolTest.java:497)
> at org.apache.cassandra.utils.memory.LongBufferPoolTest$TestUntil.call(LongBufferPoolTest.java:558)
> at org.apache.cassandra.utils.memory.LongBufferPoolTest$TestUntil.call(LongBufferPoolTest.java:538)
> at java.util.concurrent.FutureTask.run(FutureTask.java:266)
> at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1149)
> at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:624)
> at io.netty.util.concurrent.FastThreadLocalRunnable.run(FastThreadLocalRunnable.java:30)
> at java.lang.Thread.run(Thread.java:748)
> {code}
> The cause is that:
> * When evicting a normal chunk from a full MicroQueueOfChunks, local pool will try to remove corresponding tiny chunks, via {{MicroQueueOfChunks#removeIf}}.
> * If matching tiny chunk is found, tiny {{chunk.release()}} is called immediately before moving null chunk to the back of the queue.
> * Due to concurrent release from different threads, tiny {{chunk.release()}} may cause its parent normal chunk, aka. the evicted chunk in #1, to be removed from local pool and causes tiny pool to remove corresponding tiny chunks again in {{LocalPool#remove()}}.
> * In {{MicroQueueOfChunks#removeIf}}, due to previous in-progress {{removeIf}}, it throws NPE as it violate MicroQueueOfChunks's assumption which requires null chunks to be put at the back of queue.
>
> |[patch|https://github.com/apache/cassandra/pull/537]|[CI|https://circleci.com/workflow-run/a97317a0-ef21-4c01-9a97-82eaf28d7faf]|
> The fix is to put null chunks to the back of queue before releasing any chunks.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org