ZhaoYang created CASSANDRA-15726:
------------------------------------
Summary: buffer pool may NPE with concurrent release
Key: CASSANDRA-15726
URL: https://issues.apache.org/jira/browse/CASSANDRA-15726
Project: Cassandra
Issue Type: Bug
Reporter: ZhaoYang
Assignee: ZhaoYang
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 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: [email protected]
For additional commands, e-mail: [email protected]
