Paul Millar created ZOOKEEPER-2838:
--------------------------------------
Summary: race-condition when shutting down NIOServerCnxnFactory
yields CancelledKeyException
Key: ZOOKEEPER-2838
URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2838
Project: ZooKeeper
Issue Type: Bug
Components: server
Affects Versions: 3.5.3, 3.5.2, 3.5.1, 3.5.0, 3.4.10, 3.4.9, 3.4.8, 3.4.7,
3.4.6
Reporter: Paul Millar
Priority: Minor
The problem stems from closing the ServerSocketChannel before stopping the
thread(s) working with the corresponding Selector. Closing the
ServerSocketChannel will invalidate any SelectionKey objects that have been
declared. This is equivalent to calling cancel on the SelectionKey.
Therefore, after the ServerSocketChannel's close method is called, it is
possible that any thread working with a SelectionKey will experience
CancelledKeyException.
I noticed the problem with ZooKeeper v3.4.6, which resulted in the following
stack-trace:
{quote}
04 Jul 2017 15:54:15 (zookeeper) [] Ignoring unexpected runtime exception
java.nio.channels.CancelledKeyException: null
at sun.nio.ch.SelectionKeyImpl.ensureValid(SelectionKeyImpl.java:73)
~[na:1.8.0_131]
at sun.nio.ch.SelectionKeyImpl.readyOps(SelectionKeyImpl.java:87)
~[na:1.8.0_131]
at
org.apache.zookeeper.server.NIOServerCnxnFactory.run(NIOServerCnxnFactory.java:187)
~[zookeeper-3.4.6.jar:3.4.6-1569965]
at java.lang.Thread.run(Thread.java:748) [na:1.8.0_131]
{quote}
>From manually inspecting the source code, I see the problem is present in all
>currently released versions of ZooKeeper.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
