[ZOOKEEPER-2810] Racy implicit ZKDatabase creation

• Type: Bug
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 3.4.8
• Component/s: server
• Assignee: Unassigned
• Reporter: Paul Millar
• Created: 16/Jun/17 13:24
• Updated: 16/Jun/17 13:24

Description

The NIOServerCnxnFactory#startup method first starts the acceptor thread and then initialises the ZooKeeperServer instance. In particular, the call to ZooKeeperServer#startdata method creates the ZKDatabase if it does not already exist.

This creates a race-condition: if the acceptor thread accepts an incoming connection before the ZKDatabase is established then there is a NullPointerException:

java.lang.NullPointerException: null
	at org.apache.zookeeper.server.ZooKeeperServer.processConnectRequest(ZooKeeperServer.java:864) ~[zookeeper-3.4.8.jar:3.4.8--1]
	at org.apache.zookeeper.server.NIOServerCnxn.readConnectRequest(NIOServerCnxn.java:418) ~[zookeeper-3.4.8.jar:3.4.8--1]
	at org.apache.zookeeper.server.NIOServerCnxn.readPayload(NIOServerCnxn.java:198) ~[zookeeper-3.4.8.jar:3.4.8--1]
	at org.apache.zookeeper.server.NIOServerCnxn.doIO(NIOServerCnxn.java:244) ~[zookeeper-3.4.8.jar:3.4.8--1]
	at org.apache.zookeeper.server.NIOServerCnxnFactory.run(NIOServerCnxnFactory.java:203) ~[zookeeper-3.4.8.jar:3.4.8--1]
	at java.lang.Thread.run(Thread.java:748) [na:1.8.0_131]

The same problem appears to be present in release-3.5 and master branches.

The naive fix would be to start the acceptor thread last in NIOServerCnxnFactory#startup, but I can't say whether this would cause any other problems.