[CURATOR-535] TestServer random port selection has a race condition

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 4.2.0
• Fix Version/s: 5.5.0

Environment

Operating System:
  Fedora 30 (amd64)
JVM:
  openjdk version "1.8.0_212"
  OpenJDK Runtime Environment (build 1.8.0_212-b04)
  OpenJDK 64-Bit Server VM (build 25.212-b04, mixed mode)

• Assignee: Enrico Olivelli
• Reporter: Laverne Schrock
• Created: 02/Aug/19 14:21
• Updated: 06/Apr/23 07:20
• Resolved: 21/Feb/23 13:29

Description

When using one of the constructors for org.apache.curator.test.TestingServer that doesn't take a port number, the org.apache.curator.test.InstanceSpec that is constructed will chose random available ports to use. However, InstanceSpec only binds those ports during construction and then unbinds them so that they can be used when TestingServer.start() is called.

This disconnect between port selection creates a race condition where some other process (or thread) could bind the port before TestingServer is started.

I've seen this very rarely in our integration test suite that spins up and tears down TestingServer many times. I've attached a simple class for reproducing the issue. If you run it in an environment with log4j loaded and the attached log4j.properties, you should see output like the following (though it sometimes takes more iterations):

completed iteration: 0
completed iteration: 500
2019-08-02 09:47:06 ERROR TestingZooKeeperServer:162 - From testing server (random state: false) for instance: InstanceSpec{dataDirectory=/tmp/1564753624792-1, port=34707, electionPort=33621, quorumPort=45995, deleteDataDirectoryOnClose=true, serverId=1286, tickTime=-1, maxClientCnxns=-1, customProperties={}, hostname=127.0.0.1} org.apache.curator.test.InstanceSpec@59c43d10
java.net.BindException: Address already in use
    at sun.nio.ch.Net.bind0(Native Method)
    at sun.nio.ch.Net.bind(Net.java:433)
    at sun.nio.ch.Net.bind(Net.java:425)
    at sun.nio.ch.ServerSocketChannelImpl.bind(ServerSocketChannelImpl.java:223)
    at sun.nio.ch.ServerSocketAdaptor.bind(ServerSocketAdaptor.java:74)
    at sun.nio.ch.ServerSocketAdaptor.bind(ServerSocketAdaptor.java:67)
    at org.apache.zookeeper.server.NIOServerCnxnFactory.configure(NIOServerCnxnFactory.java:687)
    at org.apache.zookeeper.server.ServerCnxnFactory.configure(ServerCnxnFactory.java:76)
    at org.apache.curator.test.TestingZooKeeperMain.internalRunFromConfig(TestingZooKeeperMain.java:239)
    at org.apache.curator.test.TestingZooKeeperMain.runFromConfig(TestingZooKeeperMain.java:132)
    at org.apache.curator.test.TestingZooKeeperServer$1.run(TestingZooKeeperServer.java:158)
    at java.lang.Thread.run(Thread.java:748)
java.lang.IllegalStateException: Timed out waiting for watch removal
    at org.apache.curator.test.TestingZooKeeperMain.blockUntilStarted(TestingZooKeeperMain.java:146)
    at org.apache.curator.test.TestingZooKeeperServer.start(TestingZooKeeperServer.java:167)
    at org.apache.curator.test.TestingServer.start(TestingServer.java:148)
    at BugReproducer.main(BugReproducer.java:15)

Issue Links

• causes: FLINK-20523 OffsetsInitializerTest test failed with "java.net.BindException: Address already in use"
• links to: GitHub Pull Request #406
• links to: GitHub Pull Request #421