[ZOOKEEPER-1816] ClientCnxn.close() should block until threads have died

• Type: Bug
• Status: Patch Available
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 3.3.6, 3.4.5, 3.4.11
• Component/s: java client
• Assignee: wu wen
• Reporter: Jared Winick
• Created: 15/Nov/13 21:33
• Updated: 22/Jun/18 04:49

Description

In the testing of ACCUMULO-1379 and ACCUMULO-1858 it was seen that the non-blocking behavior of ClientCnxn.close(), and therefore ZooKeeper.close(), can cause a race condition when undeploying an application running in a Java container such as JBoss or Tomcat. As the close() method returns without joining on the sendThread and eventThread, those threads continue to execute/cleanup while the container is cleaning up the application's resources. If the container has unloaded classes by the time this code runs

ZooTrace.logTraceMessage(LOG, ZooTrace.getTextTraceLevel(), "SendThread exitedloop.");

A "java.lang.NoClassDefFoundError: org/apache/zookeeper/server/ZooTrace" can be seen.

Issue Links

• is related to: ZOOKEEPER-1394 ClassNotFoundException on shutdown of client
• relates to: ACCUMULO-1858 Backport fix for Accumulo-1379 PermGen Leak to 1.4 and 1.5