[STORM-1114] Racing condition in trident zookeeper zk-node create/delete

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.0.0, 0.10.3, 1.0.4, 1.1.1, 1.2.0
• Component/s: storm-core
• Assignee: P. Taylor Goetz
• Reporter: Zhuo Liu
• Created: 15/Oct/15 15:41
• Updated: 03/Jul/17 15:06
• Resolved: 13/Apr/17 20:16

Description

In production for some trident topology, we met the bug that some workers are trying to create a zk-node that is already existent or delete a zk node that has already been deleted. This causes the worker process to die.

We dissect the problem and figure out that there exists racing condition in trident TransactionalState's zk-node create and delete codes.

failure stack trace in worker.log:

Caused by: org.apache.storm.shade.org.apache.zookeeper.KeeperException$NodeExistsException: KeeperErrorCode = NodeExists for /ignoreStoredMetadata
        at org.apache.storm.shade.org.apache.zookeeper.KeeperException.create(KeeperException.java:119) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.zookeeper.KeeperException.create(KeeperException.java:51) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.zookeeper.ZooKeeper.create(ZooKeeper.java:783) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.CreateBuilderImpl$11.call(CreateBuilderImpl.java:676) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.CreateBuilderImpl$11.call(CreateBuilderImpl.java:660) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.RetryLoop.callWithRetry(RetryLoop.java:107) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.CreateBuilderImpl.pathInForeground(CreateBuilderImpl.java:656) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.CreateBuilderImpl.protectedPathInForeground(CreateBuilderImpl.java:441) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.CreateBuilderImpl.forPath(CreateBuilderImpl.java:431) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.CreateBuilderImpl$3.forPath(CreateBuilderImpl.java:239) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.CreateBuilderImpl$3.forPath(CreateBuilderImpl.java:193) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at storm.trident.topology.state.TransactionalState.forPath(TransactionalState.java:83) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at storm.trident.topology.state.TransactionalState.createNode(TransactionalState.java:100) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at storm.trident.topology.state.TransactionalState.setData(TransactionalState.java:115) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        ... 9 more
2015-10-14 18:10:43.786 b.s.util [ERROR] Halting process: ("Worker died")

Caused by: org.apache.storm.shade.org.apache.zookeeper.KeeperException$NoNodeException: KeeperErrorCode = NoNode for /rainbowHdfsPath
        at org.apache.storm.shade.org.apache.zookeeper.KeeperException.create(KeeperException.java:111) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.zookeeper.KeeperException.create(KeeperException.java:51) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.zookeeper.ZooKeeper.delete(ZooKeeper.java:873) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.DeleteBuilderImpl$5.call(DeleteBuilderImpl.java:239) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.DeleteBuilderImpl$5.call(DeleteBuilderImpl.java:234) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.RetryLoop.callWithRetry(RetryLoop.java:107) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.DeleteBuilderImpl.pathInForeground(DeleteBuilderImpl.java:230) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.DeleteBuilderImpl.forPath(DeleteBuilderImpl.java:215) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at org.apache.storm.shade.org.apache.curator.framework.imps.DeleteBuilderImpl.forPath(DeleteBuilderImpl.java:42) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        at storm.trident.topology.state.TransactionalState.delete(TransactionalState.java:126) ~[storm-core-0.10.1.y.jar:0.10.1.y]
        ... 12 more
2015-10-14 18:10:28.799 b.s.util [ERROR] Halting process: ("Worker died")
java.lang.RuntimeException: ("Worker died")

Issue Links

• links to: GitHub Pull Request #2064