Bruno Dumon

I follow your reasoning, until the last event: T2: processBackgroundResult(event) gets called on a closed connection There is no connection closed by NodeCache or PathChildrenCache. The point of the patch...

Hmm, I don't see any such check? Is this in NodeCache or PathChildrenCache?

In those stack traces, it is CuratorFramework/CuratorZookeeperClient which check if their state is started (which is stil true, since we have not closed them, we have closed the cache), not...

For PathChildrenCache, there was an existing test added as part of https://issues.apache.org/jira/browse/CURATOR-141, I've modified it a bit so that it is more likely to reproduce the problem. I also added...