On Aug 21, 2012, at 2:43 PM, Jordan Zimmerman <[email protected]> wrote:

> Yes, it watches the bogus node but it also blocks until the second ZK 
> instance successfully connects. I found that both of these things are 
> necessary for reliable session killing.

Unfortunately that doesn't seem to help in Python. I've set the second ZK 
instance to wait up to 30 seconds for a connection, and it never gets anything 
other than a EXPIRED_SESSION_STATE during its connection attempts. The second 
ZK instance is completely unable to connect using the sessionid/password of the 
first client. The first client never gets booted during these attempts, thus my 
original note on a possible bug either in:
1. Zookeeper
2. Zookeeper C binding
3. Zookeeper Python C binding

If you can run your session killing code in a repeating test hundreds of times 
without the second ZK instance ever timing out on its connection attempts, it 
seems its probably a bug in 2 or 3.

Cheers,
Ben

Reply via email to