zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ben Bangert <...@groovie.org>
Subject Re: Testing expired sessions bug?
Date Tue, 21 Aug 2012 21:30:22 GMT
On Aug 21, 2012, at 1:41 PM, Jordan Zimmerman <jzimmerman@netflix.com> wrote:

> Note, though, that Curator does more than what the FAQ describes for killing the session.

Ah, perhaps I'm misreading the Java code. It looks like it operates in the FAQ manner except
it adds a watch for a non-existent node so that it can register a watcher for when it gets
the next event... which is *assumed* to be a session being lost. The sessionLostWatch will
get triggered for any session event, including a connection loss. I don't see anywhere that
the original client is watched specifically to see that it has gotten a EXPIRED_SESSION_STATE
event.

In the error I described originally, the first client *will* get its connection dropped, but
will not actually be expired. If I was only watching for any event, and not specifically an
expired session state, then my tests would appear to work.

I mainly program in Python, so its likely I'm missing something in how the Java ZK client
or Curator works. Can you explain the technique here a little more?

Cheers,
Ben
Mime
View raw message