JL 2012-03-15, 17:41
Patrick Hunt 2012-03-17, 15:57
I am using Java for this.
The approach you suggest was the first one that I tried, but I was getting a session error and the ZK client was going into the CLOSED state. The server is being cleanly shutdown. As far as I can tell (from the logs and skimming through the code path), the server terminates the client session(s) during shutdown. I'm guessing it persists the session termination in the log (or elsewhere), so on restart it tells the client that this is an invalid session if the client tries to reconnect with the same session id (which I'm guessing the client does not try to do if the session termination was clean).
Nevertheless, I'll play a little bit with the timeout parameters and dive in a bit.
Thanks for the prompt reply.
On Mar 14, 2012, Mark Gius <[EMAIL PROTECTED]> wrote:
With the python client (thin wrapper around C) I was able to simulate
CONNECTING in a single node scenario. Just shutdown the only zookeeper
server and the client will notice this real quick. Turn zookeeper server
back on and client should reconnect within a few seconds. You'll need to
make sure your recv_timeout is high enough so that the client doesn't give
up and fire a DISCONNECTED.
On Wed, Mar 14, 2012 at 11:06 PM, Patrick Hunt <[EMAIL PROTECTED]> wrote:
> This is the c client? (In java I know we have a zk subclass that
> provides this.) You can use jmx on the server, there is a
> terminateConnection method. Perhaps an easier way would be to simulate
> Server failure - e.g. start a quroum of 3 servers, Have your client
> connect, then shutdown the server that the client connected to. The
> client will be disconnected and reconnect to one of the remaining 2
> servers. Our zk tests do this.
> On Wed, Mar 14, 2012 at 11:04 AM, JL <[EMAIL PROTECTED]> wrote:
> > Is there a way of forcing the ZK client to go into CONNECTING state
> after having reached the CONNECTED state, but without killing the session?
> > We are aware of the FAQ entry for "Is there an easy way to expire a
> session for testing?", but that does not achieve what we want as it kills
> the session.
> > We'd like to be able to essentially test the sequence of going:
> > - From CONNECTED to CONNECTING;
> > - followed by a transition from CONNECTING to CONNECTED.
> > Thanks,
> > -Julio
JL 2012-03-14, 18:04
Patrick Hunt 2012-03-15, 06:06
Mark Gius 2012-03-15, 06:56