[IPython-user] TaskClient reconnect

Vishal Vatsa vishal.vatsa@gmail....
Tue Dec 9 04:47:29 CST 2008


So I have been looking into this a little bit more, and as I restart
the controller,
it keeps changing the port number that binds for the task interface,
which is ok.

So I used the --client-port force the port to be the same, still no
luck though,
I am starting the task controller on a remote machine and using the block
task client in an ipython shell. I keep seeing this, when I know the
furl is not
changing.

DeadReferenceError: Calling Stale Broker


Any Ideas?

Thanks,
-vishal
2008/12/8 Vishal Vatsa <vishal.vatsa@gmail.com>:
> 2008/12/8 Brian Granger <ellisonbg.net@gmail.com>:
>>> Quick question about the behavior of foolscap connections.
>>> Say I have an async TaskClient connected to a remote TaskController
>>
>>> The TaskController goes away (network interruptions or inadvertently
>>> restarted), currently I get a Stale Reference Broker. At this point
>>> should the client tub not reconnect and get a new broker reference
>>> automatically.
>>
>> Are you saying that the actual controller process goes away or that
>> the TaskClient goes away?
>
> The task controller goes away.
>
>> Could you list the events that happen to trigger this?
>
> Mostly I have just been re starting the controller as I would like every thing
> to be loosely coupled (and we are on a shared filesystem so fresh furls are
> available)
>
>> Also, are you using the blocking (client.py) or the asynchronous
>> (asyncclient.py) TaskClient?
>
> In my app I am using asyncclient, and I also use the blocking client in
> the ipython shell.
>
>>> If this is not the case, whats the best way to deal with kind of
>>> situation, currently I am doing things in a really messy way, where I
>>> try to catch foolscap errors in defereds and explicitly try to get a
>>> new async taskclient.
>>>
>>> As it was pointed out to me, it would make more sense if this was handled
>>> at the protocol level.
>>
>> Yes, this is probably true, but I want to understand more of what
>> exactly is going on first.
>>
>> Cheers,
>>
>> Brian
>>
>>> Any Thoughts?
> Thanks,
> -vishal
>


More information about the IPython-user mailing list