[zeromq-dev] NetMQ: Poller usage

[Giacomo Tesio]

Hi Doron, thanks for your kind help!

So with an XPUB/SUB connection over an inproc transport no hand-made
syncronization is needed? Have I understand correctly?


Giacomo


On Fri, Jun 28, 2013 at 9:19 AM, Doron Somech <somdoron at gmail.com> wrote:

> Hi Giacomo,
>
> The poller is actually a copy of the zloop from czmq, actually I'm not big
> fun of the event pattern but I want to make it as close as possible to
> CLRZMQ.
>
> 1. The start method is blocking, you should call if from a dedicated
> thread.
> 2. Stop(false) just signal the dedicated thread to stop but don't actually
> wait the thread is stopped, this is good when you want to close the poller
> but continue doing something else with your code, but if for example you
> closing the application and want to wait until the poller is completely
> stopped call it with true.
> 3. Actually you cannot call Stop(true) from the dedicated thread, you can
> call Stop(false) from the dedicated thread, but I don't have to implement
> this kind of socket, you can just call Stop
> 4. Just call the stop method
>
> Regarding the bonus question, you have to do some synchronization magic to
> make it work, for the example let's call the socket which bind server and
> the sockets which connect clients. So when the server binds no client is
> connected yet (with inproc you have to bind before you can connect), now if
> you will start send messages now the client will miss them.
>
> After the client connected the client should send a signal message to the
> server, after the server bind we wait for message from client, after this
> the server can start publish messages.
>
> If we are talking about pub-sub pattern you should have a sub socket as
> the client, xpub socket as server, now the client immediately send
> subscribe message and the server wait for the first message...
>
>   Doron
>
>
> On Fri, Jun 28, 2013 at 3:23 AM, Giacomo Tesio <giacomo at tesio.it> wrote:
>
>> Hi, I'm having an hard time understanding the NetMQ.Poller usage (here:
>> https://github.com/zeromq/netmq/blob/master/src/NetMQ/Poller.cs).
>>
>> The fact is, it doesn't expose any Poll method.
>>
>> I guess that this is done in the aim of OOP desing.
>>
>> But I'm not sure about its correct usage.
>>
>> My insight is that after adding the sockets to the Poller, I should
>> simply observe the sockets' events ReceiveReady and SendReady, but if so:
>>
>>
>>    1. does the Start method block? If not, this should mean that I don't
>>    have to use it in a dedicated thread, right?
>>    2. what are the pro/cons of Stop(false)?
>>    3. one of the socket I added to the poller is a "control" one that
>>    send a "STOP" command when required. In the event handler can I stop the
>>    poller and remove the sockets?
>>    4. what should I do to cleanup things (before context disposition)?
>>
>> Moreover I'd like to know if the socket events will ever be fired from
>> sockets that aren't in a Poller, or not.
>>
>>
>> Finally a bonus question (be patient, I'm a newbie here): I've read that
>> with ZeroMQ I can loose a few messages at connection startup (even if I
>> start the receiver before the sender): does this apply to inproc sockets
>> too?
>>
>>
>> Thanks a lot.
>>
>>
>> Giacomo
>>
>> _______________________________________________
>> zeromq-dev mailing list
>> zeromq-dev at lists.zeromq.org
>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>
>>
>
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20130628/1dda3b3f/attachment.htm>