We are straying off-topic. I also use publish-subscribe in my system, but I need router-router for point-to-point interaction.
So my original problem about messages being silently discarded before the identity exchange is finalized is still relevant. Let me elaborate on the original problem with an example. The example is against https://github.com/zeromq/zeromq3-x and error handling has been removed for brevity (but it is present in my original code) void *client = zmq_socket(context, ZMQ_ROUTER); zmq_setsockopt(client, ZMQ_IDENTITY, identity, strlen(identity)); zmq_connect(client, endpoint); sleep(1); /* Wait for identity exchange to finish */ zmq_msg_t message; zmq_msg_init_size(&message, strlen(data)); memcpy(zmq_msg_data(&message), data, strlen(data)); int rc = zmq_sendmsg(client, &message, 0); assert(rc == 0); If the sleep() function above is removed, then the subsequent message is silently ignored. On 2012-07-28 20:48, Cem Karan wrote: > OK, you're doing something similar to what I'm trying to do, except that > you're planning on expanding to a WAN. I'm going to outline what I'm doing, > but I want to note that what I have are a number of pieces of test code that > let me make sure that the way I'm using zmq makes sense. I haven't tried to > push large amounts of data through yet, so it may break. With that > disclaimer in mind... > > I started out with router-router connections, but switched to pub-sub over > epgm and administratively scoped multicast addresses > (http://en.wikipedia.org/wiki/Multicast_address#Administratively_Scoped_IPv4_Multicast_addresses) > for precisely the reasons you outlined in your original email; it isn't easy > distributing the addresses. In my code, each node has one publisher socket > and one subscriber socket. All publishers and all subscribers are connected > to the same address and port. This means that I effectively have a reliable > broadcast channel (epgm ensures that all receivers get the message, in the > correct order, etc.). I wrote a simple control layer/protocol on top that > allows all nodes to communicate some minimal control information, which in > your case might include zmq URIs. That should allow router-router > connections. > > I don't know if the IANA has any addresses in the ad-hoc block > (http://en.wikipedia.org/wiki/Multicast_address#AD-HOC_block) which are open > to individual developers, but that is another thing to look at; if you can > get permission to use an address in there, then the pub-sub trick would work > across the internet. > > Thanks, > Cem Karan > > On Jul 28, 2012, at 8:11 AM, Bjorn Reese wrote: > >> On 2012-07-28 13:02, Cem Karan wrote: >>> Can you describe your system a bit more? Are the peers on the same LAN, or >>> do they need to peer across the internet? Do the peers already know about >>> each other (as in, a small group with well-known identities), or do they >>> have to discover each other? etc... >> >> Initially, all peers are on the same LAN where they discover each other >> via ZeroConf (Bonjour). These local peers constitutes one cluster. >> Later this has to be extended to several clusters communicating over >> WAN. The exact means for this has not been determined yet. >> >> >> _______________________________________________ >> zeromq-dev mailing list >> [email protected] >> http://lists.zeromq.org/mailman/listinfo/zeromq-dev _______________________________________________ zeromq-dev mailing list [email protected] http://lists.zeromq.org/mailman/listinfo/zeromq-dev
