On 05/09/13 17:43, Paul Colomiets wrote:
Hi, Basically from what discussed in the IRC: 1. Using nanomsg protocol for the data is OK
Yes, but let's also have a look at alternatives. Is there anything standard and widely-used for this kind of thing? If so, we can integrate with existing tools.
2. The application name is got from the environment, e.g. NN_PROCESS_NAME=foo
Yes, but it should be optional. It's useful where actual command line is ugly and/or non-descriptive. In other cases using command line as process name by default should work OK IMO.
3. Socket names should (probably) be devised from topology IDs. For that to work the long awaited feature of nn_connect(sock, "topology://topology-name") should be implemented first. Details for the name service will be discussed in the separate thread.
Yes. This one is not clear. We'll have to do some experimenting to see what kind of socket identifiers would be appropriate.
Martin