Hi Martin, On Sun, Sep 15, 2013 at 8:49 AM, Martin Sustrik <sustrik@xxxxxxxxxx> wrote: > On 14/09/13 23:57, Paul Colomiets wrote: > >> Then internally, it's a set of counters in socket structure. And a >> single error state per socket. >> Can we agree on this? > > > The errors should definitely exist on per-connection basis. Which means 1 > error per endpoint -- on the connecting side; not sure about the binding > side. > > I guess there may be also some errors on nanomsg socket level, although I > cannot think of one atm. > > As for statistics, I would say one set of counters per nanomsg socket should > do. Going down to the per-connection level may be an overkill. > Ok. So my strategy would be: 1. Make a branch "monitoring_internals" with counters, error states and stub functions that submit data 2. Make a branch "monitoring_impl" that submits data via ESTP + syslog. At least these protocols are easier for me to play with 3. Reiterate both branches with different semantics/protocols, until satisfaction comes :) What do you think? -- Paul