On Mon, Jul 1, 2013 at 1:09 PM, Hermann Lauer <Hermann.Lauer@iwr.uni-heidelberg.de> wrote:
Hello,
[...]After restarting all shinken processes on our monitoring machine
broker died after a while with the appended traceback, which
seems to be an pyro incompatible version on a remote machine.

Hopefully the new protocol in 1.6 will make broker more stable
between remote machines. We'll watch out here if this happens again.

Thanks for your work on improving shinken,
greetings
  Hermann

Hi,

Yes, that's one of the pyro flaw I think, that you must have a strictly same version every where. That's why you can have a look at the http-core branch, there is no more Pyro :)

But it's quite a very unstable version, so just give a look and a test, but please don't even think about pushing it in production tomorrow :p

The http-branch remove Pyro, I switch to an http connexion (wsgi module for listening connexion, with some threads and locks), and pycurl for http connexion to other daemons. I'm still fixing some timeouts things here and here, but the main idea is OK. It will be possible to add HTTPS connexions, and even to reverse-proxy some distant daemons if you want (like a unique nginx that will reverse schedulers+poller connexions for a realm).

So in the 2.0 (won't be 1.6 ;) ) there won't be any Pyro problems (and I hope no http ones :p ).


Jean