On Mon, May 05, 2014 at 12:28:38PM +0200, Henning Brauer wrote: > * Claudio Jeker <cje...@diehard.n-r-g.com> [2014-05-05 10:50]: > > On Sun, May 04, 2014 at 10:39:33PM +0200, Vincent Gross wrote: > > > I am considering to write a daemon of some kind, and I was going over > > > OpenBGPd's sources to get some good fine-grained design examples. I > > > noticed that although all IO's are asynchronous, libevent is not used, > > > but I can't figure out why. > > I'm not a libevent-fan. So I didn't use it. > > For bgpd, libevent+kqueue vs poll plain doesn't matter, the number of > sockets is too low. > > > > So, is libevent not used by accident or by design ? in the latter case, > > > what is precisely the feature/design consideration that made it > > > unsuitable ? > > Don't use bgpd as an example. It was one of the first privsep daemon we > > did and at that time it was done without libevent. ospfd and all the later > > daemons use libevent. Their event loop is therefor a lot simpler. > > So yeah not using libevent in bgpd could be considered an accident. > > not an accident. > > however, when I wrote the initial bgpd bits, I didn't think there > would be so many daemons using its framework 10 years later. so things > changed, and there's no problem with that. > > wether you use libevent or not is a matter of taste imho unless we're > potentially dealing with a very large number of sockets, in which case > kqueue has advantages over poll. >
One of the big advantages of libevent is timeout handling for free. So you don't need to reinvent it on top of poll(). When a daemon has to handle a lot of different event sources libevent tends to be a lot easier since you don't need to write all the timeout handling and there is also no need to build up the pollfd array all by yourself. -- :wq Claudio
