ah, ok - sorry, I do use Ngnix, but a bit like a caveman :) we (as in, Mesosphere) recommend that Mesos (master/agents) are run under supervision (either systemd or upstart or supervisord or your favorite supervisor) so that when the process gets killed (by accident / bug / planned restart) the supervisor will restart it (as the case may be, with the new configuration).
Maybe that is the best paradigm then. Use HA setup with Zookeeper and roll config to one master, restart, wash, rinse, repeat? yup! look, ma', no downtime :) -- *Marco Massenzio* Distributed Systems Engineer http://codetrips.com On Tue, Nov 24, 2015 at 8:20 PM, Adam Avilla <[email protected]> wrote: > On Tue, Nov 24, 2015 at 3:15 PM, Marco Massenzio <[email protected]> > wrote: > > > On Tue, Nov 24, 2015 at 10:45 AM, Adam Avilla <[email protected]> wrote: > > > > > Is there any way to support HUP reloading of config with this design > > > paradigm? > > > > > > Can you please elaborate what do you mean by "support HUP reloading"? > > I don't suppose anyone really runs Mesos in production inside a terminal? > > > > I mean like how webservers and other systems work. NGINX is an example of > the paradigm: > > http://nginx.org/en/docs/control.html > > you will see HUP spawns new workers with new config, gracefully shutting > down old workers. > > Use case is config management system copies new configs to masters, HUP > to reload, no downtime. Maybe this is already possible? If so, excuse my > ignorance :$. > > > > > Is HUPing mesos even possible / helpful? From an end user > > > perspective it sounds useful, but I am not sure if even possible for > > mesos > > > to accomplish safely. > > > > > > > If you are running Mesos in multi-Master configuration (using, eg, > > Zookeeper) yes, it is safe to re-start the process, so long as you always > > maintain a --quorum of servers running. > > In fact, that's the recommended way to do a Mesos cluster upgrade. > > > > But, again, I don't really think this is what you meant by "HUPing > Mesos"? > > (I mean, if you `kill -1 <PID>` this seems to just terminate Master - at > > least on OSX) > > > > Maybe that is the best paradigm then. Use HA setup with Zookeeper and > roll config to one master, restart, wash, rinse, repeat? >
