| jrwren | I was just reading https://insights.ubuntu.com/2018/03/20/lxd-weekly-status-39 and noticed something about cluster placement. Does anyone know when LXD got clusters and placement and where it is in the docs? | 16:21 |
|---|---|---|
| jrwren | Ah, https://lxd.readthedocs.io/en/latest/clustering/ there. no idea when this landed | 16:23 |
| rick_h_ | jrwren: it's new trying to make lxd more 'cloud-like' | 16:24 |
| jrwren | rick_h_: its great and sensible! | 16:24 |
| rick_h_ | jrwren: it's been in progress the last 2 cycles I think | 16:24 |
| rick_h_ | yea | 16:24 |
| jrwren | configured correctly it can be a nice alternative to a simple openstack. | 16:27 |
| rick_h_ | definitely | 16:27 |
| rick_h_ | less overhead by a long shot, 5 machines in a lxd cluster > those 5 machines just running OS components | 16:28 |
| rick_h_ | though when you outgrow and need SDN and storage and ... | 16:28 |
| jrwren | no need for complexity of glance, keystone, neutron (zomg complex!) | 16:28 |
| jrwren | yup, exactly. | 16:28 |
| jrwren | but still, if you are small enough... do everything you can to stay small and avoid needing neutron for as long as possible :p | 16:29 |
| jrwren | and cinder... UGH... cinder and swift are THE WORST! :p | 16:29 |
| rick_h_ | yea, it's a nice way to put something like guimaas to better work tbh | 16:30 |
| rick_h_ | take the 6 nodes into a lxd cluster and then juju deploy on top of that "cloud" across machines and such | 16:30 |
| jrwren | mmmhmmm. | 16:30 |
| jrwren | or if you are very lucky, skip juju :p | 16:30 |
| rick_h_ | :P | 16:31 |
| rick_h_ | still have to operate the stuff so <insert something here> | 16:31 |
| jrwren | operate? | 16:31 |
| jrwren | k8s I guess. :p | 16:31 |
| jrwren | but juju is probably best way to deploy k8s, and then you are back to guimass being juju managed. But I guess k8s on metal is better than k8s on Openstack. | 16:32 |
| rick_h_ | heh, back to running openstack | 16:32 |
| jrwren | so flexible! | 16:32 |
| rick_h_ | yea, definitely k8s on bare metal | 16:32 |
| rick_h_ | I don't get k8s on openstack tbh | 16:33 |
| jrwren | I do, but I don't like it. | 16:33 |
| jrwren | Its because k8s doesn't do any network security and since it is soft containers it doesn't technically do any OS security either and so you need the NOVA + Neutron for host and network security. | 16:33 |
| jrwren | It also lets you build many small k8s clusters in VMs instead of one large one. | 16:34 |
| jrwren | k8s doesn't do resource management quite as nicely as nova, again because of its container not VM nature. | 16:34 |
| rick_h_ | yea, fair enough I guess | 16:36 |
| * rick_h_ just get sad face at the resources running the stacks of stuff needed to run the application container | 16:36 | |
| jrwren | Me too. | 16:37 |
| rick_h_ | at some point the complexity has to out weigh the 'ease of deploy' of stuff | 16:37 |
| jrwren | It is why I still LOVE the stackoverflow approach. | 16:37 |
| jrwren | It doesnt' even have to be complexity. It can be pure cost. Running this stuff in the public cloud gets expensive pretty quickly. | 16:38 |
| rick_h_ | yea, I think the research was something like around 30 large instances it get more cost effective to run your own hardware? | 16:39 |
| * rick_h_ needs to look that back up, maybe it was a bit more than 30 | 16:40 | |
| jrwren | its subjective because it depends on so many things. I'm convinced you can pretty much make those number say wahtever you want them to say. | 16:40 |
| rick_h_ | fair enough, there's definitely a crossing point for folks I think | 16:41 |
| jrwren | definitely. | 16:42 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!