McL0v1n | thedac: My juju version is 2.3.1-xenial-amd64 | 02:00 |
---|---|---|
McL0v1n | thedac, maas version is 2.3.0-6434-gd354690-0ubuntu1~16.04.1 | 02:03 |
jac_cplane | is there anyway to completly wipe a juju controller/env? I made the mistake of trying to upgrade juju version - which corrupted the env. while trying to recover/remove/clean I just released the bootstrap node and uninstalled juju apt-get remove juju. then tried to start from scratch - install juju,etc. the problem is there are still remnants of the old system. I just want to wipe and start from scratch. juju | 03:38 |
jac_cplane | kill-controller, destroy-controller all fail. | 03:38 |
jac_cplane | found this on the web - will this still work ? "sudo apt-get purge --auto-remove juju" | 03:40 |
McL0v1n | jac_cplane that doesnt work, i tried | 03:42 |
jac_cplane | yep | 03:42 |
jac_cplane | I just tried. btw- i guess there is no way to update juju | 03:42 |
McL0v1n | ops, i found the error with lxc's, This is in the logs: | 03:42 |
McL0v1n | ERROR juju.worker runner.go:392 exited "0-container-watcher": worker "0-container-watcher" exited: setting up container dependencies on host machine: could not find unused subnet | 03:43 |
McL0v1n | thedac ^^ | 03:44 |
pmatulis | jac_cplane, can you open a doc issue on the wipe stuff please? https://github.com/juju/docs/issues/new | 03:45 |
jac_cplane | McL0v1n - looks like i found the way. juju unregister <controller-name> | 03:46 |
jac_cplane | then create a new one - same name overwrites the prev. then you can bootstrap it | 03:46 |
McL0v1n | ahh, forgot about that | 03:48 |
McL0v1n | also juju kill-controller | 03:48 |
jac_cplane | juju kill-controller does not work in this case i mentioned. | 03:48 |
jac_cplane | I'll add to tghe docs | 03:48 |
McL0v1n | any idea about why my lxcs are not provisioning? The error i posted above is in the host machine's logs | 03:50 |
pmatulis | jac_cplane, see it, thanks | 03:59 |
McL0v1n | I think it might have to be with: https://bugs.launchpad.net/juju/+bug/1665648 | 04:03 |
mup | Bug #1665648: Juju 2.0.3 fails to deploy LXD container lxdbr0 overlapping subnets <juju> <maas> <juju:Fix Released by hduran-8> <juju 2.1:Fix Released by jameinel> <https://launchpad.net/bugs/1665648> | 04:03 |
McL0v1n | mup we posted at the same time | 04:03 |
McL0v1n | i'm using a /8 10. subnet | 04:03 |
=== gsamfira_ is now known as gsamfira | ||
mattyw | hey folks, can you have key names like "foo-bar" in config.yaml or do you need to use underscore (foo_bar) ? | 14:30 |
jam | mattyw: we do stuff like: | 14:33 |
jam | storage-default-block-source: | 14:33 |
jam | value: maas | 14:33 |
jam | source: model | 14:33 |
jam | so foo-bar seems fine in yaml | 14:33 |
mattyw | jam, ok great thanks | 14:33 |
jam | mattyw: yamllint.com likes it fine, too | 14:34 |
mattyw | jam, I've seen lots of charms use _, even going so far as converting them back to - in the charm code | 14:34 |
mattyw | jam, so maybe it was a problem and not anymore? | 14:34 |
jam | mattyw: being ok in yaml != being ok in Mongo docs | 14:35 |
jam | though I think it is ok there, too. | 14:35 |
mattyw | jam, charm config docs? | 14:35 |
jam | as we certainly use {"model-uuid": blah} | 14:35 |
mattyw | ok great | 14:35 |
mattyw | I'll go with - for now thanks | 14:35 |
ybaumy | when i bootstrap a controller to eg vsphere/dc1 and want another one in vsphere/dc2. what do i have to do | 17:06 |
ybaumy | i mean a failover controller | 17:07 |
ybaumy | or do i have to add another controller and move it with vmotion | 17:09 |
=== frankban is now known as frankban|afk | ||
hml | ybaumy: here is some documention on juju high availablity: https://jujucharms.com/docs/2.3/controllers-ha | 17:22 |
hml | ybaumy: the juju enable-ha help page has info on specifying already created machines to be controllers. not sure if the vsphere dcs are part of constraints or not. | 17:22 |
hml | ybaumy: you’d spin up a vsphere instance where you wanted it - then use juju add-machine ssh:…. to let juju know about it | 17:23 |
hml | ybaumy: but the —to or —constaints flags would be better if possible | 17:24 |
ybaumy | hml i did now a enable ha and moved one controller to dc2 | 17:27 |
ybaumy | hml couldnt find anything that relates to creating it in dc2 in the first place | 17:27 |
ybaumy | it also would make sense to distribute kubernetes etcd and master VM's to diffrent DC's | 17:30 |
ybaumy | but appart from that and some dns problems im happy so far with it | 17:31 |
hml | ybaumy: typically juju distributes between zones, | 17:34 |
hml | ybaumy: looks like juju considers a vsphere dc as a region | 17:35 |
hml | ybaumy: not sure if a region works with a placement directive or not. | 17:36 |
ybaumy | hml so i have to create yaml to create masters in region dc1 and then dc2 ... with constraints? | 17:36 |
hml | ybaumy: you don’t specify a region with a constraint: https://jujucharms.com/docs/2.3/reference-constraints#vsphere-provider: | 17:38 |
hml | rick_h: any ideas on how to do juju ha across regions (or vsphere dc)? | 17:44 |
rick_h | hml: hmm, can you add-machine into the region and try to enable-ha --to ? | 17:47 |
rick_h | hml: I think it'll be on your to make sure the controllers are reachable on a network level there | 17:47 |
hml | rick_h: that was my thought too, just wondering… | 17:47 |
rick_h | hml: yea, not tried it but I think that should work | 17:47 |
hml | ybaumy: ^^^ | 17:47 |
hml | rick_h: regions don’t work with placement directives correct? | 17:48 |
rick_h | hml: no, they're model level when you add-model and such | 17:48 |
rick_h | hml: as all of the model is in the same region, otherwise there could be issue | 17:48 |
rick_h | hml: region->region would need to be a CMR setup and controllers don't do that atm | 17:49 |
hml | rick_h: right | 17:49 |
ybaumy | rick_h: hml thanks for clearing that up. but from a HA perspective its suboptimal | 18:00 |
rick_h | ybaumy: I'm sure there's ways to do it differently but my experience with clouds has been regions are for fallback and cross region coms can be slower/more $$ then inner region. | 18:01 |
rick_h | having a controller in different regions raises all kinds of questions about the workloads in that region/etc. | 18:01 |
ybaumy | rick_h: from your perspective its ok to handle it this way. from mine it doesnt since datacenters are only few kilometers away | 18:04 |
ybaumy | rick_h: so HA is something im looking for | 18:05 |
rick_h | ybaumy: understand, it's an interesting perspective and if the networking/latency is good I can see that. It starts to feel like regions should be zones in that situation as far as spreading all units of workloads out across them as well. | 18:05 |
rick_h | I mean if the region goes boom than any workloads the controller is managing goes with it | 18:06 |
ybaumy | rick_h: well im talking on premise cloud though | 18:06 |
rick_h | ybaumy: right, do you have zones in each region? | 18:06 |
ybaumy | rick_h: thats a different approach | 18:07 |
* rick_h is curious how it's setup | 18:07 | |
ybaumy | currently we spread masters and workers and etcd 's accross two datacenters. pods are also setup in a way to handle workloads if one DC is out | 18:08 |
ybaumy | kubernetes is what we are interessted in with juju | 18:08 |
rick_h | ybaumy: which on prem cloud is this? /me reads back | 18:08 |
rick_h | oic, vsphere | 18:09 |
ybaumy | rick_h: our companies vsphere | 18:09 |
ybaumy | rick_h: we are using openshift atm from redhat but it costs .. though im looking into other directions | 18:09 |
rick_h | yea, I'm not familiar with vsphere as much. I know in maas/openstack the we think of zones as you're doing dc's in vsphere. It's why we're hitting this mismatch as we obviously really care/want to have HA whereever possible in Juju and the workloads running. | 18:10 |
ybaumy | rick_h: we would really like to see that | 18:10 |
ybaumy | rick_h: first i made test with maas and vsphere and juju and then on top kubernetes but maas lets is also not what i want. 2.3 is really neat | 18:12 |
ybaumy | juju 2.3 | 18:12 |
ybaumy | i like that its talking directly now with API | 18:12 |
rick_h | ybaumy: I'm glad you're finding Juju interesting. I wish I had better news on the cross zone HA setup. Like I was saying, you can work around it by manually adding machines I think, but then once you go deploy things they'll only go into the original region which isn't what you're going to want | 18:12 |
ybaumy | rick_h: maybe in the future. tomorow our architects will get a presentation with juju setup from me. i hope they like it | 18:14 |
ybaumy | as much as i do | 18:14 |
rick_h | ybaumy: in the long run it'd be interesting with 2.3 and cross model relations if the charms in k8s could treat relations across the models as a way to do its HA work and so you'd just deploy two k8s and relate them and they'd combine into a super k8s | 18:14 |
rick_h | ybaumy: cool, let us know if we can be of any help | 18:14 |
ybaumy | rick_h: i will give a feedback | 18:14 |
rick_h | <3 | 18:14 |
ybaumy | im the canonical guy though .. most ppl think of canonical as desktop client OS and services | 18:15 |
ybaumy | i hope i can change that | 18:15 |
rick_h | ybaumy: same here, we've put a lot of time and effort beyond the desktop :) | 18:19 |
ybaumy | rick_h: in germany i tell you its really hard to find ppl who use ubuntu and other serivces professionaly | 18:22 |
rick_h | ybaumy: ah you're in germany? Yea, that's a tough nut there. Wasn't/isn't germany still very suse friendly as they were from there? | 18:23 |
ybaumy | rick_h: we use suse for SAP of course cause there is no competition in this case. we use redhat and centos for everything else. but you are right SuSe is really popular | 18:24 |
ybaumy | i tried their container platform lol | 18:25 |
ybaumy | and ceph storage | 18:26 |
ybaumy | its full of bugs | 18:26 |
ybaumy | you just cant use it | 18:26 |
ybaumy | SLES enterprise server is really stable for SAP and HANA | 18:26 |
ybaumy | anyways i call it a day. beer is waiting. ttyl | 18:32 |
=== thumper is now known as thumper-away |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!