[08:51] <TheAbsentOne> Hi all, any juju gui developers here? Is it possible to add nodes in the gui that represent small entities (no services but rather parts of services). I'm researching the usefulness of abstracting a database/tabel from the actual server for example in modelling techniques (like juju)
[08:53] <TheAbsentOne> So one of the goals I want to achieve is having a nodes that actually pass information from the relations. For example I would create a wordpress node, a mysql node and a pseudo database node. The relation would then be from wordpress to the db and from db to the mysql one (instead of wordpress to mysql).
[09:19] <gsimondon> What are the LXD specific configurations that conjure-up sets up when deploying CDK?
[09:20] <gsimondon> I keep on seeing that note but nobody is actually getting into what specifically that is. I do not want to use conjure-up but I'd like to understand what I have to configure manually or using some other tooling.
[09:23] <gsimondon> I've installed https://api.jujucharms.com/charmstore/v5/canonical-kubernetes/archive/bundle.yaml manually by creating lxc containers manually, adding the machines to juju and then deploying the necessary apps and settings permissions.
[09:23] <gsimondon> However, kubernetes-master is stuck in "Waiting for kube-system pods to start"
[09:24] <gsimondon> And one kubernetes-worker that I have is stuck in "Waiting for kubelet, kube-proxy to start."
[09:25] <gsimondon> Of note is that the cni relation between these two components and flannel is established but I do not see an ip address that cni is bound to on neither
[09:54] <gsimondon> So the fourth step in the CDK spell for conjure-up is dealing with CNI. However, I can see only EC2 implementation. Can any dev shed a light on this please?
[10:00] <petevg> TheAbsentOne: nodes map directly to charms, so to do what you'd want to do with the juju as it currently exists, you'd need to write a charm that acted as an intermediary for the relation data. It might make sense to write it as a subordinate to the database. That would be an unusual use case for a charm, however.
[10:00] <magicaltrout> TheAbsentOne: the pattern for that currently is that when you connect wordpress to mysql ...
[10:00] <magicaltrout> alright petevg's geeking
[10:00] <magicaltrout> i'll leave him to it
[10:01] <petevg> magicaltrout: that's pretty much all I've got. A node is a charm, so you can write a charm. :-)
[10:02] <magicaltrout> TheAbsentOne: what would probabaly make more sense is you extend wordpress or whatever, and so when a mysql relation provides connection details or whateevryou could have a hook that adds stuff to the provided datadatabase for you
[10:03] <TheAbsentOne> Yes as a first implementation I want to make it as a subordinate, do you by any chance have a good example on how to implement that? The problem I'm encountering though is exactly that (the fact that nodes map on charms) I kinda want to create nodes that represent something way smaller
[10:03] <TheAbsentOne> petevg: magicaltrout : yes exactly but in the gui you won't be able to visualise that well I think?
[10:04] <petevg> TheAbsentOne: correct. The GUI implicitly displays relation data on the edges of the graph. I'm not sure how happy that makes you, from a formal perspective, though.
[10:04] <TheAbsentOne> The goal of my research/dissertation is to actually look for a way to add a layer/functionality to juju so that non operations-people can actually model the services. And to solve the problem of 2 services connecting to the same the table
[10:05] <magicaltrout> i once got shown that gsimondon and i have no idea on github where it is :)
[10:07] <magicaltrout> petevg might know where the CDK conjure up recipe is stored
[10:07] <magicaltrout> its inside that somewhere
[10:07] <TheAbsentOne> petevg: wouldn't it be interesting to have the ability to add something on the gui that is no real service/charm but an abstraction of an entity of a service/charm? ^^ like a topic, queue for message brokers or databases/tables with database servers
[10:07] <gsimondon> magicaltrout: some teams preferring launchpad is confusing the hell out of me
[10:07] <gsimondon> :D
[10:07] <magicaltrout> join the club
[10:08] <magicaltrout> ah hold on
[10:08] <magicaltrout> progress
[10:08] <magicaltrout> https://github.com/conjure-up/spells/blob/master/canonical-kubernetes/steps/00_process-providertype/lxd-profile.yaml
[10:08] <magicaltrout> thats what i got shown gsimondon
[10:10] <gsimondon> git cloned https://github.com/conjure-up/spells/tree/master/canonical-kubernetes 30 minutes ago and jumped 4 steps ahead instead of opening that folder
[10:10] <gsimondon> let's see if that fixes the CNI setup
[10:10] <magicaltrout> ha
[10:13] <SuneK> I added a new user and granted him superuser, and access to models. However when running juju status, he gets the following error: "ERROR current model for controller mycontroller not found"
[10:13] <SuneK> Any ideas? juju show-controller works
[10:16] <petevg> TheAbsentOne: it would definitely be interesting :-) Juju tries to avoid having a deep understanding of the services that it deploys, however. Everything is just charms and relation data. That makes it flexible and powerful, but does make it trickier to do what you're looking to do.
[10:17] <petevg> gismondon, magicaltrout: the conjure-up spell for k8s lives here https://github.com/conjure-up/spells/tree/master/canonical-kubernetes  It's definitely more machine friendly than human friendly, but it does document what conjure-up does when it deploys k8s.
[10:18] <petevg> nm. I'm behind on the conversation. Looks like you already found it!
[10:20] <kjackal> gsimondon: magicaltrout: here a a short description on the lxd profile https://github.com/juju-solutions/bundle-canonical-kubernetes/wiki/Deploying-on-LXD
[10:22] <gsimondon> kjackal: thanks. useful
[10:39] <kjackal> do you know if tere is a way to figureout if the provider is localhost from within a charm?
[10:41] <petevg> kjackal: if there is a way, I don't know it. I think the best practice might be to put a flag in the config that you reference.
[10:41] <petevg> Then in the README, you say "if you have condition x, set param x in the config."
[10:41] <kjackal> :)
[10:41] <petevg> That way, you're doing something based on an environmental requirement, rather than hard coding provider specific behavior in the charm.
[10:42] <petevg> kjackal: do I want to know why you want to know the provider? :-)
[10:44] <kjackal> there is this issue: https://github.com/juju-solutions/bundle-canonical-kubernetes/issues/505 petevg,
[10:45] <kjackal> on lxd deployments k8s will probably need to set proxy-extra-args="proxy-mode=userspace"
[10:45] <petevg> kjackal: you've already got the config param, then :-)
[10:47] <petevg> kjackal: The more I learn about this, the more it sounds like something to handle at the ops/conjure-up level. A charm really doesn't want to be aware of the provider.
[10:49] <kjackal> I would also prefer to not address this inside the charm, but conjure-up is only one way to deploy bundles
[10:50] <petevg> kjackal: true. And it is frustrating to try to deploy something and get weird errors and then get told to look in conjure-up. gsimondon can probably relate :-)
[10:51] <magicaltrout> i for one try and avoid our conjure-up overlords :)
[10:51] <petevg> kjackal: does it fail to talk to the Cluster IP service in a distinct way? Or is it just a generic connection failure?
[10:51] <petevg> You could have it switch proxy modes if it failed to connect.
[10:52] <petevg> But then you might end up with a bug where you erroneously change proxy mode when you have a transient connection failure, or are still getting set up.
[10:54] <TheAbsentOne> thanks for the answers though petevg and magicaltrout when my formal problem statement is finished I'll post it here as well for the expertise of you guys ^^
[11:29] <petevg> TheAbsentOne: You're welcome! I'll look forward to the problem statement.
[11:29] <magicaltrout> don't listen to him
[11:29] <magicaltrout> he's paid to say that....
[11:32] <TheAbsentOne> as long as my dissertation leads to something I don't mind forced answers or not ;)
[11:32] <magicaltrout> what course you doing TheAbsentOne ?
[11:32] <petevg> I'm actually being paid to be at a meeting right now. IRC is just me having fun :-)
[11:33] <magicaltrout> don't invest your cash in Canonical
[11:33] <magicaltrout> they're easily distracted
[11:34] <petevg> It's multitasking!
[11:34]  * magicaltrout watches his pennies draining away
[11:34] <TheAbsentOne> well it's actually my research to obtain my master degree
[11:35] <magicaltrout> nice
[11:35] <petevg> Excellent.
[11:35] <magicaltrout> i failed university 3 times :)
[11:35] <TheAbsentOne> My origin is kinda the same origin as the tengu-team :)
[11:35] <magicaltrout> ah those weirdos
[11:35] <TheAbsentOne> hehe xD
[11:36] <TheAbsentOne> I supose most thesis goals/subjects need to be a bit weird sometimes
[11:37] <magicaltrout> i wouldn't know!
[11:37] <TheAbsentOne> It's kinda hard getting everything clear and my juju experience wasn't too big either so I hope I'll manage
[11:38] <TheAbsentOne> but it's good to know there is an IRC ^^
[11:39] <magicaltrout> indeed part of my pull towards juju was the fact you get access to the poor developers who have to write the code
[11:39] <magicaltrout> so many open source projects run by corps don't give you that facility and you end up talking to other folks on the outside whilst we all try and peer in :)
[11:41] <TheAbsentOne> very true, but right now I'm having a hard time finding the state-of-the-art examples and aproaches as juju changed/evolved throughout different versions
[13:11] <gsimondon> kjackal: petevg: the instructions for configuring LXD profile for CDK were all that's needed to get CNI running in a manual setup. just an update
[13:11] <kjackal> :)
[13:12] <petevg> Awesome :-)