/srv/irclogs.ubuntu.com/2020/06/30/#juju.txt

timClicks[m]We have an inconsistency wrt how we describe LXD04:24
timClicks[m]Within the CLI, it's the "localhost" cloud04:25
timClicks[m]But that term isn't used in the public docs04:25
jamwallyworld, kelvinliu Is there an easy way for a Unit agent to determine the name of the K8s pod names? (or for 1 unit to find the names of all pods)05:24
jamwithout hitting the K8s API?05:24
kelvinliuu mean to find that unit's workload pod?05:25
jamkelvinliu, right. AIUI, there are config options where you need to configure Mongo with the pod DNS names for its replica set05:26
jamI'm mostly acting as a proxy for the question from MarkMaglana05:27
kelvinliui assume it's a statefulset05:31
kelvinliuand the pod's names are fixed05:31
jamkelvinliu, for Mongo I'm sure it would be  (if you need to configure the name of pods before the pods exist, they have to be fixed)05:31
jamkelvinliu, I don't know the exact use case MarkMaglana is requesting. Just that he wanted to know the names of the pods, and I'd like to have general answers for them05:32
kelvinliuwe do have the pod name in `cloudcontainers` for statefulset05:34
kelvinliuim thinking he can get it05:35
kelvinliuhow the uniter can get it.05:35
jamkelvinliu, that is in the pod spec or ?05:36
kelvinliunah05:36
kelvinliuwe get the pod name after the pod has been created05:36
kelvinliuso we update it into that doc once k8s tells us the pod name05:38
kelvinliuwe can easily get the pod name by k8s API because we have proper labels and annotations to look for05:39
jamkelvinliu, sure, but we don't currently expose that to the charm, do we?05:41
kelvinliuwe have k8s python client built into our operator image already05:41
kelvinliuso charm can have a in-cluster k8s client to get access of it05:42
jamkelvinliu, ok. But that isn't talking to Juju or coordinating with Juju data. That said, it is still stuff that we can point charmers at.05:43
kelvinliuhttps://github.com/kubernetes-client/python05:43
kelvinliuI don't think currently charm can get the pod name from juju for now05:44
kelvinliubut if it's the action scripts want to know the pod name then it's easy05:45
kelvinliubecause k8s expose the pod name in ENV inside the pod itself05:45
kelvinliujuju run --unit mariadb-k8s/0  'env | grep mariadb-k8s-0'05:46
kelvinliuHOSTNAME=mariadb-k8s-005:46
jamkelvinliu, so I'm pretty sure this is more about during eg 'config-changed' than during an action.05:55
jamAnd wanting to know all pod names05:55
kelvinliuprobablly we have to let the leader to filter the pod list using application name labels05:58
kelvinliuwait u mean he want this pod name list to generate the replica string?05:59
kelvinliuthen the charm should be operator charm06:03
kelvinliufor an operator charm to provision a new instance of mongo, it might be reasonable to do it via an action?06:05
kelvinliuim not quite sure what's the current status of operator charm feature yet. might be good to discuss on mattermost further06:06
jamkelvinliu, so as mentioned I don't know the exact use case. If it was something like Mongo, I don't think it quite fits the idea of an operator charm, other than you can solve anything in some way eventually :)06:12
kelvinliuok06:34
stickupkidachilleasa, approved the integration test changes08:39
achilleasastickupkid: cool. thanks! no issues with the changes in the arg parser then?08:45
stickupkidachilleasa, nope, I tested the output of all flags08:45
achilleasamanadart_: what do you think about moving ProviderInterfaceInfo into core/network? this (https://github.com/juju/juju/blob/develop/environs/networking.go#L98) is the only thing in the Networking interface that forces you to import juju/network09:21
manadart_achilleasa: It's already duplicated. See core/network/nic.go09:24
manadart_achilleasa: So by all means, get rid of the juju/network one.09:24
achilleasaah nice! will do09:24
manadart_achilleasa: And we should be using HardwareAddress over MACAddress, because Infiniband devs do not have a MAC.09:25
achilleasamanadart_: ok, I will try to rename it... hopefully it won't break too many tests :D09:26
manadart_achilleasa: NBD if it's a hassle.09:26
manadart_achilleasa: Can you look at this one? https://github.com/juju/juju/pull/1177811:18
achilleasalooking11:18
achilleasamanadart_: running QA now; in the meantime can you look at https://github.com/juju/juju/pull/11779?11:30
manadart_achilleasa: Yep.11:31
achilleasamanadart_: seems to fail during the merge: https://pastebin.canonical.com/p/Cr25N4KymT/11:47
manadart_achilleasa: Did you change the MAC address? Let me try it again...11:51
achilleasayes, I changed name, mac (the :fe:fe at the end) and providerid11:52
manadart_achilleasa: I think omitting the _id has affected this. I ran it again with success. https://pastebin.canonical.com/p/g9sq3qFxf7/12:13
achilleasamanadart_: ah crap... forgot that IDs are generated by us... sorry12:14
achilleasaapproved12:15
=== arif-ali_ is now known as arif-ali
=== arif-ali_ is now known as arif-ali
yan0sIs it possible to add a new LXC having an interface on a space that its host didn't originally had this space?19:18
yan0sbut the host now has a manually provided interface on the subnet of this space19:19
yan0sI want to add the LXC on a new vlan interface wityhout having to redeploy the host19:22
pmatuliscan i configure a lxd profile on a per-machine basis?19:57
pmatulishml, ?19:57
hmlpmatulis:  with juju?  or in general?19:57
hmlpmatulis:  you can add a profile with more info to an existing machine, the machine may get rebooted.19:58
hmlpmatulis: not sure if juju would pick that up or not19:58
pmatulishml, ideally with juju19:59
hmlpmatulis:  juju lxd profiles are via the charm and per application.19:59
hmlpmatulis:  you can add what you want to the default profile and juju uses that for all machines20:00
pmatulisright ok20:00
pmatulisso i will have to do it manually then20:00
pmatulisthat too bad. maybe i will open a wishlist bug20:01

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!