=== med_ is now known as Guest80711 === mup_ is now known as mup [09:36] Bug #1520645 opened: Unable to enlist node in gMAAS === nagyz_ is now known as nagyz === cpaelzer is now known as cpaelzer_afk [10:14] Hello. I need help about adding a "vmware" (vcenter) node to MAAS. I added the Vcenter CA to the openssl store of the MAAS server but when I try to add a chassis, vmware I keep getting Failed to probe and enlist VMware nodes: (vim.fault.HostConnectFault) {#012 dynamicType = ,#012 dynamicProperty = (vmodl.DynamicProperty) [],#012 msg = '[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify faile$.. does MAAS have an alternativ === dooferlad_ is now known as dooferlad === cpaelzer_afk is now known as cpaelzer === smoser` is now known as smoser === lifeless_ is now known as lifeless === cpaelzer is now known as cpaelzer_afk === cpaelzer_afk is now known as cpaelzer [12:45] blake_r, are you around? === CyberJacob is now known as zz_CyberJacob [13:29] nagyz: yeah [13:31] blake_r, I was wondering about https://bugs.launchpad.net/maas/+bug/1495779 [13:32] not sure if people get notified for closed bugs, so pinged you here. :) [13:32] does my comment make any sense? if not, I'd be happy to elaborate [13:32] currently scripting setup of nodes so they use 2 interfaces for bonding and I was hoping I wouldn't need to click through 100+ nodes :P [13:33] nagyz: unconfigured on the UI is the same a mode=link_up [13:33] nagyz: so its doing the same thing [13:34] is there actually any difference if I leave it on the subnet vs unconfigured for an interface that won't have an IP actually? [13:34] nagyz: being on a subnet with mode=link_up is just like being unconfigured [13:34] nagyz: the subnet is just meta data to say this interface has access to this subnet, so it places that interface in that space, but unconfigured in that space [13:35] nagyz: it is designed that way for running something like the neutron gateway on that node, which needs its endpoint in a space, but must not have any ip address configured on that interface [13:36] ok, let me try if setting mode=link_up actually does what I want. [13:36] in the meantime, are you aware of any DNS/DHCP related issues in 1.9? [13:36] I think I'm still seeing a behaviour where enlisting just gets broken if I enable DNS support (but would need a lot of time to reproduce in a clean setup) [13:41] nagyz: that is very strange, let me know if you can get more detail on that [13:41] ok, I can try it in a clean environment later - I'd love to use the built-in DNS functionality instead of writing a sync script to designate... :-) [13:41] so going back on how to clear an interface... [13:42] basically by default after commissioning there is an interface on the subnet with an auto assign IP [13:42] my script then goes, creates a bond from two other interfaces, and sets the bond iface to static [13:42] and now this is the part where I'd like to clear the original interface. [13:43] http://pastebin.com/X8WwbLVt [13:43] this is the current iface definition [13:43] running interface link-subnet id id mode-link_up gets me an error message [13:44] "Cannot configure interface to link up (with no IP address) while other links are already configured" === zz_CyberJacob is now known as CyberJacob [13:44] if I click "unconfigured" on the web now, the links array becomesempty basically [13:50] I'm sure I'm missing something. [13:56] if I do an unlink-subnet, then the IP address association disappears [13:57] but even after I cannot run link-subnet ... mode=link_up [14:31] blake_r? :-) [14:39] nagyz: you need to remove all links and then it will go to LINK_UP automatically which is unconfigued [14:39] nagyz: you cannot set it to link_up with other links created [14:39] how would i do this from the cli? [14:39] nagyz: when you create the bond the parent interfaces will all go directly to configured [14:39] eg remove all links [14:40] nagyz: you need to call "unlink_subnet" with "link_id=" [14:40] that doesn't work as discussed in the bug report - I just get back the same as it was just with a different id [14:40] or I really don't understand something. [14:40] nagyz: when you create a bond the parents will go to unconfigured [14:40] nagyz: you dont need to do that manually [14:41] nagyz: and by default the bond will be unconfigured as well [14:42] yes, I understand that. the bond is working fine [14:47] so by default I have 4 interfaces, eth{0,1,2,3}. by default when I enlist it's using eth0 but I want to create a bond of eth1 and eth2. [14:47] bonding works, I've managed to script that fully [14:47] however, now even though it's pxebooting from eth0, it will only need to configure bond0 (and eth1 and eth2 as slaves) in /etc/network/interfaces === cpaelzer is now known as cpaelzer_afk [14:48] so what I want is to explain to maas that "please forget that eth0 had any networks" [14:49] unlink-subnet does clear the IP assignment (from the GUI I can see it going from "auto assign" to "unconfigured") [14:50] but I'm still looking for the CLI equivalent of setting the subnet to unconfigured on the GUI - and the referenced command in the bug (mode=link_up) is not that. [14:51] I'm not sure I can describe my issue any more clear :-) === cpaelzer_afk is now known as cpaelzer === cpaelzer is now known as cpaelzer_afk [14:52] is it clear now what I'd like to accomplish? === cpaelzer_afk is now known as cpaelzer === dannf` is now known as dannf [15:20] nagyz: if the only link on the interface is link_up, then it will be unconfigured in the WebUI [15:21] nagyz: if you need to remove other links use link_subnet [15:27] blake_r, there is only a single link there and no way to get rid of it from the CLI. [15:27] if I select "unconfigured" from the GUI then I see the links array as empty [15:28] nagyz: that is incorrect then from the UI, all links should never go away that is a bug [15:28] nagyz: the API is correct having one link is just like unconfigured [15:28] nagyz: that link needs to be LINK_UP [15:29] http://pastebin.com/BCjvCqvn [15:29] before and after clicking [15:29] mode stays link_up, but the subnet definition is totally gone [15:30] unlike with unlink-subnet, where it always stays there. [15:31] nagyz: if the link is "LINK_UP" the subnet is not used on the deployed node, it is only there for metadata and to place the node in that space "aka. the space the subnet belongs" [15:32] I understand that, but I'd really like to get rid of it - to do the exactly same thing as on the UI. [15:32] now, either the UI is faulty as I can set the subnet unconfigured (which removes the whole subnet definition for the interface) [15:32] or the CLI is inconsistent as it doesn't allow this [15:38] nagyz: then just do "link_subnet mode=link_up" on the interface once it has only that one "link_up" [15:39] I cannot do that as that raises the error that I refered to above [15:39] the point would be to replace clicking on the GUI to set the interface's subnet to "unconfigured" :-) [15:39] maybe I should have started with that [15:40] nagyz: well then that is a bug as you should be able to perform that operation [15:44] shall I open a bug? :-) [15:44] nagyz: yes please [16:31] Bug #1540453 opened: API doesn't indicate whether a node is deployable [16:43] hey [18:13] hi I'm trying to deploy a xenial over maas but it seem that the image is not able to load the following module when deploying : nls_iso8859_1 it is required to mount the efi partition in fat32 [18:17] boot.log show this : can't create directory '/root/lib/modules': Read-only file system [18:35] Bug #1540522 opened: Xenial deploy failed at efi mount [18:47] Bug #1540528 opened: [1.9] cannot scrub subnet information from interface [18:50] /wi/win 4 [18:56] Bug #1540528 changed: [1.9] cannot scrub subnet information from interface [19:05] Bug #1540528 opened: [1.9] cannot scrub subnet information from interface [19:29] Bug #1540522 changed: Xenial deploy failed at efi mount

[19:29] Bug #1540539 opened: MAAS installation: bind9 chokes on duplicate dnssec-validation setting [19:32] Bug #1540539 changed: MAAS installation: bind9 chokes on duplicate dnssec-validation setting [19:32] Bug #1540522 opened: Xenial deploy failed at efi mount

[19:35] Bug #1540522 changed: Xenial deploy failed at efi mount

[19:35] Bug #1540539 opened: MAAS installation: bind9 chokes on duplicate dnssec-validation setting [19:35] Bug #1540548 opened: MAAS installation doesn't create database config [19:41] Bug #1540548 changed: MAAS installation doesn't create database config [19:44] Bug #1540548 opened: MAAS installation doesn't create database config [19:47] Bug #1540548 changed: MAAS installation doesn't create database config [19:59] Bug #1540548 opened: MAAS installation doesn't create database config [21:05] I'm stuck with the "Getting started" document, I have two machines both running MAAS, I have em1 and em2 on both of them (em2 being the private network) [21:05] If someone is willing to give me a few minutes to help me get up and running, that will be wonderful and much appreciated [21:05] I'm not sure how to get the two machines to cluster, for example [21:38] why do you have two machines? [21:38] for simple installations you can run one machine [21:38] I guess depends on your node size, but for my ~120 nodes in the system, 1 maas is enough [21:39] I see [21:39] I just realized yesterday that the way this works is that you can setup MAAS itself even on a relatively weak VM [21:39] and let the hardware boot via PXE [21:40] but what's not clear to me is: the physical machine boots via PXE, how do you run multiple VM's on the same metal? [21:41] does some "layer" get booted first via PXE, which then allows running multiple VM's also via PXE? [21:41] not sure I even understand the question... [21:41] do you want to provision physical servers or VMs? [21:41] for example we use maas to provision our physical nodes but we use openstack for vms [21:42] maas is not orchestrating VM creation and whatnot [21:44] I want to use openstack for VMs [21:44] I have two relatively powerful machines with tons of storage [21:45] normally I would use something like proxmox to take as much advantage of the hardware as possible [21:45] (since all VM's are private, and absolute 100% hack proof isolation is not important) [21:45] so why not use proxmox? [21:45] for 2 machines you don't need maas... [21:46] esp since you can't use maas AND proxmox together as proxmox has it's own installer [21:46] I want to create a setup that mimics as much as possible what cloud providers offer [21:47] no cloud provider tells you how they provision their infrastructure, usually [21:47] and proxmox doesn't support any "cloud". [21:47] Skaag: maas uses VM's as if they were hardware [21:47] you can use maas to deploy the hardware and then install openstack using juju for example [21:47] roaksoax-brb, a quick question re. the bcache bug in 1.9 + the official trusty image: what's the plan there? :-) [21:48] nagyz: can you point me to the bug ? maybe the issue is not actually in MAAS but curtin [21:49] Skaag: When you tell MAAS "mannage this VM" or "This chassis (which can be a host with VMs") MAAS uses it as if it were individual hardware out there [21:49] https://bugs.launchpad.net/maas/+bug/1514094 [21:49] Skaag: MAAS won't make any distinction between a VM and a baremetal machine [21:49] right it's a curtin bug actually... [21:50] but breaks using bcache via maas :) [21:50] nagyz: seems fixed thought. What cucrtin version are you using ? [21:51] roaksoax-brb, whatever is in the image that maas uses to deploy trusty from the /releases repo :-) [21:51] nagyz: curtin is not in the images. Curtin is installed in the MAAS server actually [21:51] nagyz: dpkg -l | grep curtin in the MAAS Server [21:51] on the maas node apt-cache show tells me it's 0.1.0~bzr314-0ubuntu1 [21:53] nagyz: the bug seems fixed on rev304 and rev306 [21:53] nagyz: so that fix is in curtin [21:53] I can give it a go again but we've just tried creating bcache backed drives and it didn't work out [21:53] saw a bunch of modprobe bcache errors [21:54] nagyz: try using the MAAS daily image s? [21:54] nagyz: and please pb the intsall output, and I can raise it with the curtin guys [21:54] yeah let me give that a go [22:08] that nerve-wrecking moment when you do a "cat config-generated-from-python | ssh backbone-switch-1"... :) [22:09] the switch config was actually generated by a python script from maas's lldp discovery data [22:09] and it configures over 200 ports now [22:35] roaksoax-brb: I see, so I could in theory start an instance using whatever virtualization solution I have, and just tell it to boot via PXE [22:38] Skaag: correct [22:39] Skaag: or if you have pre-created/described say, 10 VM's in say KVM or VMWare ESXi, and you can literally add them all in one go with MAAS [22:39] Skaag: with the 'Add Chassis' from the webui [22:39] understood [22:39] so then what I need to do is install openstack directly on those two ubuntu machines, and MAAS on a separate machine, just to hold the images, and control provisioning [22:40] but then if I install openstack, what is the advantage of using a tool such as MAAS, wouldn't it be a duplicate? [22:41] Skaag: you can use MAAS/juju to install openstack directly [22:41] Skaag: openstack itself will allow you to deploy instances in the cloud, you don't really need maas for that [22:41] I see [22:41] Skaag: if you want to deploy workloads in those instances, you can use Juju with the OpenStack provider [22:42] but in my case because all I have is 2 machines dedicated to openstack, there's little benefit. I guess MAAS is really for larger outfits who deploy a LOT of hardware all the time (and dynamically) [22:42] and they can take them up/down depending on demand [22:42] and everything is 100% dynamic [22:42] Skaag correct [22:42] thank you, I feel I understand the structure much better now [22:42] Skaag: if say, you don't put OpenStack on those 2 machines, but you do have 50 VM's running, you can deploy those with MAAS [22:43] I have no idea how to do that [22:43] I have MAAS right now on those two machines [22:43] the idea with creating two instances was for redundancy [22:43] but I don't know how to make them aware of each other (cluster) [22:43] and further, I don't know how to create VM instances under maas [22:47] Bug #1540453 changed: API doesn't indicate whether a node is deployable [22:48] roaksoax-brb, would maas 1.9 be able to power on VMs on vSphere using it's API instead of WOL? [22:49] and if so, how does a "chassis" map to VMs in VMware? is it only based on prefix? or is it a folder name? [22:52] Skaag: MAAS doesn't yet have native HA [22:52] nagyz: MAAS 1.8+ IIRC supports VMWare products, but the support is based on python-pyvmomi's library, which is VMware's [22:52] nagyz: you dont really need WoL [22:53] nagyz: when you add a "chassis" you can send a prefix filter [22:54] roaksoax-brb, so if I actually want maas to start the VM, will it do it via calling the native API? [22:56] nagyz: yes [22:56] nagyz: maas never really supported VMWare until we added support on 1.8... or 1.7 even [22:56] can't recall of the top of my head :) [22:57] it would be cool to experiment with juju there instead of physical nodes:-) [22:57] (I know I know, I could use AWS or OS underneath as well) [22:58] any quick way to get back the system IDs from maas CLI sorted by the system name...? :) [22:58] tried to do it with | jq but haven't managed so far [23:01] I'd just pipe it through cut + sort [23:01] or maybe that's not very good since it would only give you the name [23:02] oh it's a json output. sorry, my bad. [23:03] Bug #1540453 opened: API doesn't indicate whether a node is deployable [23:03] | jq '.[] | {name:.hostname,id:.system_id}' | jq -s '. | sort_by(.name)' [23:03] this works [23:09] nagyz: you can use juju too i'd think [23:10] roaksoax-brb, right it only talks to maas :-) so if maas provisioning works then I can use juju [23:10] that's the current plan [23:10] ok, thanks for the help, off to bed. [23:10] I'll open some dns bugreports tomorrow ;-) [23:12] Bug #1540453 changed: API doesn't indicate whether a node is deployable [23:12] hehe [23:51] can someone point me to installing centos images in maas 1.8.3? I tried 'sudo maas admin boot-resources create name=centos/centos7 architecture=amd64/generic content@=centos7-amd64-root-tgz' do I need to generate an api key before trying to do this?