=== rtheis__ is now known as rtheis === rangerpbzzzz is now known as rangerpb [14:02] harmw, https://code.launchpad.net/~harmw/cloud-init/growfs-gpart-fixes_for_fbsd/+merge/246075 [14:02] still needed ? [16:18] rangerpb, hey. [16:18] one more question... [16:18] yessir smoser [16:19] config/cloud.cf [16:19] you set dhclient_lease to null i guess [16:19] i dont think that should be necessary [16:20] i think i need to revert the other changes to that file too [16:32] Odd_Bloke, around ? [16:33] smoser: In half an hour. [16:33] Odd_Bloke, https://git.launchpad.net/cloud-init/commit/?id=648dbbf6b090c81e989f1ab70bf99f4de16a6a70 [16:35] rangerpb set the 'agent_command' to __builtin__ [16:35] which, you had (i believe) added that support, but as far as i can tell its not used in ubuntu [16:48] smoser, the __builtin__ was the key/value that lets it go down the route of discovery on azure [16:48] without it, none of the code is used [16:48] yeah. [16:48] dhclient_lease should probably be set to something distro specific [16:49] it allows fallback to a lease file which at present was only defined for ubuntu [16:49] The __builtin__ code there was written when we were building all-snaps images for Ubuntu Core Series 15, where walinuxagent wouldn't run. [16:50] We aren't presently using it for anything, as we don't publish Series 15 images any more. [16:50] fine but it is a conditional in the current code [16:50] totally open to change [16:51] i never much liked the conditional which was there, it seems in cloud-init we chase down every possible avenue [16:51] smoser, u follow? [16:52] i have no idea what a series 15 image is [16:53] ubuntu-core [16:53] based on 15.04/10 IIRC [16:53] oh. ok. [16:53] rangerpb, does fedora/centos run the agent ? [16:54] Odd_Bloke, what does the agent do for us ? why not use the new path. [16:54] the agent still has some upside over the current [16:54] not really in provisioning but more in configuration, etc [16:54] for example, you can pass a script "through" the agent and it will end up getting executed on the vm [16:55] well, cloud-init does that too :) [16:55] smoser: I think just because we've always been relying on walinuxagent. [16:55] smoser, i think the agent can do it anytime, not just on boot [16:55] We will still need to run walinuxagent in Ubuntu images (for other Azure-specific things to work), so we'd have to check that the two don't conflict. [16:56] But walinuxagent is reasonably configurable, so if it isn't already possible to stop it doing things that we want to do, we could probably request it. [16:56] Odd_Bloke, but does the agent start itself ? [16:57] cloud-init starts it, but is that necessary ? [16:57] the answer is it depends on what you are trying to do imho [16:57] https://git.launchpad.net/cloud-init/tree/cloudinit/sources/DataSourceAzure.py#n231 <-- this was the conditional i was referring to [16:58] basically two paths to do the same thing, agree Odd_Bloke ? [16:58] Yeah, but __builtin__ doesn't rely on walinuxagent. [16:58] yeah [16:59] so it might be better? that the conditional check for the existance of the agent? [16:59] But it was only ever used in an environment where walinuxagent _would not_ run, so we don't actually know how it works in an environment where we do __builtin__ _and_ walinuxagent runs. [16:59] well i do [16:59] i didnt see any harm in running both [16:59] Yeah, I don't see any harm in it. [17:00] Unless it breaks. ;) [17:00] But, fundamentally, the two should be fine side-by-side. [17:00] It's just whether or not there's something unconsidered that needs to be cleaned up. [17:00] It sounds like there isn't. [17:00] So that's good. [17:00] the problem is that the one path *requires* an agent, which might not be there [17:01] which is why i said, if the agent is present, prob prefer it; if not, go the "newer" route [17:01] then there is no collision on provision [17:01] smoser, Odd_Bloke however, if you prefer, I can run this by MS [17:02] rangerpb, sure. i'll have a few changes to my dhclient hook cleanup that i'll point you at. [17:02] I thought you just said you'd identified that you had run the two and there wasn't a collision? Did I misunderstand? [17:02] we can have the default behavior to do as you suggest [17:02] (Or just being cautious?) [17:02] being cautious [17:02] OK, cool. [17:02] but still want it to be configurable [17:02] sort of like, where does provisioning start/stop and other stuff begin [17:02] so that you could set the agent command [17:03] Odd_Bloke, in our images, if cluod-init did not start the agent... [17:03] would it not run ? [17:03] i would like to see the configuration have something like azure_provision: cloud-init|agent [17:03] or one is override, perhaps better [17:04] i.e. default is still agent and override with cloud-init [17:04] i gotta grab some vittles, will read back when i return [17:04] So I think the only restriction we would need is that it be easy for the default to be cloud-init. [17:05] So that we can backport changes more easily, without having to carry a large patch to do that. [17:05] But even that isn't a hard blocker. [17:07] smoser: http://paste.ubuntu.com/23070971/ is the walinuxagent unit file, which I believe means it would come up anyway. [17:07] We just start it earlier. [17:07] ok [17:15] http://paste.ubuntu.com/23070985/ [17:27] rangerpb, Odd_Bloke ^ is what i'm looking at [17:28] rangerpb, that'd mean that right now fedora would have to carry the agent_command setting [17:28] but i'm open to "if agent is installed" logic [18:15] ok === rangerpb is now known as rangerpbzzzz