=== alexpilotti_ is now known as alexpilotti === harlowja is now known as harlowja_away === zz_gondoi is now known as gondoi === JoshNang_ is now known as JoshNang [17:40] smoser: ready for new release? [17:40] :) [17:40] dont' you avhe something else to do on friday night then bother me [17:40] forgive me for asking so often, I'm just eager to submit that fbsd port [17:40] hehe [17:44] you are completely justified there, harmw [18:02] * hiren_ likes harmw's persistence ;-) [18:02] lol [18:02] yea [18:45] damn i must of missed an important conversation, lol [18:45] dont you always [18:45] i think so :-P [18:46] of course right when i say that my thunderbolt crap cable comes undone, lol [18:46] thats what i don't like about the cable, it doesn't hold itself in, lol [18:46] stupid apple [18:47] :p [19:03] harlowja_: how sucky is the Y! cloud? [19:03] lol [19:04] define sucky [19:04] lol [19:04] define parameters to be sucky [19:04] haha [19:04] hiren_'s having a hard time getting cirros up and running [19:04] hmmm, not sure, haven't heard what he's doing [19:04] harlowja_: in our internal n/w [19:04] well shouldn't be that hard, idk how well cirros supports config drive [19:04] if it don't support config drive, then it may not work :-P [19:05] isn't there a normal metadataservice? [19:05] nope [19:06] just config-drive (until we revist that debate) [19:06] our security team didn't want a channel from VM -> controller layer [19:06] jeez [19:06] http://dpaste.com/184H91R [19:06] and metadata service is one such channel [19:06] wankers [19:06] ya, i tried fighting that, i gave up [19:07] Doing a metadata service in a secure and multitenant way is not an easy task :) [19:07] config-drive is nice and read-only, no two way channel... [19:07] putting an ISO partition on a VM/disk is super easy :) [19:07] so true [19:07] right, thats partially why its very attractive for security scaredfolks [19:07] harlowja_: actually, you can put ConfigDrives in vfat according to the spec [19:07] ya, vfat, iso9660 [19:07] harlowja_: so it doesn't have to be r/o... although we implement it as an iso9660 [19:07] right [19:07] well r/w it will still be a local thing [19:08] r/o is just more of an optimization [19:08] so hiren_ the following seems odd [19:08] cp: write error: No space left on device [19:08] cp: write error: No space left on device [19:08] cp: write error: No space left on device [19:08] cp: write error: No space left on device [19:08] failed to copy results from configdrive to /run/cirros/datasource [19:08] :-/ [19:08] that would seem to be an issue, lol [19:08] heh yeah [19:08] :) [19:09] brb. [19:09] harlowja_: there is dhcp or did they skip that as well? [19:09] not for vms [19:10] for the existing way baremetal is deployed, it uses dhcp [19:10] and pxeboot [19:10] so they must get their config from configdrive, right? [19:10] yes [19:10] hmk [19:11] interesting [19:11] * harlowja_ we'll move over to ironic which afaik has a similar dhcp, pxeboot, ipmi as the thing we already have (people are shifting from that team to ironic) [19:11] JayF will get a bunch of new iroinc friends (besides rloo) [19:11] soon enough, ha [19:11] :P [19:11] You guys going to use ipa? [19:11] ipa? [19:11] unsure [19:12] JayF probably, although i'm not that in touch with all that stuff, haha [19:12] whats ipa :) [19:12] ironic-python-agent [19:12] ah [19:12] the newest, sexiest deploy driver for ironic [19:12] :P [19:12] lol k [19:12] i'm still clueless when it comes to ironic :p [19:12] * harlowja_ mostly me too, ha [19:12] can't be everywhere :-P [19:13] :p [19:13] y! will likely have to figure out a different mechanism than config-drive for ironic stuff though, something like maybe what JayF and folks have done (writing the config-drive on the main disk) [19:13] attaching a cd isn't gonna work, lol [19:14] or maybe we'll revisit that whole networking metadata webservice debate [19:14] :) [19:14] Dude, we already have it handled [19:14] even better, ha [19:14] smoser: did that configdrive change for fbsd got merged yet? from raginbajin? [19:15] i dont think so. [19:15] there was some work to still dothere. [19:15] harlowja_: Yeah, I am not sure if it's fully upstreamed, but it works entirely and we'll help you get it working if you need :) [19:15] hm ok, [19:15] harlowja_: nova is the one who generates teh configdrive... it just gets passed through ironic, to the agent, into the drive [19:15] well I need it to test with configdrive, since Y! apparently requires/uses that [19:15] JayF ya, that seems nutty :-P [19:15] harlowja_: we just create a small partition at the end of the disk, and write it out [19:15] to much passing of crap around, lol [19:15] harlowja_: that's the Openstack Way(tm) [19:15] :( [19:16] for ocnfig-drive on bare metal... [19:16] harlowja_: in our original implementation, the agent built the iso :( [19:16] JayF why doesn't nova just pass the data that is the config-drive to ironic [19:16] seems saner [19:16] my opinion is that long term, nova (or deployer, whatever) has to be *very* stupid. [19:16] then passing a binary disk [19:16] and just dd stuff to the first disk [19:16] and then look for a partition on that disk named "config-drive" or re-use the EFI partition or something [19:16] and write the data in there. [19:17] in that way, the image is smart [19:17] harlowja_: because Nova is used to just writing it out itself [19:17] and tine installer is dumb [19:17] sux [19:17] and you have a standard. [19:17] smoser: that's pretty similar to what ipa does [19:17] nova's like an old grandpa or something, lol [19:17] sort of similar. [19:17] but ipa (i think) has to know how to "make it boot". ? [19:17] smoser: puts an image on a disk, looks for a partition labelled 'config-2', if it exists, we dd the configdrive into that partition, if not, we create it at the end of the disk [19:17] smoser: the bootloader is entirely in the image [19:17] it dd the disk ? [19:18] yeah [19:18] including the partition table ? [19:18] let me link you the code [19:18] yeahhh [19:18] nice. [19:18] now we'll have to support the partition-images eventually [19:18] yeah, that is theo nly way you can actually accomplish secureboot [19:18] bceause --preserve-ephemeral (hate) [19:18] meh [19:18] partition images dont have to be supported. [19:18] but I would not use it :) [19:18] smoser: tell that to Devananda, hahaha [19:18] its just not scalable. [19:19] it means that the installer has to know information about the things its installing [19:19] smoser: https://github.com/openstack/ironic-python-agent/blob/master/ironic_python_agent/shell/write_image.sh + https://github.com/openstack/ironic-python-agent/blob/master/ironic_python_agent/shell/copy_configdrive_to_disk.sh [19:19] --preserve-ephemeral could just be "i write as many blocks as are in the image. no more , no less" [19:20] you want presevation of data then keep the thing i'm writing small. [19:20] I'm not sure I agree, but I also hate the same feature you hate which is funny [19:20] haters gonna hate [19:20] where is /tmp ? [19:20] is that dis ? [19:20] smoser: that code runs from inside the ramdisk agent [19:20] and why do you "erase mbr from device?" [19:21] so qemu-img convert -O raw $IMAGEFILE /tmp [19:21] puts it into the disk :) [19:21] er, into ram [19:21] means that you have enough ram to contain the entire contents of the disk? [19:21] yeah, right now [19:21] we want to make that stream eventually [19:21] good thing most people have way more ram than disk space [19:21] oh wait [19:21] :) [19:21] heh [19:21] in this case? every server it's running on [19:21] has at least as much disk as ram [19:22] in 2/3 cases it's less disk than ram [19:22] (for the OnMetal case specifically) [19:22] curtin reads tar file extracts to stdout | dd of=/dev/foo [19:22] yeah qemu-convert doesn't output to stdout [19:22] if it did I would've done something like that [19:22] well, you dont have to support a silly format :) [19:22] someone is working on vhd support, which I think that whole bucket of things will get refactored [19:22] (and i fully realize i tell everyone they have to support qcow2) [19:23] It is what it is man :) We're making it better, but it works now, and it works pretty reliably [19:23] I have trouble finding too much fault with working software :P [19:23] yeah. [19:23] oh. [19:23] fyi, qemu-img convert -O raw http://foo.bar/my-image /dev/sda [19:24] that requires seek support in the http server [19:24] but works otherwise. [19:24] while this is a shitty answer for purposes of open source [19:24] when we implemented it, we found it was much faster do convert, then dd [19:24] because we could set dd options that made imaging on our satadoms much faster [19:24] yeah. realistically you need to just fix the image format [19:24] why did you get an image format in qcow2? [19:24] O raw http://foo.bar/my-image /dev/sda would be neat if nova hosted the image [19:24] this was when this was still teeth-agent not ironic-python-agent yet [19:24] that doesn't make any sense. [19:25] or ironic/nova [19:25] smoser: I don't know man :) that was a long time ago [19:25] harlowja_: right now we only support images from swift temp urls [19:25] harlowja_: so it's already a direct download [19:25] cool [19:25] so the easiest solution is to fix the producer of the images to create something that streams. [19:25] except for non-swift users, haha [19:25] Well yeah man, we made it work for our case [19:25] and now the gaps get filled in as ironic gets more upstream support [19:25] wfm [19:26] the agent has support for probably a half dozen things that won't land until k [19:26] sounds like openstack [19:26] lol [19:28] reminds me gotta get my plane tickets [19:28] smoser u going to paris? [19:29] * JayF will be in paris [19:29] we should have a cloud-init design thinger one evening [19:29] sure [19:29] i will. [19:29] we should. [19:30] i was wanting to say we shoudl have a hack fest or somethign. [19:30] I mean, my brain will be mush, almost certainly [19:30] i'm up for either [19:32] we can have pow wow [19:33] Paris is first week of november, right? [19:33] yeah. [19:33] sad that's not an option though [19:33] i go from paris (openstack) to re:invent (aws) [19:33] cool [19:33] harmw, :-( [19:33] yea well, it's my first week at $newjob [19:34] so not wise to jump in plane instead of meating $newppl :P [19:34] meat them [19:34] haha [19:34] those meatbags [19:34] lol [19:34] damn [19:34] :) [19:34] thats probably the most lame typo ever [20:09] nice, Jenkins finaly archived my 3 shiny cirros images === alexpilotti_ is now known as alexpilotti === gondoi is now known as zz_gondoi === alexpilotti_ is now known as alexpilotti