[00:00] Yeah, it would have to be a udeb, that was implied. [00:00] It wouldn't even have to depend on everything - just kernel-image-blah-di would be enough [00:00] Of course I'm not totally certain proposed-migration checks deps for udebs [00:00] Sure, as long as it catches all the kernel images used per arch, that would work too. [00:00] Oh and cdimage also needs to know the ABI, so it's a bit more than just that [00:00] Hence why I'd like to be able to do it in germinate somehow if possible [00:00] cdimage knows the ABI due to the seeds. [00:00] Which have to be in sync [00:00] And if you seed that udeb, it pulls in all the deps. [00:00] Hence the problem [00:00] That was my point. [00:01] Hm. Maybe. [00:01] Yeah, for that it'd have to be everything relevant [00:01] You'd probably need one per ABI [00:01] So, yeah, it would need to depend on all the ones in the manifest, not just the kernel images. [00:01] I don't see why the metapackage would need to be one per ABI, though? [00:02] The whole point is to avoid that. [00:02] Because we haven't always had all the ABIs seeded in the installer seed [00:02] I mean flavour [00:02] I'll think about it - it seems clunky to me somehow, but might be the best available [00:02] Basically, exactly what the fake "debian-installer-images" depends on today. If that were seeded, it would just work. I think. [00:03] And yeah, we could further divide by flavour, but we don't currently. The ARM server image actually contains all the flavours' udebs. [00:03] Which is a bug, to be sure. [00:03] But germinate has no concept of subarches. [00:03] So, it's a bit tough. [00:04] If cdimage udebs could pull from a metapackage, we could actually fix that. [00:05] The installer seed certainly hasn't always included all flavours, and this is deliberate and necessary for cdimage - we'd have to reflect that [00:05] You don't have to have germinate have a true concept of subarches - just have debian-installer-udebs-generic etc. [00:05] Right. [00:05] That's kinda the conclusion I was coming to in arguing against it. [00:06] I still don't quite like it in some visceral way, but it does sound workable, and it's better than what we have now. If you want to do it, be my guest :) [00:07] Just make sure that it's Priority: extra [00:07] (Or maybe optional. >= standard would be bad.) [00:07] My usual MO is optional, and wait for the archive to tell me I'm wrong. [00:07] extra would be best here, I think. [00:07] But by Policy rules, since nothing conflicts with it, optional should be right. [00:08] priority-mismatches will yell until it lands in the right tier anyway, no? [00:08] priority-mismatches doesn't care about udebs. [00:08] I've never looked at priority-mismatches' implementation, but I assume it's vaguely policy-compliant. [00:08] Oh, right. [00:08] Then extra it is. :P [00:09] And it doesn't care about enforcing optional/extra either; all it does is ensure that required/important/standard match up with the expansions of the required/minimal/standard seeds. [00:09] infinity: BTW, you said you have sbuild configured to use a giant tmpfs. Does that work with lvm-snapshot? If so, what's the simplest way to configure it? [00:09] I don't use LVM snapshots, so not sure the best way to do that. [00:09] I assume just mounting a tmpfs on /var/lib/schroot/mount won't work, since schroot overmounts snapshots under those. [00:10] I flip flop between aufs and overlayfs, both of which are just simple directory unions, so I mount /var/lib/schroot/union/overlay as a tmpfs and done. [00:10] If it doesn't something similar for lvm, that's the answer, but I dunno. [00:10] Ah, maybe I could do it in /etc/schroot/whatever/fstab [00:11] Why LVM? Just got sick of random overlayfs/aufs bugs/misfeatures? [00:11] My new server has 16GB RAM and already uses LVM for nearly everything [00:11] Fair enough. [00:11] So it seemed the logical choice [00:11] LVM snapshot removal doesn't seem totally robust, though :-/ [00:11] And yeah, 16G here too. Hence: [00:11] schroot 12G 1.7G 11G 14% /var/lib/schroot/union/overlay [00:11] schroot /var/lib/schroot/union/overlay/ tmpfs size=75% 0 0 [00:12] (It was a glorious day when I learned you could specify tmpfs size by percent) [00:18] I've mounted a tmpfs statically on /var/lib/sbuild/build and configured the schroot fstab to bind-mount that onto /build [00:18] Not perfect since you still hit disk for installing build-deps, but should help [00:20] Yeah, my solution's definitely speed for build-deps, but has the downside of aufs and overlayfs both sucking differently. :P [00:20] s/speed/speedy/ [00:21] * infinity would pay good money to not be deathly ill a day before he has to get on a plane. [00:21] Debian wheezy's kernel does have aufs, at least, so I can fall back to that if this turns out to suck too much [00:21] Pretty sure that not running sbuild on my laptop all the time is going to be FTW though. [00:21] A new laptop would also be FTW. [00:22] I've got a QA surplus one from Pete [00:22] In the process of setting it up [00:22] Oh, yay. No more laptop trying to pretend to be a tablet? [00:22] QA seems to have killed its battery though [00:22] It'll be a while before I get general life transferred, but yeah [00:23] Just don't do what I did. [00:23] I've spent a good part of the tail end of this week on sysadmin ... [00:23] Oh? [00:23] I backed up to a non-redundant disk and wiped mine out. And the backup disk died about 10% into the return rsync. [00:23] Oops [00:23] Yeah, timing++ [00:24] I do need to refresh my secondary backup disk, but I'm not generally too badly off for backups these days [00:24] With the new server I can abolish the ridiculous situation where my laptop was also the house backup server [00:24] Because the old system with 192MB RAM wasn't very good at anything involving, well, work [00:25] Heh. [00:25] An 85x RAM upgrade was pretty much what it needed [00:25] My PPC kit keeps me pretty well covered for servers, though I need to get around to slapping some big/cheap disks in the POWER5. [00:26] Which would be so much easier if the SAS controller in there wasn't the only SAS controller in the world that refuses to talk to SATA drives. :/ [00:26] These days, I really do not have disk shortage problems. Occasionally disk is in the wrong places. [00:26] So, I probably need to pick up a cheap SATA controller and hook it up to the SAS backplane. [00:27] Cause I refuse to pay the ridiculous premium for SAS drives. [00:27] This one's still half-SATA half-PATA while I migrate, but I can cope. It's an improvement. [00:27] A few pvmoves at some point won't be a big deal. [00:49] Aww. Half the fun of fixing NBS rdeps is being able to remove the binaries, and someone beat me to it. [00:49] Sad face. [00:50] Too slow, old man. [00:51] Why, I oughtta. [00:51] I did component-mismatches too. [00:51] Also: get off my lawn. [00:51] * antarus gets a lawnchair [00:52] * infinity does priority-mismatches. [00:52] At least I caught the kernel override thing at the magic mid-copy moment. [00:54] My, this HP laptop has an abysmal UEFI implementation. [00:54] You have to manually select "Boot from EFI file" and navigate your way through the EFI System Partition if you want to boot Ubuntu using UEF. [00:54] *UEFI [00:55] Yeah, apw's HP netbook is about that bad. [00:55] Spec compliance must be for other people. [00:56] is that a win8 certified system? [00:56] It doesn't have Secure Boot, so no. [00:56] (At least not AFAICS.) [00:56] It has a Windows 7 sticker on it. [00:57] as bad as they all are, so far my experience is that Microsoft at least managed to force everyone to have an half decent UEFI implementation for simple things like booting (the configuration tools they ship with are all different and all broken though) [00:57] It's fairly clear from the design and what docs exist that this UEFI implementation is just for experimentation. [00:57] Might haul the thing back to BIOS at some point. But I'm going to use it to track down this GRUB networking bug first. [00:58] I think I've reproduced it, or at least a precursor to it; I can see it sending out ARP packets and ignoring the result. [01:01] cjwatson: cool. Let me know when you have something I can test here. Oh, and on the subject, do you know if grub2 does IPv6? (next one on my list will be IPv6 UEFI PXE-like boot) [01:01] Yes, in 2.00 [01:02] At least the code exists; can't speak to testing [01:02] cool, I'll do some more testing once we have working IPv4. [01:02] Yeah, I should be able to do at least minimal testing as well. [01:02] That's a pretty major item on the UEFI implementation list. [01:03] I'm not even sure what UEFI IPv6 PXE does. My guess is that it involves DHCPv6 to get the tftp server, that or just randomly pokes the gateway and hopes to find a tftp server there. [01:04] MTFTP, I think, but yeah [01:06] * antarus will be putting Ubuntu on his Pixel this weekend [01:13] antarus: join #ubuntu-arm i'm hure hrw and ogra will be drooling over that =) [01:13] s/hure/sure/ [01:14] infinity: yeah, feel free to remove my comment. i guess a few times when images failed to build was due to new abi kernels not being promoted / out-of-sync with seeds, which you and colin discussed now anyway. [01:16] xnox: Remove it? Nonsense. We have a time-honored tradition of sniping in whiteboards. It's one of life's simple pleasures. [01:20] * xnox got level 5 in ingress, not sure what else to do. good night everyone. [01:21] Was that pretty much your last goal in life? [01:21] xnox: not arm [01:21] xnox: its just an i5 [01:21] An ARM system with a display like that would make me a very happy man. [01:22] * antarus avoids ingress [01:22] that thing will ruin my life [01:22] Of course, it would also need a ThinkPad-style trackpoint. [01:22] Which usually limits my options to... ThinkPads. [01:22] infinity: afaik the chromeos team wrote their own driver for the touchpad [01:22] (which is open source) [01:22] or so the guy told me ;p [01:22] antarus: Yes, I've been following davem's battle with it. :) [01:23] antarus: But that's hardly relevant, since I was just pointing out that I had touchpads. [01:23] had? [01:23] hate. [01:23] HATE. [01:23] Typing hard. [01:23] * antarus hands infinity an external mouse [01:23] problem solved! [01:23] Not at all. :P [01:24] The beauty of the trackpoint is never having to move your hands off the keyboard. [01:24] yeah I liked them too [01:24] but no one is making them anymore [01:24] not 'stylish' [01:24] I'll never quite grasp why they didn't become more popular. [01:24] Lenovo still ships with them on most machine, except the very low end. [01:25] you can get USB thinkpad keyboards (with trackpoint), but not that great to have to carry a usb keyboard around with your laptop ;) [01:25] ahh [01:25] the x1 doesn't have them, afaik [01:25] let me go look [01:25] I think I'll just be waiting for a thinkpad with a high res screen. [01:25] The Carbon sure does. [01:25] oh yeah [01:25] I'm wrong [01:26] * antarus doesn't play with the new hardware much ;p [01:26] The Carbon would be dangerously close to the perfect laptop if the screen res was a bit shinier. [01:26] Well, and if I could stuff in tons more RAM, but I always want that, there's never enough. [01:27] yeah the pixel has 4GB soldered on :/ [01:27] my chromebox desktop has 16, heh [01:27] My T420s has 16G. [01:27] And even that's not enough. :P [01:27] you are one of those weirdos that runs like 40 VMs on your laptop aren't you [01:28] No, I do everything in tmpfses. [01:28] Also, web browsers suck. All of them. [01:28] I'm going to tell you something that will blow your mind [01:28] there is this thing [01:28] called the cloud [01:28] it has more cpus and memory than your laptop [01:28] if you do your work there.. ;p [01:28] And irritating round trips, and sometimes fees. [01:29] And so much fiddly crap I don't really care to deal with for one-off builds. [01:29] But if you can figure out a way to offload all my web browsing to the cloud, I'm all ears. [01:30] My auto-cross-builder project made complete sense to do in the cloud. For everything else, when the cloud magics up more bandwidth for me, I might be there. [01:30] At least when Firefox started doing the on-demand tab loading, that helped a ton. [01:30] Oh yes. [01:32] cjwatson: bandwith, or latency? [01:32] (or both)? [01:32] Bandwidth. [01:32] I suppose both, but latency is not really the blocker for anything I care about. [01:32] I'm actually unfamiliar with Canonical, do they have an office, or is everyone remote? [01:32] We have offices, but most developers aren't in one. [01:32] What he said. [01:33] that is what I expected [01:33] yeah i think that would be a tough sell then [01:33] I have tons of bandwidth at home because I live in the promised land, apparently. But I still find the concept of setup/teardown and pushing my sources to the cloud, retrieving binaries and/or logs, lather, rinse, repeat, to be ludicrous for anything other than large rebuilds. [01:33] best home internet I had was 100mbit, and that was in .de in the middle of Munich [01:34] It's not an efficient way to debug and iterate. [01:34] infinity: ahh you work *all* on your laptop them [01:34] then* [01:34] my laptop is just a thin client [01:34] to my 'cloud desktop' as it were [01:34] Hence the server-specced laptop. [01:34] My row of houses is surrounded by a solid ring of areas where there is fibre-to-the-cabinet. But our area: old-fashioned copper ADSL only, and we're 4km from the exchange so 2.5Mbps is the best we can do. [01:35] Which is why I'm afraid things like the Chromebook leave me rather cold. :-P [01:35] yeah [01:35] right now I have like 25/5 business comcast [01:35] so using stuff like NX from home works pretty well [01:35] 250/25 here. [01:35] (Also, our area isn't planned for upgrade because apparently I live near too many poor people.) [01:35] promised land indeed [01:35] cjwatson: I get more than 2.5mb down over my phone ;p [01:35] (I paraphrase only very slightly.) [01:36] My phone's 3G isn't too bad, but they start imposing annoying restrictions at less than my typical monthly limit. [01:36] yeah i had ot give up my unlimited data [01:36] now i just have 2G [01:36] I found this out when I had a three-week ADSL outage last summer. [01:36] cjwatson: I assume your neighbourhood is also expected to run its own telephone exchange, including emergency services? [01:36] cjwatson: And your "ambulances" are just voluneers with wheelbarrows? [01:37] also EC2 is expensive as hell [01:37] clouds need to be cheaper [01:37] infinity: Heh. But no, we're on the Cambridge exchange. It's just our one cabinet that apparently sucks. [01:37] * antarus was speccing out on-demand buildbots on EC2 [01:37] horrible idea [01:38] I literally went and mapped out all the addresses around us with the post office's postcode checker and BT's address checker to figure out what FTTC coverage looked locally. All the surrounding cabinets: fine. Number 83: sucks to be us. [01:38] (Tedious, but BT are allergic to providing data in useful formats.) [01:38] cjwatson: I'm trying to sort out if that's sad or hilarious. [01:39] cjwatson: I'm guessing there are no reasonably-priced LTE providers or anything in the area? Pretty much anything would be better than that. Pringles can repeater to a friend a mile down the road? [01:40] cjwatson: I'm sure we can send you lamont to help replicate his old setup. [01:40] There's a cable monopoly provider. I may buckle at some point. For now, they're evil and I want nothing to do with them (e.g. mandatory intrusive web filtering). [01:40] (He did finally get cable though, so you're even more ghetto than lamont...) [01:40] * antarus makes a note never to move to what I'm assuming is the UK) [01:41] We're a bit behind. Pretty sure my 250-house block is rather worse than average, though. [01:41] antarus: cjwatson's in Cambridge, England, yeah. lamont, who was even more ghetto for ages is in Colorado. [01:41] But we did hear entertaining stories about him scaling towers to fix his internet in the middle of ice storms. [01:41] gives me bad memories of college [01:41] microwave link [01:41] The sad bit is that Cambridge is a geek-packed tech city. [01:42] went down in a storm, so we had a 56k dialup for the remote building (200 users) [01:42] and we had solaris, NIS+, and NFS homedirs [01:42] so about nothing worked [01:42] Ow. [01:42] They 'ran' on that 56k for two weeks [01:42] Yeah, I was going to comment that 56k wasn't so bad when I was in college, but not for an NIS/NFS network. [01:43] It was great at job-2 when the NFS server went down (once every couple of months) and it took IT several hours to recover it so we all got to go to the pub. [01:43] Though, while I would have put 200 users on a 64k ISDN link, my university certainly never was that bad off. [01:43] Pretty sure they had redundant T3s or better by the time I was there. [01:44] Externally. Internally was shinier. [01:44] * xnox ponders if monopolised kingston communications internets in hull was worse than connection in cambridge. [01:45] cjwatson: that sounds like Google Circa 2002 ;p [01:45] it was something like 1 Mbps. [01:45] and that's 3 years ago. [04:23] Everytime I fly back from a family visit in Switzerland it takes me a couple of weeks to get used to my slow 30Mbps/10Mbps internet again... my grand-mother is the one with the slowest internet and she has 50Mbps/25Mbps for a quarter of what I pay here... (my parents currently have 100Mbps/50Mbps for half the price I pay) [04:24] at least I'm good at mirroring stuff, so I very rarely grab big stuff directly from the internet (letting me get unlimited internet for a small premium was the worst decision my ISP ever made, currently averaging 1.5TB a month) === jetole_ is now known as jetole [20:00] stgraber: =)