Bug#390545: integrate vserver-build.fai
package: fai version: 3.0 Hi, a dedicated build method for fai-enabled vservers now exists in vservers-svn: http://svn.linux-vserver.org/viewvc.py/util-vserver/trunk/scripts/vserver-build.fai?view=markup and will hopefully released in util-vserver .211 The needed changes in fai have been send to the mailinglist but not to the BTS, they are here: http://www.mail-archive.com/linux-fai%40rrz.uni-koeln.de/msg03756.html Currently the build method in vserver has the following problems: - obsolete old path for FAI_CONFIGDIR and FAI_NFSROOT as default - relying on sams branch, doesnt work with fai 3.0 (I have no idea whats exactly needed to make it work...) - relying on the fai nfsroot to exist. I very often deploy vservers with fai without using a nfsroot and just using softupdate. So this is kind of a regression. - possible more ;) Sam, I think you could greatly help if you could describe what changes are _really_ needed in fai to make vserver-build.fai work... regards, Holger Log from the discussion on #vserver, which made me notice this: daniel_hozac mugwump: i've committed the fai to the svn now with a few fixes (check for nfsroot, removing the directories in killprocs, probably something else i forgot). -- phreak`` has quit (Quit: leaving) -- phreak`` ([EMAIL PROTECTED]) has joined #vserver h01ger daniel_hozac, whats fai in this context? daniel_hozac h01ger: Debian's fully automatic installation, or whatever it's called :) --- _[PUPPETS]Gonzo is now known as [PUPPETS]Gonzo h01ger daniel_hozac, interesting. i'm an active fai developer :) so what code did you commit to vservers (?) svn? h01ger and mugwump used to hang around in #fai too :) h01ger and i maintain my vservers with fai :) meandtheshell h01ger: http://svn.linux-vserver.org/viewvc.py/util-vserver/ :) h01ger daniel_hozac, and where there exactly? which path? daniel_hozac it's a fai build method. daniel_hozac trunk/scripts/vserver-build.fai daniel_hozac https://savannah.nongnu.org/patch/?5058 daniel_hozac h01ger: opinions appreciated :) h01ger FAI_NFSROOT and FAI_CONFIGDIR use hardcoded and deprecated paths (?) h01ger mugwump, daniel_hozac: have you tested this with fai 3.0, released sometime last week? h01ger ah, nor hardcoded daniel_hozac i haven't tested it at all. h01ger each time i see /usr/local in FAI_NFSROOT||_CONFIGDIR I get a heart-attack. i reported this stupid bug in 2003 and it was fixed last month. after three years of nagging. daniel_hozac hehe. h01ger thats why first screamed and then read further. they are at least not hardcoded... h01ger also this looks suspicious to me: h01ger echo *** Looks like you don't have the patched FAI. h01ger echo *** get it from svn://svn.debian.org/svn/fai/people/mugwump/vserver daniel_hozac how so? h01ger cause it should work with the unpatched fai?! daniel_hozac should it? would you be able to try? * daniel_hozac doesn't have any Debian hosts available to him. daniel_hozac i have no idea what mugwump's fai changes from vanilla. * h01ger neither. but i just check, the test fails with fai_3.0 h01ger checked daniel_hozac right, i suppose that's the idea. daniel_hozac but if that was commented, would it work? daniel_hozac it being the build method. h01ger i have no idea, as i dont know what feature mugwump added h01ger also i'm unhappy about the hardcoded tmpfs mounting for the vserver - i usually disable this for my vservers (as 16mb /tmp is way to little) daniel_hozac 19:58 h01ger cause it should work with the unpatched fai?! daniel_hozac that is just temporary for the build though. daniel_hozac surely a 64 MiB /tmp should suffice for building a guest? h01ger daniel_hozac, yes. but its hard to say if it will work, if i dont know what changes mugwump has in his branch. h01ger daniel_hozac, for building thats fine. didnt see that h01ger mugwump, i also dislike that this needs a setup fai nfsroot. i heavily use fai softupdate and i often dont even have a nfsroot... daniel_hozac h01ger: so what would you change? :) h01ger daniel_hozac, so this build method will show up in vserver-utils one day? daniel_hozac one day very soon. daniel_hozac i'd really want to release 0.30.211 next week. h01ger daniel_hozac, i dont know why its needed. atm i'm looking into replacing the custom fai-live-cd build scripts (which also rely on the nfsroot) with the (new) live-package from debian, which is a generic infrastructure to build live cds. and i promised to look at the munin-vserver plugins today. so i wont look much further into vserver-build.fai ;-) daniel_hozac h01ger: ok, so how would you build guests with fai? h01ger with the debootstrap method. then i just run fai softupdate and be done h01ger i have a script for this, which prepares the vserver a tiny bit after building, so i agree, a build-method makes sense. but the current implementation looks like quite a hack to me :-/ daniel_hozac i do kind of agree :) h01ger and mugwump also only send a mail to the fai list
Bug#390545: integrate vserver-build.fai
Hi Holger, Please note that this new feature does not interfere with your approach of using the debootstrap build method and then running softupdate. Perhaps we can suggest these two methods as alternatives for building vservers in the documentation. Far from being a regression, each approach has its advantages and disadvantages; the approach of using the nfsroot (which, of course, becomes a somewhat figurative term) makes the build a single logical step, and ensures consistency if the intent is to prototype real host builds using vservers. A single FAI vserver can be used to serve network installs as well as installs on the host system, and does not require that FAI be installed on the host. It also allows cross-architecture builds on systems that support it, like i386 builds on amd64 hosts. As for the default paths being obsolete, I am currently working on making this usable for the current stable release of debian before I move on to making it work with testing. I based it off FAI 2.10.1; and plan to re-base it off the latest 2.x release. Please bear in mind that I have been familiarizing myself with FAI along the way, and that now we have a working system (and the projects we are using this for - the .nz registry system and NZ Electoral Enrolment Centre offices - are in deployment phase), we are in a position to spend time to take our improvements and share them. Perhaps in the meantime you might have suggestions for detecting which version of FAI is installed, and selecting more appropriate default paths so that users won't have to override the package defaults. Holger Levsen wrote: package: fai version: 3.0 Hi, a dedicated build method for fai-enabled vservers now exists in vservers-svn: http://svn.linux-vserver.org/viewvc.py/util-vserver/trunk/scripts/vserver-build.fai?view=markup and will hopefully released in util-vserver .211 The needed changes in fai have been send to the mailinglist but not to the BTS, they are here: http://www.mail-archive.com/linux-fai%40rrz.uni-koeln.de/msg03756.html Currently the build method in vserver has the following problems: - obsolete old path for FAI_CONFIGDIR and FAI_NFSROOT as default - relying on sams branch, doesnt work with fai 3.0 (I have no idea whats exactly needed to make it work...) - relying on the fai nfsroot to exist. I very often deploy vservers with fai without using a nfsroot and just using softupdate. So this is kind of a regression. - possible more ;) Sam, I think you could greatly help if you could describe what changes are _really_ needed in fai to make vserver-build.fai work... regards, Holger Log from the discussion on #vserver, which made me notice this: -- Sam Vilain, Systems Architect, Catalyst IT (NZ) Ltd. phone: +64 4 499 2267PGP ID: 0x66B25843 -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Bug#390545: integrate vserver-build.fai
Hi Sam, On Sunday 01 October 2006 22:31, you wrote: Please note that this new feature does not interfere with your approach of using the debootstrap build method and then running softupdate. Sure. I would just be happy to throw away my approach and use yours :-) Perhaps we can suggest these two methods as alternatives for building vservers in the documentation. I think this (under the title setting up vservers) is more confusing and has no benifits. My approach is just a logical extension of softupdates, which should be covered anyway. Your approach should be the offical way to setup vservers. Far from being a regression, each approach has its advantages and disadvantages; the approach of using the nfsroot (which, of course, becomes a somewhat figurative term) makes the build a single logical step, and ensures consistency if the intent is to prototype real host builds using vservers. An nfsroot is just debootstrap plus some extra packages defined in fai.conf. So the consistency can be easily reproduced without relying on a actual existing nfsroot. I also fail to see why an nfsroot is needed for i386 builds on amd64. As for the default paths being obsolete, I am currently working on I hope you read my comment on #vserver about why I'm so upset if I see those pathes: I insisted three years to make FAI behave in a FHS compatible way and after this long time I really freak out, if I see /usr/local/share/fai - it makes my eyes bleed and my brain cry ;-) Nothing personal :) making this usable for the current stable release of debian before I move on to making it work with testing. I based it off FAI 2.10.1; and plan to re-base it off the latest 2.x release. Please bear in mind that I have been familiarizing myself with FAI along the way, and that now we have a working system (and the projects we are using this for - the .nz registry system and NZ Electoral Enrolment Centre offices - are in deployment phase), we are in a position to spend time to take our improvements and share them. Cool. Perhaps in the meantime you might have suggestions for detecting which version of FAI is installed dpkg -l fai or dpkg -l fai-client if using the split packages (which are default since etch.) regards, Holger pgp94sELDMhH5.pgp Description: PGP signature