On Mon, Apr 04, 2011 at 09:36:33AM +0200, Philipp Kern wrote: > On Sun, Apr 03, 2011 at 11:34:20PM +0200, Andreas Barth wrote: > > Except the two mipsel buildds here - well, who cares anyways? > > I generated keys for all buildds on my laptop back then. But then other > people are also approaching me with "hey, I think something broke, can you > have a look" instead of insulting me. > > > I would be happy with going on with 0.60.0 unless there are issues why > > we need to use 0.61.0 - and I think I said so in the IRC channel more > > than once. That would have saved us as well the issues that we have > > with lenny hosts plus buildd-0.61, and also the user env not being > > defined anymore. Stability is some value in running a core service. > > True enough. There was also a solution for the Lenny hosts that could've > been incorporated into the package? I wonder why that didn't happen. I'm still not entirely sure what happened here. But I don't admin any buildds myself, so I can't really diagnose what went wrong without further information from a buildd admin. One change made on the master branch is that the script-defaults.buildd file for schroot script-config=script-defaults.buildd in the configuration was replaced by a buildd/config (all in /etc/schroot) with the configuration requiring a change to script-config=buildd/config. mount-defaults.buildd becomes buildd/fstab and likewise for the other files referenced by script-defaults.buildd. This matches the changes made in the layout of the schroot configuration profiles made for schroot 1.4; the main improvement here (other than tidiness) is that because the files are generated from templates, we gain support for freebsd and other platforms rather than hardcoding linux-specific stuff in the configuration. It's important to stress that this change should not have broken anything. We were adding new set of configuration files, intended to supercede the old ones when the admin updated their schroot configuration. But because we have not removed or changed any of the /existing/ configuration, it shouldn't have broken any buildds. It looks like the main issue on the buildds was that the copyfiles-defaults and nssdatabases-defaults files referenced by script-defaults.buildd were no longer present. This is the part that caused the breakage (no passwd/shadow/group updates meant that the build user wasn't present in the chroot, causing lookup problems), and it's not clear to me why this occurred, especially on lenny systems. On squeeze chroots the old conffiles are replaced by symlinks to the new locations in the schroot pre/postinst after moving the conffiles to the new names. This keeps older configurations working transparently. But lenny chroots should still have the old conffiles. As I understand, it's a possibility that we should have provided some compatibility symlinks in the buildd package postinst to keep existing buildds working. However, this was not done, because there should never have been a need for them. TTBOMK. We still need to pin down the root cause of why the old conffiles disappeared. If they hadn't the problem would never have occurred in the first place. Could it be down to puppet? Does this introduce any differences to conffile handling on upgrades on buildds? The buildd package may have dropped the files--but they are conffiles and so should not have been removed AFAIK. If this did cause them to be removed, we'll need to add some compatibility symlinks. Regards, Roger -- .''`. Roger Leigh : :' : Debian GNU/Linux http://people.debian.org/~rleigh/ `. `' Printing on GNU/Linux? http://gutenprint.sourceforge.net/ `- GPG Public Key: 0x25BFB848 Please GPG sign your mail.
Attachment:
signature.asc
Description: Digital signature