Christian, * Christian Hofstaedtler ([email protected]) wrote: > * Stephen Frost <[email protected]> [151006 04:12]: > > That said, I do think it's worthwhile to see about fixing these > > particular install failures, and the proposed change looks like it would > > at least do that. > > I agree that it would make the situation better for some users. > > I was going to write which users are affected by this bug, but now > that I think more about it, I'm not sure which users are affected. > As you have some interest, I assume you have run into the bug? How > did your configuration look like?
Yes, I did run into this bug while upgrading the hidden PowerDNS master for postgresql.org. The configuration we have is relatively simple. My recollection of how it happened is: Installed an older version (3.4.1?), modified /etc/powerdns/pdns.conf with a few simple changes- allow-axfr-ips, master=yes, slave=yes, etc. Then at some point down the road, attempted to upgrade to 3.4.6 (in jessie backports), and it blew up on that error. To fix it, I recall modifying pdns.conf *and* pdns.conf.ucf-dist. It's possible I didn't have to change both, but I think I tried changing pdns.conf first and it didn't help, so I changed pdns.conf.ucf-dist also and that got me past the issue. Apologies for not having more detailed notes, I was a bit anxious to get it fixed. :) I'll see about building a test jessie VM, installing the version in jessie, modifying the config, and then doing an upgrade, and see if I can reproduce the error. > > > > Is there anything I can do to help? > > > > > > I'm thinking of deleting most of the code in the postinst for > > > stretch. > > > > Are you thinking about simply assuming that /etc/powerdns/pdns.d is the > > PDNSDIR ... > > Yes, because the files the postinst touches are meant to be the > files that the previous version of the pdns-server package has > shipped. Right. > > and anything else is up to the user to address? > > If the user has moved or renamed the pdns.d dir, or changed the > include= dir to point to something else entirely, then we have no > business of touching (and moving!) the users config files at all. Agreed. We just need to be able to sanely detect that and act accordingly. > As the postinst doesn't have a version check right now, it > 1) is wrong for versions after jessie, > 2) we have to look at all previously released binary packages to see > the original intent and which conffiles have previously been installed. > I haven't done that check yet. Yeah, that doesn't sound like much fun. :) > > > Not sure what to do about jessie. Given that this bug has existed > > > since 2006, maybe it's not terribly important to fix in jessie. > > > > I disagree. Perhaps I'm being naive, but having the relatively simple > > case, where /etc/powerdns/pdns.d is the directory and the configuration > > has been only mildly tweaked, failure during upgrades is not a good > > position for us to be in. > > > > I have to admit that I'm not up to speed on current policy, but I'm > > happy to try and implement whatever the correct solution is. I'm sure > > there are other packages which have include directories, is there a > > clear "right way" to handle this? > > I don't think there's much policy here except for the normal "don't > touch stuff that isn't yours" - i.e. preserve user changes, > especially if they are to/in files that aren't installed by the > package. Right. > There's some other complication - include= is the old name of the > include directory setting; jessie's pdns does include-dir= instead. [1] > > > Thanks! > > Thank you for caring about this, Absolutely, I like PDNS in general and given that we're using it to run postgresql.org, we really want it to work well. :) I'm also a DD, btw, though I haven't done much Debian-related work recently. I'd be happy to help out with maintaining PowerDNS though, as we're using it. Thanks! Stephen
signature.asc
Description: Digital signature
