| *** fungi is now known as Guest9249 | 01:33 | |
| *** kinrui is now known as fungi | 01:41 | |
| opendevreview | Karolina Kula proposed openstack/diskimage-builder master: WIP: Add support for CentOS Stream 10 https://review.opendev.org/c/openstack/diskimage-builder/+/934045 | 08:09 |
|---|---|---|
| opendevreview | Karolina Kula proposed openstack/diskimage-builder master: WIP: Add support for CentOS Stream 10 https://review.opendev.org/c/openstack/diskimage-builder/+/934045 | 09:51 |
| opendevreview | Karolina Kula proposed openstack/diskimage-builder master: WIP: Add support for CentOS Stream 10 https://review.opendev.org/c/openstack/diskimage-builder/+/934045 | 11:17 |
| opendevreview | Joel Capitao proposed openstack/diskimage-builder master: WIP: Add support for CentOS Stream 10 https://review.opendev.org/c/openstack/diskimage-builder/+/934045 | 14:16 |
| opendevreview | Karolina Kula proposed openstack/diskimage-builder master: WIP: Add support for CentOS Stream 10 https://review.opendev.org/c/openstack/diskimage-builder/+/934045 | 14:33 |
| opendevreview | Joel Capitao proposed openstack/diskimage-builder master: WIP: Add support for CentOS Stream 10 https://review.opendev.org/c/openstack/diskimage-builder/+/934045 | 16:12 |
| dtantsur | Hi folks! WDYT about using `dnf mark install` on all packages from package-installs.yaml? We have a large problem in Ironic because some of our packages get auto-removed. | 17:24 |
| JayF | https://opendev.org/openstack/diskimage-builder/commit/fd8fae299e8de8a4765dfb030ea94dcefd55b00c | 17:25 |
| dtantsur | aheeemmmm | 17:25 |
| JayF | that was recently landed, and I think caused this regression | 17:25 |
| dtantsur | ianw: please revert your revert | 17:25 |
| JayF | I'm posting it now | 17:25 |
| dtantsur | Thanks and good catch! | 17:26 |
| clarkb | are you using dib from master and not releases? | 17:27 |
| opendevreview | Jay Faulkner proposed openstack/diskimage-builder master: Reapply "Make sure dnf won't autoremove packages that we explicitly installed" https://review.opendev.org/c/openstack/diskimage-builder/+/934992 | 17:27 |
| clarkb | just asking because that merged more recently than the latest release | 17:27 |
| JayF | this is in ipa-builder jobs | 17:27 |
| dtantsur | clarkb: I'm quite sure we do | 17:27 |
| JayF | which I think uses master | 17:27 |
| JayF | explicitly to frontrun these kinda failures | 17:27 |
| dtantsur | An example for anyone curious: if we remove cloud-init (which we don't need on IPA ramdisks), dnf removes gdisk which IPA needs. | 17:28 |
| clarkb | also why does dnf installing a package not mark it as installed manually by default? this seems like a dnf bug (which ianw's original commit aludes to I guess) | 17:28 |
| clarkb | dtantsur: do you explicitly list gdisk as a package you need? I would expect that would cause dnf to mark it as explicitly installed even if it was already installed via a dependency chain elsewhere? | 17:29 |
| dtantsur | clarkb: dnf does not do that IIRC | 17:29 |
| dtantsur | (you could argue it should..) | 17:29 |
| clarkb | ya that does sound like a package manager bug to me | 17:30 |
| dtantsur | and yes, it's in our package-installs.yaml | 17:30 |
| * dtantsur needs to leave now, leaving the conversation in the hands of JayF | 17:30 | |
| JayF | I'm not convinced there's not an IPA-builder bug here too, but I'd prefer us research that not under the gun of CI failures | 17:30 |
| dtantsur | Not just CI failures: we produce IPA images that do not work. | 17:31 |
| JayF | Yeah, if the other cores here aren't willing to land this revert, in the meantime we can pin the version in ironic CI. | 17:40 |
| JayF | But I suspect we wouldn't be the only people impacted by this change in behavior, even if it is actually the package manager's bug | 17:40 |
| TheJulia | kind of digesting everything right now, it is almost like we need to re-trigger re-installs after any package removals | 17:42 |
| clarkb | I've gone ahead an approved it. I agree any improvements can happen as followups | 17:43 |
| JayF | Thanks. I feel like this is some sort of weird dnf behavior issue, because when you look at the commands that are executed in the order they're executed in out the result doesn't make sense at all | 17:45 |
| TheJulia | The fact the base source image they based the revert on lacks history is also sort of problematic because then the self packaging history/providence data is more on image trust | 17:45 |
| TheJulia | dunno, that also seems... wrong-ish | 17:45 |
| TheJulia | so there are multiple sort of weird cases which need to be navigated here | 17:45 |
| TheJulia | clarkb: ++ thanks! | 17:46 |
| ianw | we could probably drop the fedora 22 bit of that | 20:18 |
| ianw | i wonder if it has something to do with the way we create the initial chroot? | 20:21 |
| ianw | i guess it somewhat makes sense, in that we're not giving dnf the full packages info. if i was completely starting this again, a better way to do it might to be for package-installs to make a dependency-only RPM and install that ... a bit like the way builddeps and mock works | 21:57 |
Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!