| *** hwoarang has quit IRC | 01:10 | |
| *** hwoarang has joined #openstack-dib | 01:11 | |
| *** hwoarang has quit IRC | 03:50 | |
| *** hwoarang has joined #openstack-dib | 03:51 | |
| prometheanfire | johnsom: have you looked at the package-installs 'when' support? | 03:53 |
|---|---|---|
| *** hwoarang has quit IRC | 04:25 | |
| *** hwoarang has joined #openstack-dib | 04:25 | |
| johnsom | prometheanfire: yes, in the docs, not the code. It is boolean decisions from what I saw. So messy with all of the kernel options, but possible. Frankly I see three options, I proposed one to start the conversation. | 04:54 |
| johnsom | Ubuntu, for OpenStack has three useful options today, so boolean decision trees seemed messy. | 04:57 |
| prometheanfire | ya, they can be | 04:58 |
| prometheanfire | I feel like you could set three bools though | 04:58 |
| prometheanfire | and have 3 whens for each package | 04:58 |
| johnsom | I also didn’t want to go down the uninstall path of doom | 04:58 |
| johnsom | Yeah, for sure it would work | 04:58 |
| johnsom | Then you need to know the magic environment variables, etc. | 04:59 |
| prometheanfire | they should be doc'd :P | 04:59 |
| johnsom | Agreed | 04:59 |
| prometheanfire | cat likes miracle whip | 05:00 |
| johnsom | Lol | 05:00 |
| johnsom | With jello? | 05:00 |
| prometheanfire | nah, berries, just cleaning the bowl | 05:01 |
| prometheanfire | he has a sweet tooth I think | 05:01 |
| johnsom | If that is the consensus I can make it happen. I just want the 200MB back.... | 05:02 |
| prometheanfire | sure, that part I understand :D | 05:02 |
| johnsom | Breaking it down into element options seems valid too. We just need to make a decision that the other distros are ok with | 05:03 |
| prometheanfire | ya, I'd like a standard way, personally I see all the multiple debian and ubuntu elements as overkill | 05:04 |
| johnsom | Ok, so your vote is environment variables? | 05:05 |
| prometheanfire | or something else to keep it in one element | 05:05 |
| johnsom | Curious what ianw vote would be. | 05:05 |
| * johnsom wonders if he is online yet | 05:05 | |
| johnsom | prometheanfire: Thanks for the feedback. The PoC tests based on this patch all based perfectly, so I am happy with the result. | 05:08 |
| johnsom | Ugh, mobile, based->passed | 05:09 |
| *** irclogbot_2 has quit IRC | 05:30 | |
| *** irclogbot_0 has joined #openstack-dib | 05:31 | |
| ianw | johnsom: in .au, so online :) but i start super early on wednesdays to catch infra meeting | 05:44 |
| ianw | 200mb seems like a lot, are we bringing in debug symbols or something by default? | 05:46 |
| johnsom | ianw: it is all the extra kernel modules that bulk it up with the generic kernel. | 06:33 |
| ianw | hmm, we shouldn't need the -extras package though? i just kind of went through this, because the kafs modules are in there, which we wanted on production servers | 06:34 |
| ianw | and the extras package wasn't in there, so if it's coming in, maybe we can stop that? | 06:34 |
| johnsom | All of the linux-image-* packages are meta-packages, each bring in different modules. Generic is the most, followed by virtual, and then KVM. | 06:36 |
| johnsom | In the case of octavia, we don’t need the sound drivers as no one will hear the Amphora scream. Lol | 06:37 |
| johnsom | I am all about the *choice* of kernel package, it is really about how do we want to handle this across the distros. | 06:38 |
| ianw | right, but i'm talking about specifically the package like linux-modules-extra-4.15.0-22-generic | 06:38 |
| johnsom | elements, env variables, etc. | 06:38 |
| johnsom | Those only come in if the meta package depends on them or we call them out | 06:39 |
| ianw | like, per the comment, i'm not really seeing where the 200mb comes from between generic and kvm, they look not that far apart | 06:40 |
| johnsom | linux-image-generic pulls in the kitchen sink and bathroom tub | 06:41 |
| ianw | but i guess that's my point, i don't think it *has* to ... the upstream images seem to have -generic without -extra for example ... so it might be worth confirming we're not doing stuff we don't need to | 06:42 |
| ianw | certainly no dbg symbols etc should be coming in | 06:42 |
| johnsom | Here is the difference in qcow2 size: | 06:43 |
| johnsom | https://www.irccloud.com/pastebin/tSbHbF4L/ | 06:43 |
| johnsom | So comparing the old kernel to the new, size: 605397504 vs 393244160 according to glance | 06:44 |
| ianw | i'm not arguing it isn't bigger :) just that maybe we're doing something wrong if it makes such a difference | 06:44 |
| johnsom | If you look at the package dependencies, generic: https://packages.ubuntu.com/bionic-updates/linux-image-generic | 06:45 |
| johnsom | kvm: https://packages.ubuntu.com/bionic-updates/linux-image-kvm | 06:45 |
| johnsom | This is where the bulk of the difference comes from | 06:45 |
| johnsom | I dug in the images, the kernel modules are a killer on generic | 06:46 |
| ianw | right, ok, so the -generic image *does* depend on the -extra package. the kvm one doesn't | 06:48 |
| johnsom | Yes, it is a different kernel config as well to my understanding. It's not pulling in the base generic kernel package. | 06:49 |
| johnsom | Really my question comes down to "how" we want to implement kernel selection. | 06:49 |
| johnsom | I threw up a proposal to test my PoC and start the conversation. | 06:49 |
| johnsom | I see three options: | 06:50 |
| johnsom | 1. per-kernel elements. Likely some duplication. | 06:50 |
| ianw | there is probably a strong argument we should have been installing the kvm kernel all along | 06:50 |
| johnsom | 2. Environment variables that specify the kernel. | 06:50 |
| johnsom | 3. A hybrid of 1, where we layer more abstractions down and make ubuntu-minimal-kvm a child of ubuntu-minimal, which is a child of ubuntu-common. | 06:51 |
| johnsom | Agreed, but choice is good too. | 06:51 |
| johnsom | I am happy to do the leg work on any of those, I just want a consensus across the distros so we don't all do it differently. | 06:52 |
| ianw | right, well i haven't thought about it that much, but my thought is that a model like i pseudo layed out in review comment would work. have a variable to choose, but default | 06:52 |
| johnsom | I guess 4 would be change the behavior of package-install to be smarter than boolean | 06:53 |
| ianw | i don't think we really want to get that into the parsing game | 06:53 |
| ianw | before you know it, it's turing complete | 06:53 |
| johnsom | Yeah, that was my first thought as well. The boolean mess made me second guess it. | 06:53 |
| johnsom | lol O(n) | 06:53 |
| ianw | i don't think it's too much mess with a bunch of when:'s and set the default in an environment.d file | 06:53 |
| ianw | i don't think we need to think too strongly about complete cross distro (gentoo/fedora/centos/opensuse) behaviour here; they're all too different in the way they configure things like the kernel | 06:54 |
| ianw | for mine, a variable to choose the ubuntu kernel type specifically is fine | 06:54 |
| johnsom | Yeah, so my "vision" is environment.d that looks for DIB_KERNEL and then sets booleans that package-install can handle. | 06:54 |
| johnsom | Default would be generic | 06:55 |
| ianw | right, i think i'd just go for DIB_UBUNTU_KERNEL though and constrain the problem | 06:55 |
| ianw | and honestly, defaulting to kvm is probably correct ... i'm guessing it just didn't exist as a package when we started | 06:55 |
| johnsom | Well, we most likely have the same problem with the other distros. Shouldn't we do something "common"? | 06:55 |
| ianw | that can always be a follow on, anyway. i'd probably switch the infra images to it, after now having looked :) | 06:56 |
| johnsom | Yes, trusty didn't have it. | 06:56 |
| johnsom | It went from "virtual" to "virtual" == generic to "generic" and "virtual" and "kvm". lol | 06:57 |
| johnsom | Moving target for sure. | 06:57 |
| ianw | ok, yeah we can't quite kill trusty yet :/ | 06:57 |
| johnsom | They have 5-6 other kernels too, but I'm not sure they matter to OpenStack. | 06:57 |
| johnsom | Are you sure we should do distro specific variables? I will do the patch in the morning, but I have an itch that we should do a generic variable that other distros could adopt as well. | 06:59 |
| ianw | i think my preference is for distro specific. we've gotten ourselves tied up before trying to do low-level things like this in generic ways, then like with trusty etc distros change kernel patterns or packages or something, and now you have to paper over not just that but other distros etc etc | 07:01 |
| ianw | it's a pretty specific thing, choosing the exact package name for the kernel, that's why i'd suggest that | 07:01 |
| ianw | but, you know, just a suggestion. if you look at other distros and think it just fits better, i'm all ears :) | 07:01 |
| johnsom | Ok, fair enough. I will add DIB_UBUNTU_KERNEL and have environment.d parse it out so package-install can handle it. In the morning as it just rolled over midnight here and I have a 6:30am meeting. | 07:03 |
| ianw | go to bed! :) thanks | 07:03 |
| johnsom | Thanks for the feedback! | 07:03 |
| *** ivve has joined #openstack-dib | 07:07 | |
| openstackgerrit | Merged openstack/diskimage-builder master: Sync Sphinx requirement https://review.opendev.org/666023 | 07:48 |
| openstackgerrit | jacky06 proposed openstack/diskimage-builder master: Update hacking version to latest https://review.opendev.org/628466 | 07:52 |
| openstackgerrit | jacky06 proposed openstack/diskimage-builder master: Update hacking version to latest https://review.opendev.org/628466 | 08:18 |
| *** chhagarw has joined #openstack-dib | 09:14 | |
| *** yolanda has quit IRC | 09:41 | |
| *** priteau has joined #openstack-dib | 10:46 | |
| *** chhagarw has quit IRC | 10:50 | |
| *** yolanda has joined #openstack-dib | 11:17 | |
| *** chhagarw has joined #openstack-dib | 12:18 | |
| *** priteau has quit IRC | 12:27 | |
| *** yolanda has quit IRC | 13:13 | |
| *** yolanda has joined #openstack-dib | 13:17 | |
| *** ivve has quit IRC | 14:23 | |
| *** Vorrtex has joined #openstack-dib | 15:12 | |
| *** ivve has joined #openstack-dib | 18:50 | |
| *** chhagarw has quit IRC | 18:50 | |
| *** ivve has quit IRC | 19:16 | |
| *** Vorrtex has quit IRC | 20:10 | |
| *** hwoarang has quit IRC | 20:16 | |
| *** hwoarang has joined #openstack-dib | 20:17 | |
| openstackgerrit | Michael Johnson proposed openstack/diskimage-builder master: Add DIB_UBUNTU_KERNEL to ubuntu-minimal https://review.opendev.org/666063 | 21:09 |
| openstackgerrit | Michael Johnson proposed openstack/diskimage-builder master: Add DIB_UBUNTU_KERNEL to ubuntu-minimal https://review.opendev.org/666063 | 21:11 |
| openstackgerrit | Michael Johnson proposed openstack/diskimage-builder master: Add DIB_UBUNTU_KERNEL to ubuntu-minimal https://review.opendev.org/666063 | 21:12 |
| openstackgerrit | Michael Johnson proposed openstack/diskimage-builder master: Add DIB_UBUNTU_KERNEL to ubuntu-minimal https://review.opendev.org/666063 | 22:10 |
| openstackgerrit | Michael Johnson proposed openstack/diskimage-builder master: Add DIB_UBUNTU_KERNEL to ubuntu-minimal https://review.opendev.org/666063 | 23:39 |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!