<@conan_kudo:matrix.org>
17:00:29
!startmeeting CentOS Hyperscale SIG
<@meetbot:fedora.im>
17:00:32
Meeting started at 2024-11-20 17:00:29 UTC
<@meetbot:fedora.im>
17:00:32
The Meeting name is 'CentOS Hyperscale SIG'
<@conan_kudo:matrix.org>
17:00:41
!topic Roll call
<@salimma:fedora.im>
17:00:45
!hi
<@zodbot:fedora.im>
17:00:46
Michel Lind (salimma) - he / him / his
<@conan_kudo:matrix.org>
17:00:47
!hi
<@zodbot:fedora.im>
17:00:49
Neal Gompa (ngompa) - he / him / his
<@conan_kudo:matrix.org>
17:03:44
I'll wait a couple more minutes since I just pinged everyone now
<@rcallicotte:fedora.im>
17:03:56
!hi
<@zodbot:fedora.im>
17:03:57
Robby Callicotte (rcallicotte) - he / him / his
<@salimma:fedora.im>
17:04:09
we have people! yay
<@rcallicotte:fedora.im>
17:04:09
howdy!
<@rcolebaugh:matrix.org>
17:05:14
!hi
<@zodbot:fedora.im>
17:05:16
Raymond Colebaugh (rcolebaugh) - he / him / his
<@conan_kudo:matrix.org>
17:07:27
okay I think that's it, so let's move onto...
<@conan_kudo:matrix.org>
17:07:32
!topic Followups
<@salimma:fedora.im>
17:07:52
we have this perennial calendar issue
<@salimma:fedora.im>
17:08:12
because the current CentOS calendar requires PRs to update and does not support meetings pinned to local time
<@salimma:fedora.im>
17:08:44
I'm trying to sort it out over at CentOS Promo, but seems there's some sort of consensus to eventually use the Fedora calendar somehow (which makes sense since we use Fedora meeting rooms already)
<@salimma:fedora.im>
17:08:49
should we just jump there first?
<@conan_kudo:matrix.org>
17:08:53
yeah
<@conan_kudo:matrix.org>
17:09:20
I think it's just a fedora-infrastructure ticket to get a calendar created
<@salimma:fedora.im>
17:09:32
open source has both problems - people who always wait to get consensus and people who just do their own things :P
<@rcallicotte:fedora.im>
17:09:33
seems like it makes most sense to use fed cal
<@salimma:fedora.im>
17:09:48
if it's not infra I'm sure they'll tell us where to re-file
<@salimma:fedora.im>
17:10:08
do we need to vote on this? or only if someone objects
<@conan_kudo:matrix.org>
17:10:22
only if someone objects
<@salimma:fedora.im>
17:10:52
let's wait a couple of mins, then I have one more follow-up
<@conan_kudo:matrix.org>
17:11:26
do we want to ask for a centos-sig-hyperscale calendar or a global centos calendar?
<@conan_kudo:matrix.org>
17:11:33
I'm more inclined to suggest we do the former
<@salimma:fedora.im>
17:11:40
I think the SIG one
<@salimma:fedora.im>
17:12:31
Shaun is thinking about how to get a unified view, but there's probably an iCal aggregator that can be used for that. plus that means we can restrict admin rights to mess with this calendar to just us
<@conan_kudo:matrix.org>
17:12:59
right
<@conan_kudo:matrix.org>
17:14:10
!info The CentOS calendar has been problematic, leading to people missing meetings due to bugs and missing features.
<@conan_kudo:matrix.org>
17:14:31
!agreed CentOS Hyperscale SIG will move its calendar to calendar.fedoraproject.org
<@conan_kudo:matrix.org>
17:14:42
Michel Lind 🎩 UTC-6: do you want to take care of this task?
<@salimma:fedora.im>
17:15:01
yup, I'll do this
<@conan_kudo:matrix.org>
17:15:34
!action Michel Lind 🎩 UTC-6 will move the CentOS Hyperscale SIG calendar to calendar.fedoraproject.org under its own top-level calendar
<@salimma:fedora.im>
17:16:03
ok, next thing
<@salimma:fedora.im>
17:16:30
Davide and I talked to the perf maintainer about the feasibility of packaging Perf standalone back in LPC two months ago
<@conan_kudo:matrix.org>
17:17:17
oh?
<@salimma:fedora.im>
17:17:17
he's supportive and said he takes patches if we want to make the packaging more modular etc.
<@salimma:fedora.im>
17:17:17
Been busy and there's family stuff so I did not have time for this last month, but I'm close to getting the latest upstream tarball (which is sadly 6.9.0) built and we have a git.centos.org repo now
<@salimma:fedora.im>
17:17:17
<@conan_kudo:matrix.org>
17:17:40
perf wasn't the problematic one generally... bpftool is
<@conan_kudo:matrix.org>
17:17:47
it has its own versioning and the kernel spec honors this
<@conan_kudo:matrix.org>
17:18:09
but splitting perf out has some other benefits
<@conan_kudo:matrix.org>
17:18:15
we should try to get this in place in fedora too
<@salimma:fedora.im>
17:18:18
<@salimma:fedora.im>
17:18:18
- I can make head or tail of the weird way this thing consumes build flags. Currently I just chose to not override CXXFLAGS and LDFLAGS because otherwise the build fails
<@salimma:fedora.im>
17:18:18
- we can get 6.12 built (I tried generating the tarball myself and some files are missing IIRC, for 6.11, so there's some things we need to fix)
<@salimma:fedora.im>
17:18:18
I think I'll keep this in experimental for now until
<@salimma:fedora.im>
17:18:47
but yeah standalone packaging makes it much easier to iterate. e.g. enabling demangling support is trivial and this builds in a few minutes
<@salimma:fedora.im>
17:18:53
yes
<@salimma:fedora.im>
17:18:56
sorry, meant to get to that
<@salimma:fedora.im>
17:19:12
once we promote this out of experimental we should have a Fedora Change Proposal to switch Fedora over too
<@salimma:fedora.im>
17:19:35
I'll ask jforbes on the side if he's around today and invite him to co-own the CP / comaintain
<@salimma:fedora.im>
17:20:00
it was not problematic to build. it's problematic if you want a newer perf which Meta often does
<@conan_kudo:matrix.org>
17:20:04
he seems to be around in #kernel:fedoraproject.org
<@jforbes:fedora.im>
17:20:17
I am around, looking at context
<@salimma:fedora.im>
17:20:22
or if you want added features (i have a months-old request to turn on cxa demangle that I utterly failed at when perf was built in kernel-tools)
<@salimma:fedora.im>
17:20:30
ohai! :)
<@conan_kudo:matrix.org>
17:20:32
bpftool is a problem because koji's NVRA uniqueness check has been failing builds
<@salimma:fedora.im>
17:20:44
you must be watching mentions of your username or something :)
<@conan_kudo:matrix.org>
17:21:18
for the 6.12 rebase, I'll be mangling the release to include a suffix with the kernel version
<@jforbes:fedora.im>
17:21:27
I do get notifications
<@conan_kudo:matrix.org>
17:21:29
for the bpftool package
<@salimma:fedora.im>
17:21:30
perf also has funny versioning
<@salimma:fedora.im>
17:22:11
the soname is 0.0.1 and the python module is 0.1 ... and the Java binding has an invalid name according to rpmlint :P
<@salimma:fedora.im>
17:22:11
<@salimma:fedora.im>
17:22:11
but once we get 'rpmlint parity' with the Fedora package I think it's good to come out of experimental
<@jforbes:fedora.im>
17:22:14
So bpftool is supposed to be a separate package. The RHEL maintainer was working on that, and was supposed to let me know when it was approved in Fedora so that I could merge his MR to turn it off in the kenrel build
<@salimma:fedora.im>
17:22:22
also upstream seems to build with -O6 by default which is... huh
<@conan_kudo:matrix.org>
17:22:39
yeah, the MR needs rebasing: https://gitlab.com/cki-project/kernel-ark/-/merge_requests/3338
<@conan_kudo:matrix.org>
17:22:52
oh crap
<@conan_kudo:matrix.org>
17:22:57
I need to review the package 😂
<@conan_kudo:matrix.org>
17:23:11
I'll do that today
<@jforbes:fedora.im>
17:23:37
So yeah, it's your fault :)
<@salimma:fedora.im>
17:23:41
so jforbes would you like to co-own a change proposal to split off perf in Fedora? maybe targeting F42 or F43 depending on how fast I can get this done
<@jforbes:fedora.im>
17:23:42
haha
<@salimma:fedora.im>
17:24:08
this should be a self-contained change, right? it should not affect end users
<@jforbes:fedora.im>
17:24:13
I do not see the advantage to splitting off perf. bpftool is being maintained in a different way
<@jforbes:fedora.im>
17:24:31
While the perf maintainers are big on maintaining it inside the kernel.
<@jforbes:fedora.im>
17:24:59
did you speak to acme about this at LPC?
<@salimma:fedora.im>
17:25:05
yes
<@salimma:fedora.im>
17:25:14
he seems keen on people using the standalone tarballs
<@salimma:fedora.im>
17:26:05
<@salimma:fedora.im>
17:26:05
b) stuck in experimental because we can't sign it yet (so perf is also stuck there)
<@salimma:fedora.im>
17:26:05
a) sometimes complicated to build
<@salimma:fedora.im>
17:26:05
the pain point for us - at least in Hyperscale - is that we often need the latest perf, but the whole kernel itself is
<@jforbes:fedora.im>
17:26:06
Let me confer with acme about it, but as long as he is in favor, I am happy to do so.
<@jforbes:fedora.im>
17:26:45
Well hyperscale is all about the CS kernel right?
<@salimma:fedora.im>
17:26:52
it does not matter so much in Fedora, except that turning on extra features seems difficult (I have not tried since it moved back from kernel-tools to kernel, but I spent a day trying to turn on cxa-demangle and because of the funny way perf consumes build flags, that was super tricky)
<@salimma:fedora.im>
17:27:06
sounds good! and no rush
<@salimma:fedora.im>
17:27:13
not just though
<@salimma:fedora.im>
17:27:22
the biggest HS consumer, Meta, runs its own kernel internally
<@salimma:fedora.im>
17:27:31
but we get a lot of userspace tooling (including perf) from HS
<@conan_kudo:matrix.org>
17:27:37
Nope. We use the fedora-based kernel now.
<@salimma:fedora.im>
17:27:49
oh right, I forgot about that too
<@jforbes:fedora.im>
17:27:55
Cool
<@jforbes:fedora.im>
17:28:18
So yeah, will talk to acme, but I am okay with it.
<@salimma:fedora.im>
17:28:51
awesome! thanks. I'll probably have some patches I'll send him sometime this week too
<@salimma:fedora.im>
17:29:10
anyone else has a follow-up?
<@jforbes:fedora.im>
17:29:17
Honestly, will make it easier to coordinate.with the dynamic linking too
<@conan_kudo:matrix.org>
17:29:24
I just posted a review comment for bpftool: https://bugzilla.redhat.com/show_bug.cgi?id=2310159#c3
<@conan_kudo:matrix.org>
17:29:34
fairly straightforward fixes and then it can be approved
<@jforbes:fedora.im>
17:31:14
I would almost rather it not be as a Fedora version change though. Once it is split out, we should do so on all supported releases.
<@jforbes:fedora.im>
17:31:28
Otherwise I have to do some wonky patch stuff
<@conan_kudo:matrix.org>
17:31:52
didn't we do a Change for kernel-tools even though it affected all releases?
<@salimma:fedora.im>
17:31:58
yeah, I think it's fine backporting. I'm thinking of the Python 3.9 change proposal a few releases back
<@jforbes:fedora.im>
17:32:02
We did not
<@salimma:fedora.im>
17:32:11
the proposal basically says "we will push this to stable releases too"
<@jforbes:fedora.im>
17:32:47
Yeah, I am not against the actual change process, just as long as we can do it on all supported releases
<@conan_kudo:matrix.org>
17:33:06
we could just do a simple devel-announce thing
<@salimma:fedora.im>
17:33:13
<@salimma:fedora.im>
17:33:13
if we want to do it more quietly that's fine as well. but it might be good for comms to have it listed
<@salimma:fedora.im>
17:33:13
agreed, I think people expect the Fedora kernel changes to be backported so this counts as that too
<@conan_kudo:matrix.org>
17:33:18
there's no actual impact to anyone for the split
<@salimma:fedora.im>
17:33:19
that works
<@jforbes:fedora.im>
17:33:34
correct, it only impacts the maintainers.
<@salimma:fedora.im>
17:34:41
shall we move on to the next topic?
<@jforbes:fedora.im>
17:34:48
go ahead
<@salimma:fedora.im>
17:34:50
oh... has anyone been working on openssh
<@conan_kudo:matrix.org>
17:35:32
jforbes: let the kernel person know about bpftool internally plz :)
<@conan_kudo:matrix.org>
17:35:38
so that we can process that quickly
<@rcolebaugh:matrix.org>
17:36:14
Yes Vishal has been working on the new openssh build for c10
<@salimma:fedora.im>
17:37:30
hopefully it's almost ready
<@rcolebaugh:matrix.org>
17:37:40
He's merged from upstream and has been sorting out merge conflicts. Uploading the new lookaside sources worked out just fine btw! Right now he's verifying which patches were legit removed upstream. Hopefully we'll be set to start testing internally soon
<@salimma:fedora.im>
17:37:44
stream 10 and epel 10 are slated for official reveal on... Dec 3, I think is the latest estimate
<@conan_kudo:matrix.org>
17:37:55
yeah
<@conan_kudo:matrix.org>
17:38:21
it doesn't matter much for us other than hopefully we can have hyperscale 10 images ready by then
<@conan_kudo:matrix.org>
17:38:40
then shaunm can put a screenshot of hyperscale 10 kde for the new website
<@salimma:fedora.im>
17:39:48
you don't need openssh for the hyperscale image though right Conan Kudo
<@salimma:fedora.im>
17:40:13
that's more on the meta side - we ideally want to make a splash internally and announce something workable for people to test
<@conan_kudo:matrix.org>
17:41:01
yeah
<@salimma:fedora.im>
17:41:02
any other followups?
<@conan_kudo:matrix.org>
17:41:28
we don't ship openssh in the standard repos right?
<@conan_kudo:matrix.org>
17:41:35
it's only in the -fb tags
<@salimma:fedora.im>
17:41:44
yup
<@salimma:fedora.im>
17:42:20
some of the changes might make sense for a public build, but those need to be sorted out
<@salimma:fedora.im>
17:42:31
and ideally the good parts all get upstreamed anyway and everyone gets it
<@conan_kudo:matrix.org>
17:43:47
right
<@conan_kudo:matrix.org>
17:45:16
I don't think we have any other followups?
<@salimma:fedora.im>
17:45:58
what's next... announcement?
<@salimma:fedora.im>
17:46:23
if we have not mentioned it we should get the Connect CFP on the meeting records as a reminder
<@conan_kudo:matrix.org>
17:46:31
so we can move onto announcements
<@conan_kudo:matrix.org>
17:46:38
!topic Announcements
<@conan_kudo:matrix.org>
17:47:28
!info CentOS Connect 2025 CFP is open until December 16, 2024
<@conan_kudo:matrix.org>
17:47:31
!link https://cfp.fedoraproject.org/centos-connect-2025/cfp
<@salimma:fedora.im>
17:47:56
!info deadline is Dec 16
<@salimma:fedora.im>
17:48:13
woops, I have not submitted anything
<@conan_kudo:matrix.org>
17:48:21
!info The CentOS Showcase talk Davide Cavalca and I did on Hyperscale is now published: https://www.youtube.com/watch?v=-nHy8pf0AYE
<@conan_kudo:matrix.org>
17:48:25
!link https://www.youtube.com/watch?v=-nHy8pf0AYE
<@conan_kudo:matrix.org>
17:49:02
!info The CentOS Showcase talk Michel Lind 🎩 UTC-6 did on using Hyperscale to develop improvements for Fedora is published.
<@conan_kudo:matrix.org>
17:49:06
!link https://www.youtube.com/watch?v=sCvawNYcfgA
<@conan_kudo:matrix.org>
17:50:52
Any other announcements?
<@conan_kudo:matrix.org>
17:52:36
okay then...
<@conan_kudo:matrix.org>
17:53:16
!topic Tickets
<@conan_kudo:matrix.org>
17:53:24
I don't think we have anything here, do we?
<@conan_kudo:matrix.org>
17:53:45
nope, nothing new... though we definitely need to do some spring cleaning
<@salimma:fedora.im>
17:53:59
speaking of tickets I am pretty sure the update notification tickets are dead
<@conan_kudo:matrix.org>
17:54:23
yeah
<@salimma:fedora.im>
17:55:22
if anyone wants to brave openshift :
<@salimma:fedora.im>
17:56:01
oh and remember the Hyperscale hangout is today
<@salimma:fedora.im>
17:56:17
is it 5 PM ET / 3 PM PT or did I get the time wrong again?
<@conan_kudo:matrix.org>
17:56:31
that's correct
<@conan_kudo:matrix.org>
17:56:44
my personal calendar item for it says it's at that time
<@salimma:fedora.im>
17:57:51
alright, we should wrap this up and jump to the EPEL meeting in... 3 mins
<@conan_kudo:matrix.org>
17:58:25
yep
<@conan_kudo:matrix.org>
17:58:36
so...
<@conan_kudo:matrix.org>
17:58:37
!topic Membership
<@conan_kudo:matrix.org>
17:58:43
Nothing for this, we have no new requests
<@conan_kudo:matrix.org>
17:58:50
so...
<@conan_kudo:matrix.org>
17:58:52
!topic Miscellaneous
<@conan_kudo:matrix.org>
17:59:04
Anything anyone wants to bring up in... one minute?
<@rcolebaugh:matrix.org>
17:59:20
nothing else on my end :)
<@conan_kudo:matrix.org>
17:59:26
alrighty then
<@conan_kudo:matrix.org>
17:59:31
3...
<@conan_kudo:matrix.org>
17:59:33
2..
<@conan_kudo:matrix.org>
17:59:34
1.
<@conan_kudo:matrix.org>
17:59:37
!endmeeting