<@nirik:matrix.scrye.com>
16:05:21
!startmeeting riscv-sig
<@meetbot:fedora.im>
16:05:24
Meeting started at 2025-11-11 16:05:21 UTC
<@meetbot:fedora.im>
16:05:24
The Meeting name is 'riscv-sig'
<@davidlt:matrix.org>
16:05:27
!hi
<@nhanlon:beeper.com>
16:05:27
!hi
<@zodbot:fedora.im>
16:05:28
Neil Hanlon (neil) - he / him / his
<@nirik:matrix.scrye.com>
16:05:29
hey folks.
<@zodbot:fedora.im>
16:05:44
David Abdurachmanov (davidlt)
<@davidlt:matrix.org>
16:06:28
Should I give a quick update of F43 progress?
<@nirik:matrix.scrye.com>
16:06:36
yeah, that would be good...
<@davidlt:matrix.org>
16:07:30
According to my old scripts we have 3835 older NVRs and 532 missing (might be not available for riscv64, never had a build in general, FTBFS, etc.).
<@davidlt:matrix.org>
16:07:58
There are recent larger updates that landed (QT5, QT6 + Plasma) which will block some other packages until finished rebuilding.
<@nirik:matrix.scrye.com>
16:08:23
the qt6/plasma stack is so big. ;(
<@davidlt:matrix.org>
16:08:35
I pinged Mark for `debugedit` issue, and got initial response, but nothing more since then.
<@davidlt:matrix.org>
16:09:18
`catatonit` and OpenJDK 22 and 23 has hit by the same issue related to linker, compiler or flags, or a combo.
<@davidlt:matrix.org>
16:09:32
New GCC that landed recently didn't change anything.
<@davidlt:matrix.org>
16:09:57
Not having OpenJDK 25 (current LTS) of course blocks some packages.
<@davidlt:matrix.org>
16:10:16
I might end up bootstraping everything on F42 and tagging into F43 to unblock.
<@davidlt:matrix.org>
16:10:37
I should investigate the issue with toolchain (I suspect linker), but that would take quite some time.
<@davidlt:matrix.org>
16:11:03
We had a couple of downtimes recently, but upstream Koji instance was affected too.
<@nirik:matrix.scrye.com>
16:11:04
yeah, the way they do java on primary is weird. :)
<@nirik:matrix.scrye.com>
16:11:27
yeah, sorry about that. Trying to get stuff all working.
<@davidlt:matrix.org>
16:11:44
kashyapc has been learning and helping to build packages too.
<@davidlt:matrix.org>
16:12:20
Two bigger unfinished items to figure out would be forgejo sig stuff and compose things.
<@davidlt:matrix.org>
16:13:00
I believe my brain had removed almost everything related to Pungi and composes from my LLC cache in the brain by now 😄
<@davidlt:matrix.org>
16:13:48
I used today's downtime to do quite a lot of firmware (U-Boot flashing), and similar to various boards. Somehow that kills motivation.
<@nirik:matrix.scrye.com>
16:14:10
yeah, hardware is... so slow and painful. always
<@nirik:matrix.scrye.com>
16:14:14
on pungi...
<@nirik:matrix.scrye.com>
16:14:48
I think we had some p550's that were earmarked all for centos sigs. I wonder if we could look at getting 1 for composes in fedora and use that...
<@nirik:matrix.scrye.com>
16:15:08
we could also try mock forcearch
<@davidlt:matrix.org>
16:15:47
Alternative would be to do VM, but I think you mentioned that RHEL 9 or 10 (RHEL in general) can only do KVM.
<@nirik:matrix.scrye.com>
16:16:36
well, we can / already have a compose-x86-riscv01 thats a fedora vm on a rhel hypervisor
<@nirik:matrix.scrye.com>
16:16:43
so mock forcearch should work there
<@nirik:matrix.scrye.com>
16:16:47
since it's fedora
<@davidlt:matrix.org>
16:17:04
but Koji will not use forcearch.
<@nirik:matrix.scrye.com>
16:17:52
I think it might be able to, since it's trasparent. If you tell koji 'do a riscv build on this builder' and the builder is x86, it might just work...
<@davidlt:matrix.org>
16:17:55
Correction, it might.
<@davidlt:matrix.org>
16:19:39
Yeah, so kojid can check extra config for forcearch
<@nirik:matrix.scrye.com>
16:19:53
worth a shot I think, might not work
<@nirik:matrix.scrye.com>
16:19:57
or might be too slow
<@nirik:matrix.scrye.com>
16:20:09
but for just runroot tasks for composes... might be ok
<@nirik:matrix.scrye.com>
16:20:33
anyhow, I think thats all I had
<@davidlt:matrix.org>
16:20:38
From my current understanding there might be 0 runroot tasks as we aren't enabling the targets that trigger those. I am not 100% sure.
<@davidlt:matrix.org>
16:21:57
What worries me is that forcearch is probably a tag thing.
<@nirik:matrix.scrye.com>
16:22:50
I think it's supposed to be transparent... ie, if you run mock telling it to do a riscv build on a x86 machine it just transparently uses qemu-static...
<@davidlt:matrix.org>
16:22:56
In that case it would also applies on native host (riscv64).
<@nirik:matrix.scrye.com>
16:23:02
but I actually haven't tested it, so I could be totally wrong
<@davidlt:matrix.org>
16:23:10
Not sure how mock reacts to that.
<@davidlt:matrix.org>
16:23:40
From kojid point of view it doesn't check anything, just passes that to mock.
<@nirik:matrix.scrye.com>
16:26:32
yeah
<@davidlt:matrix.org>
16:27:29
Also bootstrap is native.
<@davidlt:matrix.org>
16:28:22
I mean bootstrap chroot step (if enabled).
<@davidlt:matrix.org>
16:28:44
That would be x86_64 from what I can see.
<@davidlt:matrix.org>
16:29:12
That depends on `use_bootstrap`. I think, that's disabled right now.
<@nirik:matrix.scrye.com>
16:29:56
by default mock uses containers for that, but it can make bootstrap chroots which is I think what we do in primary
<@davidlt:matrix.org>
16:29:58
Booting a VM would sound easier 😄 This requires more code reading & testing locally to figure out exactly how this work.
<@davidlt:matrix.org>
16:30:10
native and non-native, as forcearch would be set for both.
<@nirik:matrix.scrye.com>
16:30:42
a riscv vm on x86 hypervisor? sadly tho we run into the rhel9/10 cannot do that problem. ;(
<@nirik:matrix.scrye.com>
16:30:55
I guess I could put fedora on the hypervisor...
<@davidlt:matrix.org>
16:32:21
So the problem with `use_bootstrap` would be that we have no x86_64 repos.
<@davidlt:matrix.org>
16:33:08
It's basically a tiny chroot with DNF to be used to create chroot, and that needs to be host native.
<@davidlt:matrix.org>
16:33:50
https://pagure.io/fedora-infra/ansible/blob/main/f/roles/koji_builder/templates/builders/site-defaults.cfg#_8
<@davidlt:matrix.org>
16:34:05
So that basically means it needs x86_64 repos too.
<@nirik:matrix.scrye.com>
16:34:21
no? shouldn't it use riscv repos/rpms to make that?
<@davidlt:matrix.org>
16:34:39
No, because the 1st chroot is a small one container DNF to be used to create riscv64 chroot.
<@davidlt:matrix.org>
16:34:49
No, because the 1st chroot is a small one containing DNF to be used to create riscv64 chroot.
<@davidlt:matrix.org>
16:35:19
That's because you don't want to use old DNF (or YUM) from the host.
<@nirik:matrix.scrye.com>
16:35:25
yes, but it must also be riscv
<@nirik:matrix.scrye.com>
16:35:37
it's just to get around the versioning issues I thought?
<@davidlt:matrix.org>
16:35:40
No, because that runs natively.
<@nirik:matrix.scrye.com>
16:35:56
oh, I see what you are saying, yeah...
<@nirik:matrix.scrye.com>
16:36:10
it uses the host stack to make that bootstrap, so if thats x86_64...
<@nirik:matrix.scrye.com>
16:36:22
but the forcearch stuff gets around that somehow?
<@davidlt:matrix.org>
16:36:38
1st you get a new DNF (native). Use that native DNF to create riscv64 chroot. Then you are in QEMU static + riscv64 chroot.
<@nirik:matrix.scrye.com>
16:36:47
anyhow, I think this will take some testing/looking more.
<@nirik:matrix.scrye.com>
16:36:53
yeah, sounds right.
<@nirik:matrix.scrye.com>
16:37:12
so indeed a vm sounds much simplier.
<@davidlt:matrix.org>
16:37:15
> Enforce host-native repo architecture for bootstrap chroot
<@davidlt:matrix.org>
16:37:20
from code
<@davidlt:matrix.org>
16:37:46
There is `bootstrap_forcearch=True` but comment says it should never be used 😄
<@nirik:matrix.scrye.com>
16:37:58
ha, nice
<@davidlt:matrix.org>
16:38:00
Of course, Koji doesn't know anything about it too.
<@davidlt:matrix.org>
16:38:22
VMs would be easier 😄
<@davidlt:matrix.org>
16:38:37
We can think more. This forced my brain to wake up a bit 😆
<@nirik:matrix.scrye.com>
16:38:40
I guess I can look at redoing the hypervisor with fedora so it can run a emulated vm
<@nirik:matrix.scrye.com>
16:38:52
not sure when I would get to it tho. things are... crazy
<@davidlt:matrix.org>
16:39:13
There is virt-main COPR SIG that provides Fedora Rawhide libvirt/qemu stack too, but probably Fedora hypervisor would be better.
<@davidlt:matrix.org>
16:39:28
Yeah, seems infra gets broken every few days.
<@davidlt:matrix.org>
16:39:41
Composes would be nice to have, but not a top priority.
<@nirik:matrix.scrye.com>
16:39:46
yeah, sorry... we are trying to get things back to normal
<@davidlt:matrix.org>
16:39:59
Could you set maxjobs=512 on both VMs, and restart kojid?
<@nirik:matrix.scrye.com>
16:40:19
anyhow, I need to go feed cats, they are meowing at me. ;)
<@davidlt:matrix.org>
16:40:32
I am now pushing queues up to 1K and I need more capacity to eat those build, waitrepo, newRepo, etc. tasks.
<@davidlt:matrix.org>
16:41:15
Once I hit the current limit of 256 it becomes hard to handle it with priorities alone.
<@nirik:matrix.scrye.com>
16:42:33
oh yes, adding to todo
<@nirik:matrix.scrye.com>
16:42:52
shall we head back to the main channel?
<@davidlt:matrix.org>
16:43:45
Yes
<@nirik:matrix.scrye.com>
16:44:31
pushing that change now
<@nirik:matrix.scrye.com>
16:44:36
Thanks everyone
<@nirik:matrix.scrye.com>
16:44:38
!endmeeting