blackboxswSRU uploads queued for 23.3 into Focal, Jammy and Lunar -proposed. Thanks falcojr for the PRs and uploads there. We will probably start an SRU verification this week then00:03
blackboxswdbungert: as a heads up, this is the cloud-init SRU that we know will require a livecd-rootfs set of changes to drop the After=snapd.seeded.service systemd dependency ordering for Jammy00:04
dbungertblackboxsw: ack, I started the livecd-rootfs side today00:04
blackboxswfalcojr is going to manage the SRU release for 23.3. I'll assist w/ subiquity-specific testing where needed.00:04
blackboxswppa:cloud-init-dev/proposed has cloud-init 23.3 uploaded and built for focal, jammy, lunar and mantic if that helps anyone for testing.01:10
senzillaHello 👋 Any OpenBSD people here? 😊 I have run into some issues related to dhcp and network bring-up on OpenBSD16:59
senzillaI chatted to minimal yesterday, and I will open an issue on GitHub tomorrow and describe the issues in more detail17:00
minimalmeena: ping!17:02
senzillameena: Thanks for all the FreeBSD improvements to cloud-init! I think it's brought OpenBSD a lot closer as well! 🙂17:14
meenasenzilla: hi17:59
meenaminimal has relayed your issues, but i haven't had much time to do stuff today18:00
meenakiddo is back to school on Friday… 18:00
meenaSo, essentially, we need to split this into a base class, and one for each divergent BSD18:03
senzillameena: Right. We might also need a new Dhcp class. IscDhclient might not cut it for OpenBSD since it's based on dhclient which is being phased out in OpenBSD. It would be better to use dhcpleasectl instead18:15
inflatadorminimal thanks for the heads-up on that18:16
minimalsenzilla: dhcpcd and udhcpc support was added recently as part of catering for dhclient EOL18:16
senzillameena: dhclient is still in OpenBSD base, but it's not working well with assumptions in cloud-init (pid files and lease files are not compatible I think)18:16
senzillaYesterday, I started stubbing out a new Dhcp class for dhcpleased/dhcpleasectl . I won't find much time in the next couple of days, but I think it'll be pretty straight forward18:18
minimalsenzilla: is that IPv4 only?18:20
senzillaYeah, unfortunately I think so18:22
senzillaI think the OpenBSD distro class could set preferred dhcp class priority based on ipv6 setting18:37
meenasenzilla: how well would dhcplease(d|ctl) work for the net_ops class?18:41
* meena is trying to update her dev VM to 14.0 ALPHA18:42
senzillameena: I need to dive into the code base a bit more to answer that. Don't have it in front of me right now18:48
senzillameena: and, good luck with the VM update :)18:51
meenanot netops, netops should work.18:51
meenaalready ran out of memory twice on compiling LLVM.18:51
senzillaYes, I think netops will work. But I don't think IscDhclient will work18:51
holmanbsenzilla: hi o/18:52
senzillaI think we need a new class, something like Dhcpleased(DhcpClient) which works entirely with dhcpleased and dhcpleasectl18:53
senzillaholmanb: 👋18:53
holmanbsenzilla: thanks for digging into this, your help on this is welcome and appreciated :D18:53
holmanbsenzilla: sounds like you're on the right track18:53
senzillaI'm not super18:54
senzilla...familiar with python18:55
meenasenzilla: do you have any sysvinit files for OpenBSD you can contribute?18:55
senzillais it only the @classmethod methods that count as the actual "public" Dhcp class interface?18:55
meenasenzilla: we'll happily help… but, i wasn't super familiar with Python before starting to work on cloud-init (and iocage)18:56
holmanbsenzilla: fyi, the Udhcpc class is incomplete and won't fully function on azure (and maybe others, would need to check), but it's a really good starting place to get a dhcp_discovery() implimented for whichever class is needed for OpenBSD18:56
holmanbsenzilla: no classmethod just means that the method can be called with the class without instansiating an object instance18:57
senzillameena: I do not have any sysvinit files. I actually think the best way to integrate cloud-init is to patch /etc/rc. I don't like patching base, but I think it makes sense here. Because OpenBSD have intentionally hard coded a lot of the init and service startup procedures. And getting cloud-init inserted early enough would require patching /etc/rc18:58
meenai didn't realize18:59
meenabut i think goneri does it similarly18:59
senzillaAlternatively, a cleanly packaged version could create a couple of sysvinit files. But cloud-init will only come in right at the end after all base services18:59
senzillameena: goneri is writing the cloud-init boot stages to /etc/rc.local which is only just before user defined services. Still way after base services start up19:00
senzillameena: Yeah, I have studied goneri's work a fair bit19:01
meenaon FreeBSD we're very early. basically, after network is up (including ntpd, i think) and after the critical part of the filesystems is mounted.19:02
senzillaYeah, one have to patch /etc/rc to get inserted that early in OpenBSD. The hard coded and fully predictable start up order in BSD is great for robustness, predictability and security guarantees19:03
senzillaIt's less good for integration :P19:03
meena*nod* *nod*19:04
senzilla But hey, rc is placed in /etc... It's free for all right :P19:05
meenaare there any other ports that modify /etc/rc? 19:09
senzillaDon't think so. I think the port should *not* patch /etc/rc . I doubt that would get accepted. It's probably best to leave that as a documented manual intervention19:16
senzillaI have got a set of Packer templates for building cloud-init OpenBSD VMs. Similar to what goneri is doing.19:17
senzillaI haven't reached out to goneri yet, but it would make sense to collaborate with him on a standard pre-patched cloud-init OpenBSD image19:19

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!