| JayF | Should we disable building new images when a branch transitions to unmaintained? That feels like the correct thing to do, especially in context of something like IPA, which now we might not be able to build all the old platforms reliably anymore | 02:16 |
|---|---|---|
| TheJulia | Likely a good idea, tbh | 02:26 |
| opendevreview | Adam McArthur proposed openstack/ironic master: api: Add schema for portgroups API https://review.opendev.org/c/openstack/ironic/+/959881 | 04:54 |
| opendevreview | Adam McArthur proposed openstack/ironic master: api: Add schema for volume_target/_connector API https://review.opendev.org/c/openstack/ironic/+/993284 | 04:54 |
| opendevreview | Jacob Anders proposed openstack/ironic-specs master: Add spec for multi-component batched firmware updates https://review.opendev.org/c/openstack/ironic-specs/+/992208 | 04:56 |
| *** jroll01 is now known as jroll0 | 07:23 | |
| opendevreview | Jacob Anders proposed openstack/ironic-specs master: Add spec for multi-component batched firmware updates https://review.opendev.org/c/openstack/ironic-specs/+/992208 | 09:24 |
| opendevreview | Jacob Anders proposed openstack/ironic-specs master: Add spec for multi-component batched firmware updates https://review.opendev.org/c/openstack/ironic-specs/+/992208 | 10:24 |
| opendevreview | Jacob Anders proposed openstack/ironic-specs master: Add spec for multi-component batched firmware updates https://review.opendev.org/c/openstack/ironic-specs/+/992208 | 10:26 |
| janders | TheJulia JayF iurygregory as a follow-up to the mid-cycle I updated ^, when you have a moment I would appreciate feedback. Also CC cardoe | 10:39 |
| janders | I dropped deferred reboot, Multipart and Targets but added a much-requested feature downstream - batching BIOS settings changes (seems solid based on my lab testing, which I will do more of soon) | 10:46 |
| opendevreview | Esther Domfeh proposed openstack/ironic master: Add request context user tracking to node history records https://review.opendev.org/c/openstack/ironic/+/993901 | 11:45 |
| iurygregory | good morning ironic | 12:08 |
| iurygregory | janders, ack | 12:08 |
| opendevreview | David Nwosu proposed openstack/ironic master: Add project field to node history https://review.opendev.org/c/openstack/ironic/+/983349 | 13:59 |
| opendevreview | Pierre Crégut proposed openstack/ironic master: Conditional check of remaining pending attributes https://review.opendev.org/c/openstack/ironic/+/993770 | 14:07 |
| TheJulia | Question, and I'm sure someone is pinging folks who work on openshift metal3, what if we had a way for IPA to ignore the BMC link local addresses? | 14:18 |
| JayF | I don't have any direct objection to it... but it's the wrong fix for the person having that issue. | 14:28 |
| JayF | If you can't get your BMC NIC to not appear to the OS, that's not hardware that's going to be a good experience with Ironic. | 14:28 |
| JayF | Most hardware I've seen configured this way can have their behavior changed | 14:29 |
| TheJulia | Well, the issue in part, is if you get even just one host enrolled into ironic, you can't get the additional hosts enrolled | 14:29 |
| TheJulia | Because lookups then fail for any hardware becasue it looks like a certian vendor I once worked for, stamps the same MAC potentially. | 14:30 |
| JayF | lol | 14:30 |
| JayF | I don't understand why the answer isn't "turn around these standards-breaking hardware at the loading dock" | 14:31 |
| JayF | that's going to cause WAY more problems than just Ironic port creation | 14:31 |
| JayF | Each BMC gets it's own L3 subnet just so they aren't all arp'ing for each other? | 14:32 |
| * JayF afk for a couple hours | 14:32 | |
| TheJulia | the tl;dr is an ethernet port, likely presented via the USB bus for that BMC L3 segment. I've told folks in the past with stuff like this to explicitly turn off the interface, but if they don't know its there.... | 14:37 |
| rpittau | TheJulia: what if we add a validation to IPA with a clear error message? | 14:45 |
| TheJulia | Well, I'm sort of thinking maybe a thing to do, if someone has a library with known EUI/OUIs to just use that to prune down the list | 14:46 |
| TheJulia | and then if no MACs exist, maybe surface an error as a result | 14:46 |
| TheJulia | and make it configurable?!? | 14:46 |
| TheJulia | netaddr has an integrated EUI validator which will spit out the OUI, I wonder if it includes an allocation database | 14:48 |
| TheJulia | err, no releases in 2 years | 14:50 |
| TheJulia | but it does have a DB | 14:51 |
| TheJulia | so good news, the OUI the host sends is not in IEEE's list | 15:23 |
| TheJulia | bad news their server doesn't indicate it's date and macaddress and netaddr's cached copy seems to be the most up to date otherwise | 15:24 |
| TheJulia | Maybe an exclude list? | 15:24 |
| rpittau | yeah, that should work | 15:26 |
| rpittau | (sorry for the slow reply, meeting downstream) | 15:27 |
| TheJulia | no worries, on the redfish forum call at the moment myself | 15:29 |
| TheJulia | so, starting with FA is apparently a signal that "this is locally done" | 15:43 |
| TheJulia | which means non-standard, I think IPA should just ignore interfaces starting as such | 15:43 |
| TheJulia | I think there is an IPA change an an ironic change here | 15:45 |
| TheJulia | jfyi, I keep forgetting that tomorrow is a holiday in the states | 15:58 |
| rpittau | TheJulia: yes, it looks like it (IPA + ironic change) | 15:58 |
| rpittau | IPA should drop these interfaces during discovery | 15:58 |
| rpittau | I did a quick research, we should think about excluding all the FC00::/7 block which are Unique Local Addresses | 15:58 |
| rpittau | and FE80::/10 addresses (starting with FE) are Link-Local Addresses, which are completely non-routable off the local wire | 15:58 |
| TheJulia | I wouldn't really do it via the IP address, and I'm not sure we can in this case because we have no insight into the reporters addressing | 15:59 |
| TheJulia | that being said, I've got claude figuring out a mac based filter | 15:59 |
| TheJulia | I think the kernel also guesses FC00 on no slaac | 16:00 |
| rpittau | ok, that makes sense | 16:02 |
| rpittau | I guess we have predictable patterns for most of vendor or standard local addresses | 16:02 |
| TheJulia | yeah, the purpose for the recording of interfaces is to record what seem like valid physical links addresses can be allocated to | 16:03 |
| JayF | TheJulia: the idea of "ignore things with this mac" as a global + node-level-override config (THINK ABOUT THE SECURITY) makes a lot of sense and is generally applicable | 16:03 |
| JayF | TheJulia: if such a config took wildcards/globs, it would replace a patch we run downstream | 16:03 |
| TheJulia | I wouldn't bother with node level, since honestly, thats too late | 16:03 |
| TheJulia | like, waaaay too late | 16:03 |
| JayF | for your use case :) | 16:03 |
| JayF | Usually ours are shaped more like "don't inventory anything that's not a bone standard ethernet interface" (excluding fc, things exposed as DPU) | 16:04 |
| TheJulia | but the FA interface flag does seem to be standardized and at least one lookup tool suggested it was a "administrator assigned address" | 16:04 |
| TheJulia | Oh, thats a good point, I do wonder what the PF's mac ends up being on the otherside of the pci bridge with a dpu | 16:04 |
| TheJulia | oooh, claude is going to do a superhero style landing through auto-including a release note | 16:06 |
| TheJulia | it has been chewing on my problem statement and ideas for a while and went into plan mode | 16:07 |
| TheJulia | https://bugs.launchpad.net/ironic/+bug/2150450 should be ready for it's embargoed notification reveal today, backports posted, draft ossa also posted to it as well. | 16:11 |
| TheJulia | For https://bugs.launchpad.net/ironic/+bug/2150458 I can do backports today if we're comfortable with it as-is. | 16:11 |
| JayF | TheJulia: ECHAN | 16:11 |
| TheJulia | shoot | 16:11 |
| TheJulia | DOH | 16:11 |
| opendevreview | Esther Domfeh proposed openstack/ironic master: feat: populate node_history state fields https://review.opendev.org/c/openstack/ironic/+/993208 | 16:52 |
| shermanm | how does IPA inventory infiniband? does it just treat it as IPoIB if the driver is loaded? | 16:55 |
| JayF | I don't recall, I think part of the issue my downstream has is around first runs them being configured differently than they set them up to be later? I'm not 100% sure | 16:58 |
| TheJulia | That is a super good question | 17:01 |
| TheJulia | historically I woudl think they would just have been viewed as IPoIB interfaces, but today I don't know | 17:01 |
| TheJulia | janders might have thoughts here, fwiw | 17:01 |
| shermanm | part of my issue is just having a really wide range of older and newer IB cards. the newer ones load `mlx5_core` and show up as ipoib, but the older ones need `mlx4_en` loaded manually. | 17:09 |
| shermanm | tbh I'm leaning towards a plugin to run tools like `ibv_devinfo` and get richer output anyway | 17:09 |
| TheJulia | I guess you could force the module to be loaded | 17:16 |
| TheJulia | ... its not an awful thing | 17:16 |
| TheJulia | but yeah, a plugin may be good, I just don't know if they would get loaded or could be loaded into network itnerfaces, I guess there is a hook there you could override and merge results into | 17:17 |
| JayF | Tomorrow I'm out for the US holiday, and I'm leaving early today to recharge my batteries with fresh sunlight. o/ | 17:45 |
| JayF | have a good one folks | 17:45 |
| TheJulia | you too | 17:49 |
| TheJulia | I'm not far behind you, I have one more meeting today | 17:49 |
| opendevreview | Esther Domfeh proposed openstack/ironic master: feat: add state, target_provision_state, and duration_seconds to node history https://review.opendev.org/c/openstack/ironic/+/989994 | 18:24 |
| opendevreview | Julia Kreger proposed openstack/ironic-python-agent master: Filter interfaces with locally-administered MAC addresses https://review.opendev.org/c/openstack/ironic-python-agent/+/993999 | 20:27 |
| opendevreview | Julia Kreger proposed openstack/ironic master: Add configurable OUI ignore list for MAC filtering https://review.opendev.org/c/openstack/ironic/+/994000 | 20:29 |
| TheJulia | rpittau: iurygregory: I think the IPA side change is slightly cleaner at the cost of visibility. | 20:29 |
| TheJulia | Anyway, food for thought! | 20:29 |
| iurygregory | TheJulia, ack will take a look | 20:51 |
| janders | shermanm CC TheJulia I didn't use inspection with my IB fleet in my HPC days, I deployed without inspection (uniform fleets). Custom IPA/hardwaremanager may be the go in your situation. Worth checking what https://opendev.org/openstack/ironic-python-agent/src/branch/master/ironic_python_agent/hardware_managers/mlnx.py already does. | 23:11 |
| janders | if you're going down this path, should be easy enough to deal with module-loading issue across different adapter generations | 23:12 |
| TheJulia | Also, patches always welcome to that hardware manager! | 23:14 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!