| opendevreview | halfcrazy proposed openstack/sushy master: Fix error handling for non-object JSON error bodies https://review.opendev.org/c/openstack/sushy/+/1005824 | 04:03 |
|---|---|---|
| opendevreview | halfcrazy proposed openstack/sushy master: Fix error handling for non-object JSON error bodies https://review.opendev.org/c/openstack/sushy/+/1005824 | 04:05 |
| opendevreview | halfcrazy proposed openstack/sushy master: Fix error handling for non-object JSON error bodies https://review.opendev.org/c/openstack/sushy/+/1005824 | 04:08 |
| rm_work | cardoe: are you actually trying out that http-RPC driver? I noticed you backported a fix that really only matters if you’re doing that 😂 | 06:19 |
| rm_work | I’m also on vacation in Germany this week but busy working on OSS, so I feel that sentiment pretty hard lol | 06:20 |
| opendevreview | Verification of a change to openstack/ironic-python-agent master failed: Maintain plugin requirements as optional dependencies https://review.opendev.org/c/openstack/ironic-python-agent/+/1000668 | 08:38 |
| *** hroy| is now known as hroy | 08:44 | |
| rpittau | hello everyone! I'm checking the open patches for the services final releases, please double-check my statements: (sorry for the wall of text!) | 08:48 |
| rpittau | in bifrost https://review.opendev.org/c/openstack/bifrost/+/968363 is in the verification queue, https://review.opendev.org/c/openstack/bifrost/+/1002083 would nice to merge | 08:48 |
| rpittau | in IPA 3 patches are in the verification queue, no others left open | 08:48 |
| rpittau | in IPE all is merged, ready to release | 08:48 |
| rpittau | in ironic I see 2 patches in the priority list with no votes (bug fixes) | 08:48 |
| rpittau | I will propose the final releases tomorrow at the latest | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/ironic-python-agent-builder stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/ironic-python-agent-builder/+/1005836 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/ironic-python-agent-builder stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/ironic-python-agent-builder/+/1005837 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/ironic-python-agent-builder master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/ironic-python-agent-builder/+/1005838 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/ironic-ui stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/ironic-ui/+/1005839 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/ironic-ui stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/ironic-ui/+/1005840 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/ironic-ui master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/ironic-ui/+/1005841 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/networking-baremetal stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/networking-baremetal/+/1005842 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/networking-baremetal stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/networking-baremetal/+/1005843 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/networking-baremetal master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/networking-baremetal/+/1005844 | 08:48 |
| opendevreview | OpenStack Release Bot proposed openstack/networking-generic-switch stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/networking-generic-switch/+/1005845 | 08:49 |
| opendevreview | OpenStack Release Bot proposed openstack/networking-generic-switch stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/networking-generic-switch/+/1005846 | 08:49 |
| opendevreview | OpenStack Release Bot proposed openstack/networking-generic-switch master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/networking-generic-switch/+/1005847 | 08:49 |
| opendevreview | Merged openstack/networking-baremetal master: tox: Drop redundant passenv https://review.opendev.org/c/openstack/networking-baremetal/+/1004209 | 08:50 |
| opendevreview | Merged openstack/ironic-ui stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/ironic-ui/+/1005839 | 09:36 |
| opendevreview | Merged openstack/networking-generic-switch stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/networking-generic-switch/+/1005845 | 09:36 |
| opendevreview | Merged openstack/networking-generic-switch stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/networking-generic-switch/+/1005846 | 09:36 |
| opendevreview | Merged openstack/networking-baremetal stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/networking-baremetal/+/1005842 | 09:39 |
| opendevreview | Merged openstack/ironic-ui stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/ironic-ui/+/1005840 | 09:39 |
| opendevreview | Merged openstack/ironic-ui master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/ironic-ui/+/1005841 | 09:39 |
| opendevreview | Merged openstack/networking-baremetal stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/networking-baremetal/+/1005843 | 09:40 |
| opendevreview | Merged openstack/networking-baremetal master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/networking-baremetal/+/1005844 | 09:40 |
| opendevreview | Merged openstack/networking-generic-switch master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/networking-generic-switch/+/1005847 | 09:42 |
| opendevreview | Merged openstack/ironic-python-agent-builder stable/2026.2: Update .gitreview for stable/2026.2 https://review.opendev.org/c/openstack/ironic-python-agent-builder/+/1005836 | 09:44 |
| opendevreview | Merged openstack/ironic-python-agent-builder stable/2026.2: Update TOX_CONSTRAINTS_FILE for stable/2026.2 https://review.opendev.org/c/openstack/ironic-python-agent-builder/+/1005837 | 09:44 |
| opendevreview | Merged openstack/ironic-python-agent-builder master: Update master for stable/2026.2 https://review.opendev.org/c/openstack/ironic-python-agent-builder/+/1005838 | 09:44 |
| iurygregory | Folks, today morning/afternoon I will be out, need to travel for a doc appointment | 09:45 |
| dtantsur | dxterslab: hey, we need to talk re metal3 networking. I think we have a lot of bits in cooking. | 09:46 |
| iurygregory | I just noticed Ironic wasn't included in the "October 2026 PTG Team List Annoucement" email from Kendall, I've sent an email to ptg@openinfra.org asking to include Ironic on it. | 09:47 |
| dtantsur | dxterslab: https://github.com/metal3-io/metal3-docs/blob/main/design/ironic-standalone-networking-service.md will add basic switch management to Metal3. https://review.opendev.org/c/openstack/ironic-specs/+/995954 is talking about evpn. | 09:47 |
| dtantsur | dxterslab: realistically, we can add evpn outside of the scope of the 2nd spec. Now, kube-ovn also supports evpn. So at least in theory this whole effort should get you where you want to be. | 09:48 |
| dtantsur | dxterslab: we need any help we can get to push it forward though. | 09:48 |
| opendevreview | Merged openstack/ironic-python-agent master: Replace deprecated pyparsing interface https://review.opendev.org/c/openstack/ironic-python-agent/+/1002971 | 10:02 |
| opendevreview | Merged openstack/ironic-python-agent master: Remove burnin-network-kazoo extra https://review.opendev.org/c/openstack/ironic-python-agent/+/998721 | 10:02 |
| opendevreview | Merged openstack/ironic-python-agent master: Provide more context about inspection on lookup https://review.opendev.org/c/openstack/ironic-python-agent/+/968475 | 10:02 |
| priteau | iurygregory: I was just coming here to ask about it :) | 10:31 |
| TheJulia | iurygregory: Thanks! I was going to ask | 14:05 |
| TheJulia | dtantsur: I was talking to dxterslab last night as well, I think an action item is likely path and configuraiton modeling for the DPUs to treat them as "switch" devices, but what that actually looks like configuration modeling wise might be a little more work that is likely going to require someone to go heads-down | 14:06 |
| dtantsur | TheJulia: I'd rather not get DPU involved, but if that's the requirement.. | 14:06 |
| dtantsur | .. then yes, it's a whole new feature development | 14:07 |
| * TheJulia again discuss NVMe-OF | 14:07 | |
| TheJulia | well, yeah, the other key aspect is workflow, and just the variety of options, we've got solid foundations built already which account for some of it, is_smartnic being one of them, but that is also built on an entirely neutron model as well. I'm all for moving that forward in general, but there is also just a ton of work there and ... yeah | 14:08 |
| * TheJulia ponders going back to bed | 14:08 | |
| dtantsur | Going back to bed is always the wise choice | 14:09 |
| dtantsur | On the DPU topic, is it still a one vendor show? | 14:09 |
| TheJulia | it is an each vendor does their own thing show, which is very much just like switches, which is why I keep thinking n-g-s as a good mechanism/place to hide the variety of attachment/config logic | 14:10 |
| dtantsur | Yeah, so the point I'm trying to make is: switch support is coming. DPU support requires starting from nearly ground up. | 14:11 |
| TheJulia | On a plus side, AMD's DPUs *do* AIUI, allow you to run SONiC on them | 14:11 |
| TheJulia | sort of, what might be a good thread to pull on is to see if https://ovn-kubernetes.io/master/installation/launching-ovn-kubernetes-with-dpu/ can work on supporting non-k8s workloads at all on the actual physical machine. There is some suggestion of knowing the host side physical function cross-over interface, which makes me worried that it is a no, but the root issue is everyone does something different. The base aspect there is | 14:14 |
| TheJulia | if there is a semi-standard mechanism to control that configuration and say a k8s cluster is running across DPUs directly and it can support non k8s workloads on the host, then success. If not, Then we just maybe set the expectation "you need to join the cards to an EVPN fabric, and ironic keeps the logic at "login to device, configure VNI locally for usage (so the card loads the routes), and then bind that vni to a vlan/virtual or | 14:14 |
| TheJulia | physical function to be presented to the overall host" | 14:14 |
| TheJulia | which is... almost exactly what we do on switches today | 14:14 |
| TheJulia | (and, fwiw, I believe there is a spec document someplace which lays that out as well) | 14:14 |
| TheJulia | I guess the issue is largely where to draw the lines, because as the physical host orchustrator, we also need to ensure the dpu is online in advance of attempting to configure it.... | 14:16 |
| dtantsur | Yeah, and Metal3 doesn't currently model child nodes | 14:16 |
| TheJulia | which is why we have the existing is_smartnic stuff, and it is not clear to me the number of vendors which supports the card operating on a single core on standby power | 14:16 |
| TheJulia | child nodes was also largely for direct overall management, if your treating it as a switch, thats less of a thing then but still needing to navigated. At least treating it as a switch abstracts some of it away | 14:17 |
| TheJulia | the issue is these stupid things are so infinitely configurable | 14:17 |
| TheJulia | And, no, nobody get any ideas of embedding infinity stones in them. kthxbai | 14:18 |
| dtantsur | damn! | 14:18 |
| TheJulia | Next the AI core will take it over, and suddenly we'll have ultron mk2 | 14:18 |
| * TheJulia is still fond of remembering the customer that wanted to run BGP route reflectors on them.... | 14:20 | |
| opendevreview | Julia Kreger proposed openstack/ironic master: Fix file descriptor leak in neutron client connections https://review.opendev.org/c/openstack/ironic/+/1005807 | 14:20 |
| TheJulia | I asked nobodycam to take a look at ^ since he is the one who mentioned that to me over dinner one night and claude he noted there were two other places to fix. | 14:21 |
| TheJulia | looks like we don't have a previous bug. I think the issue is he just hit a scale where he really felt this issue, less so post-eventlet though for the code base but it is also likely hurting thread reclaiming a little so its likely a bit better to just take the more explicit approach. | 14:26 |
| TheJulia | same basic pattern of issue appears to exist in some of the other cases where we fire up tenant aware clients, claude is doing the leg work while I feel like going back to bed. :) | 14:27 |
| dxterslab | TheJulia, dtantsur: I would treat DPUs as compute nodes with the capabilities of network functions. DPUs are being looked at not just for networking offload capabilites but also for storage enhancement capabilities. There are companies looking to create agents that run directly on them to bring storage services. I also saw some documents around NVIDIA DOCA about virtiofs improvements under BF3 DPUs. In the AI space, NVIDIA has | 14:28 |
| dxterslab | made it part of their reference design to have DPUs as part of it | 14:28 |
| dtantsur | I don't doubt that NVIDIA is quite keen on it ;) | 14:28 |
| TheJulia | oh yeah, they have basically forced oems to bundle their cards in | 14:29 |
| dxterslab | yup... they have forced just not oems but customers to only accept their ref arch... unfortunate world to live i | 14:29 |
| TheJulia | Yeah, I have a customer super keen on magically having storage volume attachments through dpus as well | 14:30 |
| * TheJulia twitches | 14:30 | |
| dtantsur | They just need to deploy some AI agents on DPUs and tell them to manage storage | 14:30 |
| dtantsur | Boom, problem solved | 14:30 |
| dxterslab | The only DPU i have seen that can be treated as a switch since it uses P4 to program its FPGA is AMD pensando | 14:30 |
| TheJulia | lol | 14:30 |
| TheJulia | well, "formally" as a switch :) | 14:31 |
| dxterslab | But, I don't know what other apps you can run on it. They market storage performance... so i do believe DPUs should be treated like any other host and some type of label needs to exist to identify its role if ironic is planning to keep track of that | 14:31 |
| TheJulia | since their FPGA supports p4 | 14:31 |
| TheJulia | dxterslab: Yeah, we have this whole concept of a child node, but it has never really panned out. I did work in that direction because I had customers wanting us to deploy OSes to the devices and be able to model the linkage. But the DPU plays by vendors have shifted from bring your own os to "we provide certified oses" | 14:32 |
| dtantsur | Hmm, as an aside, should we deprecate child nodes then? | 14:33 |
| TheJulia | There are a pile of reasons there I don't want to get into, but it boils down to "those oses and packages have particular patches which are not in mainlines" | 14:33 |
| dxterslab | the way i have treated ironic is as a provisioner and the correlation between DPU and baremetal node exist in my CMDB. We use netbox to model this. Does ironic need to be able to model this or should this be the job of a cmdb? | 14:33 |
| TheJulia | dunno, there feels like there is a whole separate discussion here | 14:33 |
| TheJulia | The linkage really depends on how you model inventory and link it in | 14:34 |
| dxterslab | ah!!! i see | 14:34 |
| TheJulia | cmdbs, accounting, asset tracking, thats all opinionated | 14:34 |
| TheJulia | (and largely driven as well by accounting rules applied to your business and inventory) | 14:35 |
| dxterslab | dtantsur: for your development of network-orchestration, what does n-s-g do? Does it only configure the port or can it also configure the overlay before configuring the vlans on the ports on the switch? reason i ask is because we are looking to automate the entire process of creating the overlay network for a tenant i.e (vpc) and allocate the node to the right VLAN on the ports it is connected to. We are also likely planning to do | 14:40 |
| dxterslab | this using gnmi or rest | 14:40 |
| JayF | dxterslab: n-g-s uses generic methods (today; ssh'ing to a switch and configuring it) to setup switches for baremetal | 14:41 |
| TheJulia | Restconf and netconf methods are under development as well | 14:41 |
| JayF | dxterslab: there is some capability to do meta-work, like mapping down vxlan networks, but that interface is only utilized by neutron integration | 14:41 |
| dxterslab | is neutron a requirement to be able to use n-g-s? | 14:42 |
| TheJulia | it doesn't make assumptions about what style of VXLAN you have, but some drivers have logic to handle needing to treat multicast vxlan or evpn style configurations as well. | 14:42 |
| TheJulia | dxterslab: not explicitly, although evpn/vxlan support is not in place yet for the direct ironic-networking invocation of n-g-s | 14:42 |
| JayF | dxterslab: ngs is sorta "an API to configure switches". Neutron and standalone both use it, but the neutron integration supports more advanced networking (this is mainly related to maturity/age rather than design -- and neutron gives us some of it for free) | 14:43 |
| TheJulia | alegacy_: fyi, you may be interested in this discussion. | 14:43 |
| cardoe | We missed PTG sign up. | 14:48 |
| TheJulia | iurygregory already followed up | 14:48 |
| TheJulia | cardoe: out of curiosity, are you seeing any orphaned or stale fd's talking to neutron on your prod ironic ? | 14:51 |
| cardoe | Yes after you started you patch I looked. Conductor held a lot. | 14:52 |
| cardoe | Lots of chatter about network stuff. About to hit the road through. | 14:52 |
| TheJulia | cardoe: okay, good to know, I've got a patch for neutron, patches in progress for swift and cinder. | 14:58 |
| TheJulia | noted on the bug I just created | 14:59 |
| TheJulia | https://bugs.launchpad.net/ironic/+bug/2167506 | 15:00 |
| opendevreview | Julia Kreger proposed openstack/ironic master: Fix file descriptor leak in neutron client connections https://review.opendev.org/c/openstack/ironic/+/1005807 | 15:04 |
| cardoe | Network stuff is something I’ve got feels but don’t have time to talk today. On PTO. | 15:17 |
| TheJulia | yes, go pto | 15:17 |
| TheJulia | pto == gooooood | 15:17 |
| dxterslab | i don't know you but PTO extremely important | 15:18 |
| dxterslab | Moreso if you are doing extreme activities like MTB or skiing | 15:18 |
| TheJulia | MTB ? | 15:20 |
| dxterslab | mountain biking lol | 15:21 |
| dxterslab | recently got into it and love it | 15:21 |
| TheJulia | oh, heh. I have a neighboring seat holder at the local hockey stadium who does that. It sounds fun, but it sounds like he has broken a few bones in non-fun ways doing it. | 15:23 |
| JayF | I have seen some videos of folks doing it first person. It looks eerily similar to falling down a mountain, just some wheels are involved. No thanks. | 15:35 |
| dtantsur | I could use an MTB given the landscape around. But not for trails, no | 15:39 |
| TheJulia | Here, I would be afraid the wind blown sand would just strip the paint :) | 15:48 |
| rm_work | Yeah but … when I’m on PTO I get bored and want to do OSS work lol | 15:51 |
| dtantsur | Going to places without internet helps a lot | 15:55 |
| JayF | When I take PTO I have to resist the urge to never come back again. | 15:57 |
| JayF | lol | 15:57 |
| dtantsur | So true | 16:26 |
| dtantsur | Folks, wdyt about backportability of https://review.opendev.org/c/openstack/ironic/+/1005010 and https://review.opendev.org/c/openstack/ironic/+/1005083? | 16:28 |
| JayF | I don't think they are good for backporting. | 16:44 |
| JayF | It's going to change inventory value formats in a way someone wouldn't expect in a stable branch | 16:44 |
| JayF | those aren't like, technically an API, but I would be very upset as an oper if I had a script break on a "stable" branch because an output like this changed | 16:45 |
| dtantsur | That's https://review.opendev.org/c/openstack/ironic/+/1005083, but what about https://review.opendev.org/c/openstack/ironic/+/1005010? | 16:45 |
| JayF | the accelerators.py change is clearly backportable | 16:46 |
| JayF | but it's paired with something borderline -- support for a new storage api | 16:46 |
| JayF | so I'm on the fence. The fix is good enough that it should be backported but it shouldn't have been paired with a feature | 16:47 |
| dtantsur | "support for a new storage api" may be a bit of an overstatement. We already support both Storage and SimpleStorage, we just don't pick extra fields from the former. | 16:47 |
| JayF | I mean, I'm 50/50 if it's a "new API" | 16:53 |
| JayF | so that for sure pushes it to JFDI | 16:53 |
| TheJulia | if it makes things better, then I can go along with jfdi | 16:56 |
| iurygregory | I was pretty sure I've signed up ironic when kendall sent the first email about the ptg, but maybe I forgot to click submit... | 20:21 |
| iurygregory | I didn't receive any confirmation yet after reaching out to ptg@openinfra.org this morning... | 20:22 |
| JayF | All that matters is that we're listed now. | 20:22 |
| JayF | I emailed them as well, no response | 20:22 |
| iurygregory | ok | 20:22 |
| iurygregory | just got back from my travel | 20:22 |
| iurygregory | time to work :D | 20:23 |
| TheJulia | Kendall may also be in some crazy time zone | 20:49 |
| iurygregory | ack | 21:03 |
| opendevreview | Julia Kreger proposed openstack/ironic master: Fix file descriptor leak in Swift client connections https://review.opendev.org/c/openstack/ironic/+/1005971 | 23:37 |
| opendevreview | Julia Kreger proposed openstack/ironic master: Fix file descriptor leak in Cinder client connections https://review.opendev.org/c/openstack/ironic/+/1005972 | 23:37 |
| opendevreview | Julia Kreger proposed openstack/ironic master: Fix file descriptor leak in Cinder client connections https://review.opendev.org/c/openstack/ironic/+/1005972 | 23:38 |
| TheJulia | okay, there we go. | 23:39 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!