| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2025-06-12 | |||
| 16:48:22 | TheJulia | heh, wrong channel :) | |
| 16:48:54 | opendevreview | Balazs Gibizer proposed openstack/nova stable/2024.1: Add functional reproducer for bug 2102038 https://review.opendev.org/c/openstack/nova/+/952484 | |
| 16:48:55 | opendevreview | Balazs Gibizer proposed openstack/nova stable/2024.1: Return HTTP400 for multi spec pci alias if PCI in Placement https://review.opendev.org/c/openstack/nova/+/952485 | |
| 16:48:56 | opendevreview | Balazs Gibizer proposed openstack/nova stable/2024.1: Multiple spec per PCI alias limitation https://review.opendev.org/c/openstack/nova/+/952486 | |
| 16:48:57 | opendevreview | Balazs Gibizer proposed openstack/nova stable/2024.1: Validated that PCI alias has proper ids https://review.opendev.org/c/openstack/nova/+/952487 | |
| 16:48:58 | opendevreview | Balazs Gibizer proposed openstack/nova stable/2024.1: Validate [pci]alias at service startup https://review.opendev.org/c/openstack/nova/+/952488 | |
| 16:48:59 | opendevreview | Balazs Gibizer proposed openstack/nova stable/2024.1: Cache [pci]alias parsing https://review.opendev.org/c/openstack/nova/+/952489 | |
| 16:53:21 | TheJulia | no worries | |
| 17:16:40 | sean-k-mooney | JayF: so on that | |
| 17:17:00 | sean-k-mooney | not inclduing the MTU in the metadta was specificly request by ironic a number of cycles ago | |
| 17:17:02 | JayF | I meant to tell you: welcome to the claude-code club :) It was nice to see someone not-me using it :D | |
| 17:17:21 | JayF | Sounds like we've corrected a previous mistake, then. | |
| 17:17:36 | sean-k-mooney | we do it condtionally | |
| 17:17:53 | sean-k-mooney | if the network has dhcp we dont incldue it and allow it to be managed vi dhcp | |
| 17:18:01 | JayF | Julia is the expert in this bit moreso than I am :) | |
| 17:18:07 | sean-k-mooney | btu if there is no dhcp enabled we includde it for people doing static ip | |
| 17:18:37 | sean-k-mooney | ack i was just wonderign why its only coming up now it think we didi it 2 or we release ago | |
| 17:19:45 | sean-k-mooney | TheJulia: context was includign the MTU infroamtion in teh metadat made it much harder to go form ml2/ovs to ml2/ovn becacuce that requried reducing the mtu | |
| 17:20:04 | sean-k-mooney | so neuton asked us to only set it if there was no dhcp | |
| 17:21:09 | sean-k-mooney | JayF: on claude-code ya i decied to pay for it for a year and see where the state of the art is vs local llms ectra | |
| 17:21:35 | JayF | sean-k-mooney: so, you want some advice since I've been poking at it for a few weeks? I made a openstack-dev dir, put a claude.md in it, and checked out my other projects under it | |
| 17:21:38 | sean-k-mooney | so far im likeing the terminal first approch not that i have done a lot with it | |
| 17:21:48 | TheJulia | sean-k-mooney: we've had reports of mtu weirdness and we're seeing a huge increase in not using dhcp so just makes sense to address it. | |
| 17:21:52 | JayF | the pycharm ide integration is nice, it uses the jetbrains diff stuff | |
| 17:22:07 | sean-k-mooney | JayF: you mean like this https://github.com/SeanMooney/grian-dev | |
| 17:22:12 | TheJulia | since, the mismatch between what is known and what an OS can assume otherwise introduces a huge amount of risk | |
| 17:22:13 | JayF | if you wanna use mine as a start https://www.irccloud.com/pastebin/sL4RIxrg/CLAUDE.md | |
| 17:22:29 | sean-k-mooney | JayF: i also have created https://github.com/SeanMooney/openstack-ai-style-guide | |
| 17:22:41 | sean-k-mooney | which i might rename because i want ot evolve that ot be more hten just style | |
| 17:22:51 | JayF | okay I think yours wins for quality lol | |
| 17:23:07 | JayF | but I will say: the pattern of "start claude in a dir that has all your openstack stuff checked out" has been really nice | |
| 17:23:20 | sean-k-mooney | i got claude to help write it | |
| 17:23:21 | TheJulia | plus, cases where dhcp may not even be accessible without network data out of the box and where config drive must be relied upon to get started | |
| 17:23:42 | JayF | I gave it a sample thing during trying it out, pointing it to a feature that'd require implementation in Ironic + modification of hte nova virt driver, and it did a decent job of navigating it | |
| 17:24:03 | JayF | TheJulia: sean-k-mooney: plus the uncomfortable reality that often Ironic knows more about the physical network situation than neutron/nova :( | |
| 17:24:04 | sean-k-mooney | TheJulia: so that odd becasue we do still include the mtu info if dncp is disabled on neutron | |
| 17:24:26 | sean-k-mooney | TheJulia: if we are not then that a regresion we shoudl fix in nova too | |
| 17:24:55 | TheJulia | sean-k-mooney: dhcp only matters if it is reachable out of the box, that is not always true. It is clear cut for VMs, but you can have greater hurdles with physical machines | |
| 17:25:24 | sean-k-mooney | TheJulia: ah so this is also related to prot groups and bond/vlan netowrking | |
| 17:25:33 | JayF | sean-k-mooney: everything is also related to that :) | |
| 17:25:45 | JayF | sean-k-mooney: there's very few people running ironic in the real world not using at least one of those | |
| 17:25:51 | sean-k-mooney | well if we know its ironic, which we shoudl be cause it vnic type=baremetal | |
| 17:25:58 | sean-k-mooney | we may be able ot make ti work better, | |
| 17:26:05 | sean-k-mooney | but if you have already fixed it on the ironic side | |
| 17:26:09 | sean-k-mooney | then thats fine too | |
| 17:26:23 | JayF | Having Ironic own some of the newtork metadata is a feature we'd need for standalone anyway | |
| 17:26:35 | JayF | and gives us more flexibility since (as I said before) we often know more than you all do about the physical world | |
| 17:26:37 | sean-k-mooney | ya that makes sense | |
| 17:26:58 | TheJulia | Yeah, we're seeing more asks for smoothing that general experience so, just makes sense to spot/address the highly specific items to make things better | |
| 17:26:59 | sean-k-mooney | i also think thie idea of ironic haveign the ablity to generate or enrish the config drive is also useful for that case | |
| 17:27:12 | TheJulia | So... | |
| 17:27:16 | TheJulia | Funny you should mention that! | |
| 17:27:23 | JayF | and honestly, the more I think about ^^ (the physical world not always being perfectly modelled in nova) is a ... feature not a bug? The entire benefit of ironic+nova integrated is keeping the experience simple ... and the real world is complex as hell lol | |
| 17:27:33 | TheJulia | we've semi-wondered outloud about "should we offer ssh key injection?!" | |
| 17:28:03 | sean-k-mooney | well keyparis works with ironic already right | |
| 17:28:08 | TheJulia | oh yeah | |
| 17:28:08 | sean-k-mooney | but you mean in standalone | |
| 17:28:11 | TheJulia | yeah | |
| 17:28:27 | sean-k-mooney | i sort of feel like the keypari really shoudl be in keystone | |
| 17:28:31 | JayF | we already support it in a supremely roundabout way | |
| 17:28:40 | sean-k-mooney | and hten we jsut get it form there or in barbican | |
| 17:28:42 | JayF | sean-k-mooney: notably that doesn't help most of our standalone users who use basic auth :) | |
| 17:28:43 | TheJulia | sean-k-mooney: agree, and that would make it sort of easier | |
| 17:28:43 | sean-k-mooney | btu legacy | |
| 17:28:48 | TheJulia | yup | |
| 17:28:50 | JayF | you can do arbitrary file injection in a deploy template | |
| 17:28:58 | JayF | but that's a lot of crap to do just to get an ssh key in | |
| 17:29:01 | TheJulia | yeah | |
| 17:29:28 | sean-k-mooney | i mean teh format of the config drive/metadata is pretty standarsied | |
| 17:29:38 | sean-k-mooney | so having the ablity for ironic to do that is not a bad thing | |
| 17:29:46 | JayF | TheJulia: the killer potential feature hiding under the surface there: Ironic somehow being able to inject troubleshooting-keys into IPA at boot time :D | |
| 17:29:53 | sean-k-mooney | woudl we use it in the combined case proably not but maybe | |
| 17:30:12 | TheJulia | JayF: I'd have concerns about doing that, but yeah, I've had asks for it in the past | |
| 17:30:15 | sean-k-mooney | JayF: cant it od that or do you have to buidl them in | |
| 17:30:30 | JayF | TheJulia: I have concerns about the whole thing but I'm assuming we can find a way to have the cake and eat it too :D | |
| 17:30:30 | sean-k-mooney | its been a long time since i deployed with bifrost | |
| 17:30:39 | TheJulia | So one thing on this entire subject, I'd <3 if nova just shipped us the dicts, but I'd have to double check the constraints and modeling for user input in nova again | |
| 17:30:42 | sean-k-mooney | but i remeber sshing into the ipa in the past to debug somthing | |
| 17:30:47 | TheJulia | its been years since I've walked that code path | |
| 17:31:24 | sean-k-mooney | TheJulia: JayF may rememebr this better then i but i think we just send you json blobs today | |
| 17:31:40 | sean-k-mooney | i woudl have to look but i dont think we provide a filesytem but i oculd be wrong | |
| 17:31:42 | TheJulia | sean-k-mooney: I don't think y'all actually do | |
| 17:31:43 | JayF | I remember nothing :) | |
| 17:31:47 | JayF | but we can check! | |
| 17:31:50 | TheJulia | I think we get handed an iso9660 file right now | |
| 17:32:02 | sean-k-mooney | oh ok | |
| 17:32:23 | sean-k-mooney | i mean we can always change that if you come up with a design | |
| 17:32:42 | TheJulia | the factory which builds it in the overall virt utils, if memory serves just hands us the file pre-formed | |
| 17:32:53 | sean-k-mooney | i might be conflating this with the auto leasee feature | |
| 17:32:58 | TheJulia | yeah | |
| 17:33:28 | sean-k-mooney | so the reaosn i was quistioning it is we currently store it in memcache | |
| 17:33:35 | sean-k-mooney | if you turn that one | |
| 17:33:47 | sean-k-mooney | so we ahve it in a dict like serisable datascuture | |
| 17:34:10 | TheJulia | Isent there a mechanism on the nova side to do file injection? | |
| 17:34:28 | sean-k-mooney | yes its deprecated and ahs been for about a decaade or so | |
| 17:34:31 | TheJulia | oh, good | |
| 17:34:41 | sean-k-mooney | its called personality files | |
| 17:35:11 | sean-k-mooney | you can also pass a user-data cript for cloud init that is what was ment to repalce arbity file injection | |
| 17:35:21 | TheJulia | okay, as I mentioned, it has been ages | |
| 17:35:28 | TheJulia | in the user-data payload right? | |