| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-04-06 | |||
| 14:57:51 | bauzas | we need to enable something and then discuss on a separate change | |
| 14:57:58 | stephenfin | any unknown keys | |
| 14:58:07 | stephenfin | e.g. hw:cpu_policyyy | |
| 14:58:13 | stephenfin | which is why I wanted strict | |
| 14:58:30 | dansmith | but people can use their own custom keys for scheduler behavior yeah? so we can't reject *everything* | |
| 14:58:35 | bauzas | stephenfin: it can still be a recommended choice | |
| 14:58:37 | johnthetubaguy | yeah, I think if we are strict for all known namespaces, it would be much better | |
| 14:58:43 | bauzas | stephenfin: but this would only be a doc thing | |
| 14:59:00 | bauzas | johnthetubaguy: we had this discussion 3 years ago | |
| 14:59:19 | johnthetubaguy | bauzas: we probably did, I might have been wrong then too | |
| 14:59:20 | stephenfin | dansmith: yes, and I've documented how they can add their own validators for whatever extra specs they want | |
| 14:59:27 | bauzas | johnthetubaguy: and we said we had to be accepting *any* key because we can't assume whether it's a typo or a custom key | |
| 14:59:28 | dansmith | wait, | |
| 14:59:32 | stephenfin | assuming they don't want to simply opt of validation | |
| 14:59:35 | sean-k-mooney | dansmith: right hence the idea of having "custom:" namespace where you can add stuff we woudl never validate | |
| 14:59:38 | dansmith | so in strict mode they have to write code in order to be able to use things like the json filter? | |
| 15:00:04 | bauzas | the 'custom:' namespace is good but this requires some proper signaling (and at least one cycle of deprecation) | |
| 15:00:15 | sean-k-mooney | dansmith: json filter uses shcduler hint | |
| 15:00:28 | bauzas | ie. we say "okay, look, custom filters will continue to work without any change in U and V" | |
| 15:00:41 | dansmith | there's some filter we can hit custom fields on the flavor isn't there? | |
| 15:00:53 | bauzas | "but starting from W, you will need to change your flavors to use a custom: prefix" | |
| 15:01:09 | sean-k-mooney | dansmith: yes compute capablity filter and the aggartioninstnaceType filter | |
| 15:01:14 | stephenfin | yes, 'CapabilitiesFilter' and 'AggregateInstanceExtraSpecsFilter' filters | |
| 15:01:17 | sean-k-mooney | dansmith: both are supproted | |
| 15:01:22 | stephenfin | and I've enumerated the known values for both | |
| 15:01:23 | sean-k-mooney | it jsut allows any key | |
| 15:01:31 | sean-k-mooney | if you use there namespaces | |
| 15:01:34 | stephenfin | *known keys | |
| 15:01:50 | stephenfin | the values are basically wildcards because of how flexible thoseare | |
| 15:01:51 | openstackgerrit | Lee Yarwood proposed openstack/nova master: workarounds: Add option to disable native LUKSv1 decryption by QEMU https://review.opendev.org/708030 | |
| 15:01:52 | openstackgerrit | Lee Yarwood proposed openstack/nova master: workarounds: Add option to locally attach RBD volumes https://review.opendev.org/708029 | |
| 15:01:53 | stephenfin | *those are | |
| 15:01:58 | melwitt | frickler: yes https://docs.openstack.org/releasenotes/nova/stein.html#known-issues | |
| 15:02:13 | sean-k-mooney | stephenfin: you allow any key in there namespace right | |
| 15:02:42 | sean-k-mooney | because those filters allow arbitray key names as long as they are in the same namespaces | |
| 15:02:49 | bauzas | sean-k-mooney: dansmith: stephenfin: honestly, I think I'm +2 on a deprecation policy for those filters | |
| 15:03:07 | bauzas | like, no longer supporting those wildcards at W | |
| 15:03:16 | bauzas | and the same goes for custom filters | |
| 15:03:20 | sean-k-mooney | bauzas: that a sperate topic | |
| 15:03:27 | stephenfin | sean-k-mooney: not the Capabilities filter anyway - those have to map to a field on the HostState object or ComputeNode o.v.o | |
| 15:03:27 | bauzas | sean-k-mooney: no, that's tied | |
| 15:03:38 | frickler | melwitt: ha, thx for pointing me to the obvious docs | |
| 15:03:42 | bauzas | we can't be strict on enforcing keys until we fixed those messed keys | |
| 15:03:44 | sean-k-mooney | stephenfin: ya i was thinking of the other one | |
| 15:03:55 | bauzas | (internal team meeting FTW) | |
| 15:04:07 | dansmith | okay I see the a.i.e.s filter already has a namespace, I didn't realize | |
| 15:04:07 | sean-k-mooney | bauzas: we can they have there own namespace | |
| 15:04:08 | stephenfin | sean-k-mooney: yes, that's wildcarded | |
| 15:04:17 | sean-k-mooney | but ya i guess i should join intrenal call | |
| 15:04:21 | stephenfin | yup, since Grizzly (thanks, Intel) | |
| 15:04:40 | bauzas | that bears me (joke) | |
| 15:05:34 | sean-k-mooney | dansmith: yep i made sure those fileter would still work in the spec review | |
| 15:05:51 | sean-k-mooney | sicne they used to be used for dpdk/numa stuff alot | |
| 15:05:56 | openstackgerrit | Lee Yarwood proposed openstack/nova master: DNM - Test stable device rescue tests with BFV instances https://review.opendev.org/710050 | |
| 15:06:48 | bauzas | honestly, that's now 4 hours we're discussing over this and I'm out of steam now | |
| 15:07:21 | dansmith | bauzas: wait until people start asking about how to do this in osc in a year :) | |
| 15:07:24 | bauzas | so, again, either we make this permissive now or we just punt this for now | |
| 15:08:01 | bauzas | ... | |
| 15:08:20 | stephenfin | johnthetubaguy: so in this permissive for unknown namepaces only model, would we reject e.g. every unrecognized 'hw:' extra spec? | |
| 15:08:28 | bauzas | dansmith: I just feel we made too many gifts in the past with custom and in-tree filters | |
| 15:08:39 | johnthetubaguy | stephenfin: I think that is a yes, to prevent the most obvious typos | |
| 15:08:40 | bauzas | we should ask for some return | |
| 15:08:57 | johnthetubaguy | stephenfin: or rather, for the user to know things are validated if its in a known namespace | |
| 15:09:48 | stephenfin | johnthetubaguy: okay, in that case do we need to continue to provide a way to disable validation in case there are people there with e.g. 'hw:something_custom' right now? | |
| 15:09:54 | johnthetubaguy | stephenfin: bonus, we would only need a microversion to add a new namespace | |
| 15:10:03 | openstackgerrit | Lee Yarwood proposed openstack/nova master: libvirt: Remove MIN_LIBVIRT_MULTIATTACH https://review.opendev.org/710238 | |
| 15:10:10 | johnthetubaguy | stephenfin: I think we would just drop that flag for now, and see how people take it | |
| 15:10:44 | johnthetubaguy | I like the simpler API of there being a list of validated namespaces | |
| 15:10:59 | johnthetubaguy | stops all the mess of migrating people towards "custom" something | |
| 15:11:26 | stephenfin | okay, so unless our hand was forced we'd be essentially saying "you need to change this if you ever want to use this microversion" | |
| 15:12:03 | johnthetubaguy | I think so, yes | |
| 15:12:44 | johnthetubaguy | well, you can use the new API version to add your new key, and delete the old bad key | |
| 15:12:46 | stephenfin | Cool. Last one. Do we still need strict in this model (or that parameter in general), seeing as we are effectively strict for all recognized namespaces | |
| 15:12:58 | johnthetubaguy | I don't think so | |
| 15:13:03 | johnthetubaguy | at least not to start with | |
| 15:13:09 | stephenfin | Okay, so I can drop that parameter | |
| 15:13:25 | johnthetubaguy | you could typo the namespace... but, hey, at least we are helping you more now | |
| 15:13:52 | stephenfin | That all works for me. gibi, bauzas, sean-k-mooney, dansmith: any significant concerns with that before I do the needful ^ ? | |
| 15:14:25 | stephenfin | (last 15 lines or so of scrollback) | |
| 15:14:59 | openstackgerrit | Lee Yarwood proposed openstack/python-novaclient master: Microversion 2.86 - Stable device boot from volume rescue https://review.opendev.org/714956 | |
| 15:15:08 | johnthetubaguy | stephenfin: for completeness, the old microversions are still unchanged, no validation at all there | |
| 15:15:22 | stephenfin | Yeah, agreed | |
| 15:16:25 | johnthetubaguy | sorry that all took so long, but I think what we have at the end is better | |
| 15:16:43 | johnthetubaguy | and easier to document and use :) | |
| 15:16:49 | sean-k-mooney | ill read back in a few minutes. | |
| 15:17:13 | lyarwood | stephenfin: ah, both of us are going for 2.86 again, want me to move to 2.87? | |
| 15:17:45 | stephenfin | lyarwood: Yup :) Your call but it might make sense | |
| 15:18:55 | lyarwood | stephenfin: yup no issues, I wasn't paying attention | |
| 15:20:42 | bauzas | stephenfin: pardon my French but I may have misunderstood the agreement in between you and johnthetubaguy | |
| 15:21:14 | bauzas | what's the outcome when it's said "drop this flag" ? drop the microversion ? | |
| 15:21:39 | stephenfin | keep the microversion, but drop the '?validate' argument | |
| 15:21:50 | bauzas | and the default being ? | |
| 15:22:12 | bauzas | permissive as I can understand crom 'I don't think so' from johnthetubaguy ? | |
| 15:22:17 | bauzas | from* | |
| 15:22:32 | stephenfin | strict for all known namespaces (hw:, os:, vmware:, ...) | |
| 15:22:39 | stephenfin | don't care for anything outside that set | |
| 15:24:09 | bauzas | I'm cool with this | |
| 15:24:15 | bauzas | stephenfin: ^ | |
| 15:24:30 | bauzas | no upgrade impact, pretty clear | |
| 15:24:47 | bauzas | stull leaves nothing unchanged except for already-defined namespaces | |