| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2023-01-10 | |||
| 17:04:57 | sean-k-mooney | bauzas: did we settel on ironic is not supprot for host aggreates a while ago when chatting downstream | |
| 17:06:35 | bauzas | sean-k-mooney: I thought we said given we only support hostnames in aggregates (and not nodenames), I hardly understand how this can work | |
| 17:07:15 | bauzas | but you can target a specific nova-compute using aggregates so that can still work with sharded ironic nodes | |
| 17:07:41 | sean-k-mooney | when you add a comptue host to thet api https://docs.openstack.org/api-ref/compute/?expanded=add-host-detail#add-host | |
| 17:07:46 | sean-k-mooney | you are adding the comptue service | |
| 17:07:59 | sean-k-mooney | and ironic is currently loadbalanceing compute nodes between compute services | |
| 17:09:37 | bauzas | agreed, that's what I said | |
| 17:09:51 | bauzas | but you can't target a node, that's my point | |
| 17:10:19 | bauzas | the good news is that placement aggregates support nodes | |
| 17:10:45 | sean-k-mooney | right so what they are trying to do in that patch is incluse all the ironic RPs created for the ironic servier in the the placement aggreate | |
| 17:10:50 | bauzas | so theorically, you could do some flavorey query about it | |
| 17:11:16 | sean-k-mooney | you coudl but if we proceed with this patch we are baissly saying we now supprot host aggrates with ironic again | |
| 17:11:29 | sean-k-mooney | im pretty sure we did in the past and that broke at some point | |
| 17:11:59 | sean-k-mooney | where somepoint is a long long time ago | |
| 17:12:47 | bauzas | long long time ago, we were not having placement aggregates | |
| 17:13:09 | bauzas | and if I blame the existing add_host_to_agg code, I don't think I'll see a difference | |
| 17:14:02 | sean-k-mooney | didnt we used to have multipel compute services for ironic ndoes or something | |
| 17:15:40 | bauzas | I don't think so but I could be wrong | |
| 17:16:43 | sean-k-mooney | ok your right we did not | |
| 17:16:45 | sean-k-mooney | https://specs.openstack.org/openstack/nova-specs/specs/juno/implemented/add-ironic-driver.html#proposed-change | |
| 17:16:53 | sean-k-mooney | it was intially one compute service | |
| 17:17:20 | sean-k-mooney | then we supported muple in newton https://specs.openstack.org/openstack/nova-specs/specs/newton/implemented/ironic-multiple-compute-hosts.html | |
| 17:20:56 | sean-k-mooney | i guess based on https://specs.openstack.org/openstack/nova-specs/specs/newton/implemented/cells-aggregate-api-db.html | |
| 17:21:11 | sean-k-mooney | the aggreate api si ment ot be aggreate to compute node mappings | |
| 17:21:21 | sean-k-mooney | in which case then yes ironci shoudl work | |
| 17:21:37 | sean-k-mooney | you just need to add each node by the ironic uuid | |
| 17:21:49 | sean-k-mooney | however that is what we tried downstream and it did not seam to work | |
| 17:21:58 | sean-k-mooney | and i recal that that used to work before | |
| 17:23:19 | bauzas | that's what I remember when we discussed about placement aggs | |
| 17:23:41 | bauzas | in general, we only support the first node | |
| 17:23:48 | sean-k-mooney | well for nova aggreates ignoring placment that is what used to work in newton | |
| 17:24:10 | sean-k-mooney | we coudl map compute nodes to aggates by the host key | |
| 17:24:30 | sean-k-mooney | which presumable was hypervior_hostname | |
| 17:24:39 | bauzas | no | |
| 17:24:51 | bauzas | host was in general the Host conf opt | |
| 17:25:03 | sean-k-mooney | that would have been the compute service then not the comptue node | |
| 17:25:08 | bauzas | always | |
| 17:25:14 | sean-k-mooney | the usecase says "Operators wish to apply the same host aggregate to compute nodes in multiple cells." | |
| 17:25:25 | bauzas | that was always the service | |
| 17:25:42 | bauzas | because hostname key is about the n-cpu service | |
| 17:25:55 | sean-k-mooney | the spec does not say service. an by the way im agreein gi tought this was also the service not the comptue node | |
| 17:26:45 | sean-k-mooney | so the problem is context the pec that created the db table repataly says compute node | |
| 17:26:55 | sean-k-mooney | so maybe the spec is wrong or not what we implemnted | |
| 17:27:15 | bauzas | "Aggregates may be applied to compute nodes in multiple cells and are a global concept." | |
| 17:27:20 | bauzas | this phrase is wrong | |
| 17:27:40 | sean-k-mooney | right but this spec neverf mention compute services | |
| 17:27:40 | bauzas | Placement aggregates relate to resource providers | |
| 17:27:43 | sean-k-mooney | only compute nodes | |
| 17:27:52 | bauzas | Nova aggregates related to nova-compute services | |
| 17:27:57 | sean-k-mooney | so if i read the spec the intent seams to be to mapp compute nodes | |
| 17:28:00 | bauzas | relate* | |
| 17:28:20 | bauzas | sean-k-mooney: the intent of that spec was to put the aggregates table to the API DB | |
| 17:28:28 | bauzas | not about desiging placement aggs | |
| 17:29:58 | sean-k-mooney | i am not talking about placment aggs at all | |
| 17:30:04 | sean-k-mooney | just nova | |
| 17:30:18 | sean-k-mooney | im also reading our newton docs https://docs.openstack.org/nova/newton/aggregates.html | |
| 17:30:26 | sean-k-mooney | whcih again talk about comptue nodes not compute services | |
| 17:30:54 | bauzas | sure | |
| 17:31:07 | bauzas | we also talk about ephemeral storage, remember ? :) | |
| 17:31:29 | bauzas | but the fact is, I know how scheduling works and I know what we return | |
| 17:31:51 | sean-k-mooney | yep but i know we have downstream docs about addign ironic does to host aggretes by uuid and i have seen that used in other place too | |
| 17:32:06 | bauzas | we return a destination which is a hostame for calling the RPC API | |
| 17:32:54 | sean-k-mooney | we return more then that | |
| 17:33:31 | sean-k-mooney | or at least we provide more then that to the filters | |
| 17:33:40 | sean-k-mooney | its not so much about what we return to the conductor | |
| 17:34:33 | opendevreview | Merged openstack/nova master: Support evacuate with PCI in placement https://review.opendev.org/c/openstack/nova/+/854615 | |
| 17:37:01 | sean-k-mooney | so lookign at the newton code it is in terms of the compute service | |
| 17:37:10 | sean-k-mooney | we do https://github.com/openstack/nova/blob/808d36475103e373f1deb3344b6829ce68d6cdd5/nova/db/sqlalchemy/api.py#L533-L542 | |
| 17:38:32 | bauzas | sean-k-mooney; I need to stop today's work :( | |
| 17:39:19 | sean-k-mooney | bauzas: that ok | |
| 17:39:41 | sean-k-mooney | anyway we need to look at the patch carfullly as it sound more liek a feature then a bug and im not sure it work fully | |
| 17:40:06 | bauzas | sean-k-mooney: I was about to say it | |
| 17:40:19 | bauzas | looks to me a new behavioural support | |
| 17:40:23 | bauzas | hence not a bug | |
| 17:40:32 | bauzas | and I'd prefer us to design it correctly | |
| 17:40:39 | bauzas | as there could be corner cases | |
| 17:41:03 | sean-k-mooney | well i think there are defintly usecase for supprotiing ironic with host aggreates | |
| 17:41:23 | sean-k-mooney | but ya i think it needs a spec espcislly with the other ironic work that is in flight | |
| 17:41:56 | bauzas | I'll quickly add a comment on that patch into that direction | |
| 17:42:15 | bauzas | I won't use the -2 hammer yet as it's late | |
| 18:19:13 | opendevreview | Kirill proposed openstack/nova-specs master: new spec: support of vnc console for ironic https://review.opendev.org/c/openstack/nova-specs/+/863773 | |
| 18:31:18 | opendevreview | Dan Smith proposed openstack/nova master: Add virt/node module for stable uuids https://review.opendev.org/c/openstack/nova/+/863915 | |
| 18:31:19 | opendevreview | Dan Smith proposed openstack/nova master: Add get_available_node_uuids() to virt driver https://review.opendev.org/c/openstack/nova/+/863917 | |
| 18:31:19 | opendevreview | Dan Smith proposed openstack/nova master: Pass service ref to init_host(), if exists https://review.opendev.org/c/openstack/nova/+/863916 | |
| 18:31:20 | opendevreview | Dan Smith proposed openstack/nova master: WIP: Make resource tracker use UUIDs instead of names https://review.opendev.org/c/openstack/nova/+/863919 | |
| 18:31:20 | opendevreview | Dan Smith proposed openstack/nova master: Persist existing node uuids locally https://review.opendev.org/c/openstack/nova/+/863918 | |
| 18:31:21 | opendevreview | Dan Smith proposed openstack/nova master: WIP: Detect host renames and abort startup https://review.opendev.org/c/openstack/nova/+/863920 | |
| 18:31:53 | dansmith | gibi: I still need to do the undelete explicit test in ^ so I marked that penultimate one as WIP with a note, I just want to get a bunch of work pushed up for insurance | |
| 18:32:04 | gibi | dansmith: ack | |
| 18:42:28 | artom | Besides the tox thing (that is now fixed?) there are no further known gate issues? | |
| 18:42:57 | sean-k-mooney | beseide the multiple tox thigns not that comes to mind | |
| 18:43:25 | sean-k-mooney | artom: are you seing somthing or wonderifn if you can recheck | |
| 18:43:32 | artom | Latter | |
| 18:43:41 | sean-k-mooney | then i belive yes | |
| 18:43:57 | artom | ✊✊ | |
| 18:44:02 | sean-k-mooney | on master at least | |
| 18:44:22 | sean-k-mooney | i dont know if all the pin patches are mergd on all the sable branches | |
| 18:45:04 | gmann | pin patches only required on python-novaclient and others are taken care by the pinning in common place https://review.opendev.org/q/I442568a5f5900e593feb2b5527109e0aa79e5aa7+status:open | |
| 18:45:58 | gmann | nova stable branches are all green https://review.opendev.org/q/topic:tox4-pin-testing | |
| 18:46:12 | sean-k-mooney | cool ill appove the client patches now so | |