Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2022-09-30
14:18:23 frickler bauzas: where do we have that? the use case is securing the initial SSH connection to a fresh instance, yes
14:59:41 bauzas frickler: https://docs.openstack.org/api-ref/compute/?expanded=show-keypair-details-detail#show-keypair-details
14:59:58 bauzas when you import a pubkey, then it creates the fingerprint
15:02:42 frickler bauzas: ah, but the issue is about the hostkey, not the user key. cloud-init can output it on the console but I would want a nicer solution https://stackoverflow.com/questions/66658406/openstack-how-to-find-out-vms-key-fingerprint-before-first-ssh-session
15:05:54 frickler the other option would be to have something similar like AWS IAM host roles and via that give cloud-init credentials with which it could talk to designate
15:09:26 frickler but maybe you are right and this is more an issue for neutron than for nova
15:11:31 clarkb frickler: yes mordred brought this up probably 8 years ago at this point. I think the idea then was maybe to have the hypervisor scan the hostkey since it can talk directly to the correct interface then report that back through the api
15:14:30 frickler clarkb: ah, scanning from the outside instead of relying on some action from the instance would make this more generally useful, that's a good point, too
15:26:02 bauzas frickler: hah, about the hostkety
15:26:12 bauzas I mean the compute node
15:26:30 bauzas then, as I said, not a nova usecase IMO
15:31:15 clarkb bauzas: why isn't that a nova use case?
15:31:28 clarkb nova is the only entity in a position to accurately query the info
15:31:54 bauzas clarkb: that's port-dependent on the host, right?
15:32:31 clarkb yes, you'd need to make a network request over the correct interface/port to accurately retrive the information
15:32:55 clarkb and you can't traverse additional network segments or you lose the accuracy
15:33:13 bauzas correct
15:33:24 bauzas so you can't rely on nova itself
15:33:52 bauzas but I'm open to discuss the usecase in a spec
15:33:56 bauzas or even at the PTG
15:34:15 bauzas I guess this is for blessing the instance ssh connection ?
15:34:23 clarkb ya ultimately I don't really care what implements it, but I do think being able to get the host key from nova show and be able to trust it is a very useful feature
15:34:41 clarkb bauzas: yes, it is so that you don't have to take a leap of faith the first time you make an ssh connection that you are not being mitm'd
15:35:09 bauzas clarkb: but then the connection to the port is managed thru neutron
15:35:47 bauzas but maybe that's something we can sort out at the PTG
15:35:48 clarkb but it is a host attribute. Your ports may come and go but the hostkey doesn't necessarily change
15:35:52 bauzas I understand the usecase
15:36:02 clarkb it may ultimately require cooperation between the compute and network layers
15:36:04 bauzas this is OS-specific tho
15:36:14 clarkb well it is ssh specific
15:36:16 bauzas actually no
15:36:18 bauzas yeah that
15:36:28 clarkb windows and solaris and linux and osx can all run an sshd
15:36:33 bauzas yeah
15:37:17 bauzas anyway, I'm not eventually against the usecase, I think we can agree it's a valid one
15:37:24 bauzas now the problem is how to get this
15:37:27 bauzas and how to present this
15:44:15 clarkb yup, I think the raeson it hasn't been done is that it is complicated
22:06:01 atmark how to properly to delete a VM stuck on BUILD status? openstack server delete throws no server with name or ID exists
22:07:07 sean-k-mooney that is the correct way if we have gotten to the point of creating a instance object in the db
22:07:30 sean-k-mooney if its stuck before that point then there should be no recored to delete
22:08:14 atmark server list still list the instance
22:08:35 atmark how can I get rid of the record?
22:08:58 sean-k-mooney you should be able to just do a server delete
22:09:16 sean-k-mooney if you cant then you have a broken deployment and you need to triage why the request is failing
22:09:45 sean-k-mooney did you try deleteing with the uuid
22:09:52 sean-k-mooney rahter then the name
22:10:12 sean-k-mooney if that fails you should check the nova-api to see if there is an internal excption
22:10:37 atmark i'm deleting by uuid but response is no name or ID or exists
22:11:07 atmark i think this an old instance from last year
22:11:19 sean-k-mooney ack there is obviouly an error internally without knowing what that is there is not much more i can suggest
22:11:32 sean-k-mooney if you take the request id you shoudl be able to find it in the api logs
22:11:47 atmark i'm doing cleanup atm and just caught one on weird state
22:12:07 sean-k-mooney its possible that it exists in teh api db but not in the cell db
22:12:22 sean-k-mooney if you do a server show does it have a host
22:12:29 atmark no
22:12:57 sean-k-mooney ok so its before the schuiler has selected one most likely
22:13:43 atmark https://paste.openstack.org/show/bXZv2CMnmfdGeswoUKUC/
22:14:23 atmark VM is octavia amphorae
22:14:48 sean-k-mooney --all is all tenants
22:15:04 sean-k-mooney is that owned by your current project
22:15:18 johnsom Yep, the instance is owned by the Octavia service account.
22:15:49 sean-k-mooney no what i mean is when openstack server delete 6fb27c67-539c-4525-8747-e26487d15e75 is run
22:15:49 atmark yes, i have admin access to all tenants
22:15:52 sean-k-mooney what user is that
22:16:04 sean-k-mooney ok so thats an admin user
22:16:41 sean-k-mooney i think you should be able to delete the vm then even if your current token is not for the correct project due to admin right if you are not using new policy
22:17:30 sean-k-mooney the server show is implying that its not in the current project
22:17:46 sean-k-mooney so i was wondering if the delete was failing for the same reason
22:18:04 johnsom There is a --all-projects for the server delete command too, just like for list.
22:18:13 sean-k-mooney i tought admins could bypass that check
22:18:25 johnsom I thought so too honestly
22:19:04 sean-k-mooney its been a while since i have done that so cant recall if you need to set the project id somehow or not
22:20:17 atmark openstack server --os-project-id 848fe125a93a408ba8a8044fb87e9cdf delete 6fb27c67-539c-4525-8747-e26487d15e75
22:20:48 atmark 848.. is tenant ID of octavia
22:22:19 sean-k-mooney that would require your current user to have the member or admin role on that project for keystone to be able to issue the token
22:22:40 sean-k-mooney the uuid shoudl be enough if your an admin
22:22:51 sean-k-mooney --all-project is only required to delete by name
22:23:06 sean-k-mooney on a differnt project
22:23:15 atmark i'm able to list VMs `openstack server list --os-project-id 848fe125a93a408ba8a8044fb87e9cdf`
22:23:27 sean-k-mooney ack
22:23:33 sean-k-mooney and i assume the delete didnt work
22:23:38 atmark it didn't
22:23:51 sean-k-mooney you could try --force
22:23:56 sean-k-mooney but that is not for this usecase
22:24:10 sean-k-mooney its for forceing the delete now if you have soft delete enabled
22:24:26 sean-k-mooney you could also try reset-state
22:24:28 sean-k-mooney then delete
22:24:41 sean-k-mooney so reset the state to error
22:24:44 sean-k-mooney then delete form error
22:25:46 atmark throws an no ID found on reset-state
22:26:42 sean-k-mooney ya so i think what is happenign is there is a record in teh api db
22:26:58 sean-k-mooney but there is no cell mapping for the instnace and no recored in teh cell db
22:27:18 sean-k-mooney reset-state woudl try and update the instance in the cell db
22:27:32 sean-k-mooney but that fails because it was not created
22:28:01 atmark this is what's in the db https://paste.openstack.org/show/bbS8V1bZK0VcFO7WBD7p/
22:28:15 sean-k-mooney if you have db access your could confirm that by checkign the api db to see if the build request exists and/or instance_mapping
22:28:45 sean-k-mooney ok thats in which db
22:28:48 sean-k-mooney cell0
22:28:51 sean-k-mooney or api

Earlier   Later