Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-08-08
15:05:01 mriedem you mean your customer would
15:05:12 mriedem let's front-load validate all incoming IDs because of one red hat customer!
15:05:13 mriedem yay!
15:06:02 artom I actually don't think there's a customer
15:06:24 artom It's more a matter of aligning with our api-ref docs: https://docs.openstack.org/api-ref/compute/?expanded=detach-interface-detail#detach-interface
15:06:29 artom Which state a 404 could happen
15:06:35 artom ... or we should change the doc
15:06:43 mriedem technically the 404 is for the instance not being found
15:06:47 mriedem the port not being found is a 400
15:06:55 mriedem according to what i've learned over the years from cdent and edleafe
15:06:57 mriedem and the api sig
15:07:09 mriedem missing things that aren't the main resource are 400s
15:07:29 edleafe mriedem is a good student :)
15:07:39 mriedem i only had to ask about 10 times
15:08:34 artom Ah, so since the api-ref doens't state a 400 is a possible response, we're technically good?
15:08:40 artom Still feels weird
15:09:12 artom Although I suppose there's precedent
15:09:17 artom With live-migration, for example
15:12:19 stephenfin efried: You should look at this too, I'd say https://review.opendev.org/#/c/666032/
15:13:15 efried dansmith: mriedem: FYI I'm +2 all the way up on aggregate isolation https://review.opendev.org/#/c/671075/
15:13:29 efried stephenfin: I've been avoiding that one
15:13:34 efried I never understood the original spec
15:14:01 efried but I'll have a look at some point if you get desperate.
15:14:48 stephenfin efried: TBH, it's what I'm doing anyway so it really just needs a sanity check
15:16:03 efried point is, if I do manage to review the code, I'll be doing it against whatever the spec says. So if I rubber-stamp the spec based on the fact that that's what you're coding... https://previews.123rf.com/images/sudowoodo/sudowoodo1708/sudowoodo170800024/84484656-ouroboros-symbol-snake-eating-its-own-tail-modern-alchemy-logo-vector-illustration-.jpg
15:30:53 openstackgerrit Shilpa Devharakar proposed openstack/nova-specs master: Update spec: filtering of alloc candidates by forbidden aggregates https://review.opendev.org/675384
15:37:06 aspiers how can I debug placement returning no allocation candidates on devstack, when debug is already set to True and showing nothing useful?
15:37:14 aspiers GET /placement/allocation_candidates?limit=1000&required=%21COMPUTE_STATUS_DISABLED&resources=DISK_GB%3A20%2CMEMORY_MB%3A2048%2CVCPU%3A1
15:38:27 aspiers I've already gone through the three resource classes via openstack resource provider inventory show $rp_uuid
15:38:51 aspiers each one is showing a total and max_unit greater than what is being asked for
15:38:55 efried aspiers: You'll want to show usages
15:39:09 aspiers isn't that what I just did? ^^
15:39:15 efried You showed inventories
15:39:39 efried usages: "how much of it is consumed"
15:39:39 efried inventories: "how big is it"
15:39:44 aspiers ok
15:40:34 aspiers surely placement should offer to DEBUG-log details of why it's rejecting candidates?
15:40:49 efried that's not nearly as easy as you would think
15:40:56 aspiers "why am I getting NoValidHost" is the #1 FAQ I've ever heard from operators in 7 years of OpenStack
15:41:02 efried ditto
15:41:03 aspiers of course it wouldn't be ;-/
15:41:19 aspiers otherwise someone would have done it, I guess
15:41:30 efried And this would be an absolutely overwhelming amount of data
15:41:38 efried but mainly
15:41:38 aspiers well sure
15:41:43 aspiers that's why it would be at DEBUG level
15:41:50 efried we reject the vast majority of things based on a database query
15:42:10 aspiers well even logging what's left after that would be a start
15:42:18 efried we do have some debug logs for that
15:42:29 efried and the fact that you're not seeing any means... we've done all the filtering at the db level :)
15:42:46 aspiers ok
15:43:10 aspiers looks like I've used 20 out of 35 DISK_GB and am asking for another 20
15:43:17 efried that'd do it
15:43:17 aspiers next question is why is it so damn low
15:43:30 efried you can show allocations
15:43:55 aspiers "openstack resource provider allocation show $rp_uuid" shows nothing
15:44:07 efried allocation show takes an instance UUID
15:44:16 efried I think you should be able to do allocation list
15:44:17 openstackgerrit Matt Riedemann proposed openstack/nova master: Fix rebuild of baremetal instance when vm_state is ERROR https://review.opendev.org/523559
15:44:18 mriedem melwitt: addressed your comments i think ^
15:44:22 mriedem efried: or a migration uuid :)
15:44:37 edleafe artom: mriedem: I just looked at the API you were discussing earlier for missing ports, and I think it *should* be a 404, since the resource you are requesting is the port interface. 400 would be correct if you were searching for a server that had that port.
15:44:48 mriedem jroll: TheJulia: could use ironic subteam re-review on this https://review.opendev.org/523559
15:45:12 mriedem edleafe: both are on the path
15:45:16 mriedem server_id and port_id
15:45:49 mriedem 404 cool for all things on the path then?
15:45:56 melwitt mriedem: that's a blast from the past. will have to reload a lot of context when I go back to review
15:46:42 edleafe mriedem: is this the GET or DELETE call?
15:46:46 mriedem edleafe: anyway, it's essentially dead code anyway
15:46:48 mriedem DELETE
15:47:07 mriedem https://docs.openstack.org/api-ref/compute/#detach-interface
15:47:44 edleafe The call doesn't delete the server. The resource it is concerned with is the port. If that resource doesn't exist, then 404 is correct
15:47:46 aspiers efried: I mean, why is the *total* inventory so damn low, not the usage
15:47:56 aspiers I guess that's in n-cpu
15:48:04 efried yes
15:48:07 mriedem edleafe: ack
15:48:11 edleafe Of course, if you subscribe to the idempotent DELETE school of thought, 204 would be correct
15:48:20 efried aspiers: I assume libvirt driver?
15:48:21 mriedem like i said, it's dead code at this point unless we do a proxy neutronv2.api.show_port call on that port_id
15:48:29 openstackgerrit Shilpa Devharakar proposed openstack/nova-specs master: Update spec: filtering of alloc candidates by forbidden aggregates https://review.opendev.org/675384
15:49:05 openstackgerrit Balazs Gibizer proposed openstack/nova master: allow getting resource request of every bound ports of an instance https://review.opendev.org/655110
15:49:14 aspiers efried: virsh shows no pools
15:49:31 efried I don't know what that means
15:49:37 aspiers virsh pool-list
15:49:38 efried look at _get_local_gb_info
15:49:45 aspiers ok thanks
15:49:51 efried what's your [libvirt]images_type?
15:50:26 mriedem artom: you know, could just move this validation to the api https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L6373
15:50:34 sean-k-mooney aspiers: i dont think openstack uses libvirt storage pools
15:50:42 aspiers sean-k-mooney: yeah that was a guess
15:50:43 mriedem compute is already checking the info cache before doing anything with the port at the neutron level
15:51:29 sean-k-mooney aspiers: mybe we might for lvm image backend? but im pretty sure tis not used for qcow/raw/flat iamge type
15:51:59 aspiers libvirt.images_type = default
15:52:03 efried aspiers: assuming your libvirt.images_type is unset or anything other than lvm/rbd, it'll be looking at the free space on your instances_path
15:52:12 aspiers that makes more sense
15:54:24 efried aspiers: ...which looks like it'll be by default whatever file system your nova installation is
15:54:57 efried if your nova is in /foo/bar/nova, your instances_path will by default be /foo/bar/instances, I think.
15:55:13 efried so
15:55:23 efried df -h /foo/bar ought to tell you how much space it thinks you have.
15:55:41 openstackgerrit Balazs Gibizer proposed openstack/nova master: Pass network API to the conducor's MigrationTask https://review.opendev.org/655111

Earlier   Later