Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2022-02-02
11:44:28 MrClayPole When I live migrate any instances that has more than 1 volume attached the windows VMs sometimes lose access to their volumes and then hang. The logging stops on the host until after they are restarted. I'm not sure how to troubleshoot this further or if live migration of instances with more than one disk is supported? Any help would be appreciated.
11:45:12 MrClayPole The VMs in question are using NFS storage via the NetApp cinder driver
12:01:22 ralonsoh gibi, hi, this is about https://bugs.launchpad.net/neutron/+bug/1959749
12:01:34 ralonsoh please check https://review.opendev.org/c/openstack/neutron/+/468982/5/neutron/services/qos/drivers/manager.py
12:02:03 ralonsoh "openstack network qos rule type list" will return only those rules accepted by all loaded mech drivers only
12:02:37 ralonsoh You can create a rule of any type, but you won't be able to assign it
12:08:17 gibi ralonsoh: do you mean assigning it to a port?
12:08:29 ralonsoh if the port is bound
12:08:45 ralonsoh because that means you'll call the mech driver qos extension
12:09:20 opendevreview Stephen Finucane proposed openstack/nova master: docs: Follow-ups for cells v2, architecture docs https://review.opendev.org/c/openstack/nova/+/827336
12:10:38 gibi ralonsoh: I have a port with a qos policy that uses the min pps rule type and I can boot a VM with that port and the port is bound successfully
12:11:01 ralonsoh gibi, in your deployment, what backend are you using?
12:11:07 ralonsoh or backends
12:12:15 gibi ovs and sriov
12:12:18 opendevreview Dmitrii Shcherbakov proposed openstack/nova master: Introduce remote_managed tag for PCI devs https://review.opendev.org/c/openstack/nova/+/824834
12:12:19 opendevreview Dmitrii Shcherbakov proposed openstack/nova master: Bump os-traits to 2.7.0 https://review.opendev.org/c/openstack/nova/+/826675
12:12:19 opendevreview Dmitrii Shcherbakov proposed openstack/nova master: [yoga] Add support for VNIC_TYPE_SMARTNIC https://review.opendev.org/c/openstack/nova/+/824835
12:12:20 opendevreview Dmitrii Shcherbakov proposed openstack/nova master: Filter computes without remote-managed ports early https://review.opendev.org/c/openstack/nova/+/812111
12:12:27 gibi ovs support the min packet rate rule
12:12:38 ralonsoh gibi, yes, right, as I weas guessing
12:12:52 ralonsoh the problem is how we build the rule type set, as I commented in the bug
12:13:04 gibi looking...
12:13:13 ralonsoh we return the intersection of all mech driver supported types
12:13:18 ralonsoh instead of returning the union
12:13:30 sean-k-mooney gibi: ovn does not support qos fully yes i think
12:13:42 gibi sean-k-mooney: yes, I uses ovs :)
12:13:49 sean-k-mooney so if you are using the new default it might not work unless you reverted to ml2/ovs
12:13:51 sean-k-mooney ok
12:13:57 gibi sean-k-mooney: I reverted, yes :)
12:14:08 ralonsoh sean-k-mooney, we do support qos in OVN
12:14:10 ralonsoh fully
12:14:12 gibi ralonsoh: ohh so because I have sriov I cannot see the min pps
12:14:18 ralonsoh gibi, right
12:14:23 gibi ralonsoh: let me check that in another devstack that only hase ovs but not sriov
12:14:27 ralonsoh I think this should be reconsidered in the API
12:14:32 sean-k-mooney ralonsoh: that new this cycle right
12:14:33 gibi ralonsoh: I agree
12:14:47 ralonsoh I'll push a patch today
12:14:53 gibi ralonsoh: this sounds incorrect that I can boot with a qos rule but the rule type list does not show it
12:14:59 gibi ralonsoh: thank you!
12:15:01 ralonsoh sean-k-mooney, that was supported since wallaby
12:15:05 ralonsoh and in D/S in OSP16
12:15:32 sean-k-mooney oh ok
12:16:33 sean-k-mooney ralonsoh: this is not the first time this api design choice has come up
12:16:49 ralonsoh sean-k-mooney, yeah... I think the current implementation is wrong
12:16:58 sean-k-mooney well i ment in general
12:17:11 sean-k-mooney neutron also has the same problem with vlan transparncy
12:17:34 sean-k-mooney to work around that vlan transpace was set to yes for sriov
12:17:44 sean-k-mooney event though it really done not support it properly
12:18:04 ralonsoh the aim of the API is to return only what is supported by all drivers
12:18:06 sean-k-mooney i think in general neutron need to list the capyablity per ml2/driver
12:18:19 ralonsoh for example: https://review.opendev.org/q/3299cdffae5cd7196a1676da103da5e2e413ec21
12:18:23 sean-k-mooney ralonsoh: ya i know but that has never felt useful to me
12:18:33 ralonsoh it was changed before and then reverted
12:18:57 sean-k-mooney the api shoudl idally list the qos polices per driver
12:18:59 ralonsoh sean-k-mooney, then what we can do is to create another API call
12:19:11 ralonsoh return all_supported_qos_types
12:19:15 ralonsoh or something similar
12:19:52 sean-k-mooney perhaps
12:20:10 ralonsoh I'll propose a new API
12:22:14 sean-k-mooney i still think having the new api return a dictonaly keyd by either the driver or vnic-type with the support qos polices as the values would be the way to organsie that api but just add a property to the exising one to list all is less work
12:23:00 ralonsoh I think we can do this with the current API, just adding a new parameter to the CLI call
12:23:04 sean-k-mooney the issue i have with all_supported_qos_types is that you cant tell if a port you create will work with any specific policy
12:23:30 sean-k-mooney ralonsoh: actully at the end of the day what we really need is schduler support
12:23:37 ralonsoh yes
12:24:06 gibi minimum pps / bw has scheduler support :)
12:24:12 sean-k-mooney neutron need to use traits to model which host support which polcies and nova need to shcdule the prot to such a host based on the requst
12:24:24 sean-k-mooney gibi: it does but im thinking for dscp ectra
12:24:35 gibi yeah for dhcp it is a different game
12:24:40 sean-k-mooney e.g. the non quntitive qos
12:24:50 sean-k-mooney its a fair point about min*
12:25:01 sean-k-mooney those are ahead of the game
12:25:28 stephenfin sean-k-mooney: Care to finally get these in? It's only been 16 months :) https://review.opendev.org/c/openstack/nova/+/705792/ https://review.opendev.org/c/openstack/nova/+/754448/
12:25:29 gibi and supporting the non quantitative is problematic by the resourceless request group problem in placement
12:26:02 sean-k-mooney stephenfin: im looking at https://review.opendev.org/c/openstack/nova/+/814562 now but i can look at them after
12:26:16 stephenfin Okay, sweet. ty :)
12:26:45 sean-k-mooney stephenfin: o nnueton refactoring ya ill review those too
12:35:45 sean-k-mooney ah you have a follow up for the cells doc cool i was going to ask if you were doing a new reviesion
13:12:10 gibi ralonsoh: I'm not sure I understand the reason of the wontfix on https://bugs.launchpad.net/neutron/+bug/1959749
13:16:06 gibi ralonsoh: you #1) point is what I would need to work. So that the rule type list returns all rule types not just rule types that are supported by every configured driver
13:32:09 ralonsoh gibi, sorry, I don't know why I set this flag
13:32:25 ralonsoh confirmed, this is the correct value
13:32:37 gibi ralonsoh: that is better, thanks :)
13:48:09 admin1 hi guys .. is there a way to "transfer ownership" of an instance from one project to another ?
14:04:57 opendevreview yuval proposed openstack/nova master: Lightbits LightOS driver https://review.opendev.org/c/openstack/nova/+/821606
14:58:51 opendevreview Dmitrii Shcherbakov proposed openstack/nova master: Document remote-managed port usage considerations https://review.opendev.org/c/openstack/nova/+/827513
16:45:59 gibi if somebody wants a change of (code) scenery then I can suggest looking at the placement code review series to support any-traits queries in microvarsion 1.39. The series starts here https://review.opendev.org/c/openstack/placement/+/825846/3 :)
17:01:15 opendevreview Merged openstack/nova master: docs: Add a new cells v2 document https://review.opendev.org/c/openstack/nova/+/814562
17:15:10 melwitt gibi: I will look at some new code scenery :)
17:27:53 dmitriis gibi, sean-k-mooney: mostly been getting unrelated gate failures so I am waiting for some rechecks to complete. I made a functional change to the patch that introduces the remote_managed tag here https://review.opendev.org/c/openstack/nova/+/824834/8/nova/pci/devspec.py#322 to include a check for the presence of a serial number when a device is
17:27:53 dmitriis tagged as remote_managed which is something I overlooked in the previous iteration and updated testing to reflect that.
17:28:38 dmitriis I started working on the docs and started a doc review but most of the docs will be in Neutron under the OVN driver guide similar to how it's done today with OVS hardware offload.
17:42:44 gibi melwitt: thanks! :)
17:43:12 gibi dmitriis: I will read back tomorrow I have to go now
17:43:32 dmitriis gibi: np, thanks a lot for the help so far
18:13:35 sean-k-mooney dmitriis: ack
18:14:07 sean-k-mooney dmitriis: im looking at some downstream stuff currently but ill try to take a look proably tomorow at this point
18:14:48 sean-k-mooney dmitriis: most of the doc make sense for the neutorn guide but we shoudl detail how to use the remote managed flag ectra in nova
18:16:56 dmitriis sean-k-mooney: ack, ta.
18:16:57 dmitriis sean-k-mooney: I currently describe some of it in the latest doc change and reference the option docstring but I can expand the description in the docs themselves too.

Earlier   Later