| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-07-25 | |||
| 13:36:16 | jaypipes | bauzas: alloc_reqs_by_rp_uuid is a map of compute node UUID to list of allocation requests | |
| 13:36:19 | dtantsur | yeah, I suspect so | |
| 13:36:27 | dtantsur | the final word is for TheJulia though :) | |
| 13:38:58 | TheJulia | I think so | |
| 13:40:52 | bauzas | jaypipes: sure, but what if the first compute node UUID is not corresponding to the first host from the ones given by get_sorted_hosts() ? | |
| 13:41:08 | jaypipes | bauzas: it's a dict... | |
| 13:41:30 | jaypipes | bauzas: not sure what you're asking, sorry... | |
| 13:42:14 | bauzas | jaypipes: oh sorry, just saw edleafe's comment | |
| 13:42:23 | bauzas | https://review.openstack.org/#/c/483566/10/nova/scheduler/filter_scheduler.py@202 | |
| 13:42:52 | bauzas | we're getting a list of allocation requests *per* compute node | |
| 13:43:28 | jaypipes | bauzas: right | |
| 13:43:46 | sdague | mriedem: what's up now? | |
| 13:43:47 | bauzas | jaypipes: I was confused by your TODO in https://review.openstack.org/#/c/483566/10/nova/scheduler/filter_scheduler.py@270 | |
| 13:43:48 | edleafe | bauzas: don't feel bad - I had to read that over several times before it made sense to me, too | |
| 13:44:00 | jaypipes | bauzas: that alloc_reqs_by_rp_uuid is a dict, keyed by compute node UUID, of allocation requests that contain that compute node | |
| 13:44:07 | bauzas | jaypipes: so the same compute node could have more than one allocation request ? | |
| 13:44:23 | bauzas | just tbc | |
| 13:44:32 | edleafe | bauzas: there could be several allocs for a given compute node. The TODO is about being smarter about picking which one to use | |
| 13:44:56 | bauzas | the real problem I had is that allocation_candidates just was discussed when I was in and out, and now I'm paying the price by giving you silly comments :/ | |
| 13:44:58 | jaypipes | bauzas: yep. imagine in the future, nested r-p's there may be dozens of allocation requests that partially allocate resources on the compute node. | |
| 13:45:09 | jaypipes | bauzas: same with things like shared storage | |
| 13:45:25 | jaypipes | bauzas: not a problem, don't worry about it. | |
| 13:45:25 | mriedem | sdague: https://review.openstack.org/#/c/215385/ | |
| 13:45:28 | bauzas | jaypipes: ah, right | |
| 13:45:39 | bauzas | nested RPs is understandable by me | |
| 13:46:01 | bauzas | because you could consume more than one thing for a specific compute node | |
| 13:46:10 | bauzas | ah, and I see your point with shared storage | |
| 13:46:22 | jaypipes | bauzas: so, imagine a compute host with 2 NUMA nodes and 4 PCI devices, affined to the different NUMA cells. We might potentially have a lot of allocation requests (candidates) for those child providers plus resources on the compute host itself (like VCPU, etc) | |
| 13:46:30 | bauzas | we could potentially have the compute node having allocation request for local resources + shared resource ? | |
| 13:46:42 | jaypipes | bauzas: yep | |
| 13:46:46 | bauzas | jaypipes: that, I'm clear for nested RPs | |
| 13:46:53 | bauzas | okay, nevermind all my comments then | |
| 13:46:56 | bauzas | I'm on board now :) | |
| 13:46:57 | jaypipes | bauzas: one alloc request might have local disk, another consuming from shared disk. | |
| 13:47:00 | sdague | mriedem: ah, that one. Approved | |
| 13:47:09 | jaypipes | bauzas: of course, that's not currently possible, but you see the idea | |
| 13:47:09 | bauzas | jaypipes: thanks for explaning it | |
| 13:47:14 | jaypipes | no worries | |
| 13:48:08 | mriedem | dtantsur: TheJulia: what is the story on https://review.openstack.org/#/c/468353/ ? | |
| 13:48:14 | mriedem | required for pike? | |
| 13:49:37 | sdague | melwitt: https://bugs.launchpad.net/nova/+bug/1706310 that's addressed by your new quota counting? | |
| 13:49:38 | openstack | Launchpad bug 1706310 in OpenStack Compute (nova) "Quota usage is not updated if an instance is deleted while booting" [Undecided,New] | |
| 13:49:42 | dtantsur | mriedem: I think this is nice-to-have, but still optional | |
| 13:49:56 | dtantsur | I guess it may cause some longer discussions | |
| 13:49:58 | TheJulia | mriedem: the consensus is not required this cycle, it just allows for greater use case flexibility ultimately shipping a source IP address off to cinder in the event there is firewalling present. | |
| 13:50:03 | bauzas | jaypipes: +2d FTW | |
| 13:50:25 | sdague | mriedem: that one is a little less straight forward, and honestly probably a good ptg conversation | |
| 13:50:29 | sdague | it adds virt driver methods | |
| 13:50:54 | mriedem | right i'd like to defer it | |
| 13:51:00 | sdague | mriedem: agreed | |
| 13:51:01 | TheJulia | sdague: good point | |
| 13:51:07 | mriedem | as something else - ironic bfv + network-managed storage | |
| 13:51:30 | jaypipes | edleafe: I can clean those nits up later. at this point, I'm nervous about kicking off another gate recheck odyssey ;) | |
| 13:52:00 | cdent | jaypipes: you might like these then: https://review.openstack.org/#/c/486237/ | |
| 13:52:17 | edleafe | jaypipes: totally agree | |
| 13:52:47 | ildikov | mriedem: oh ok, tnx for the info :) | |
| 13:56:47 | mriedem | sdague: about that quota bug, yes counting quotas would address that, but also, it could already be a known issue fixed in ocata, | |
| 13:56:52 | mriedem | the reporter didn't say what version they were using | |
| 13:56:55 | mriedem | i'm leaving comments | |
| 14:06:57 | openstackgerrit | Jan Gutter proposed openstack/nova master: Add VIFHostDevice support to libvirt driver https://review.openstack.org/486426 | |
| 14:09:10 | openstackgerrit | Jan Gutter proposed openstack/nova master: Add VIFHostDevice support to libvirt driver https://review.openstack.org/486426 | |
| 14:11:52 | mriedem | sdague: the dependencies for the cells fleetify devstack change are all merged https://review.openstack.org/#/c/477556/ | |
| 14:12:00 | mriedem | sdague: thoughts on putting that in right now? | |
| 14:12:12 | mriedem | or, | |
| 14:12:18 | mriedem | maybe i should run a novaclient change against it quick? | |
| 14:12:29 | dansmith | damn the torpedos | |
| 14:12:34 | mriedem | i think it blew up novaclient functional ci on the first go around | |
| 14:12:45 | mriedem | b/c it wasn't dependent on the quotas stuff | |
| 14:12:48 | mriedem | and that job is single tenant | |
| 14:13:42 | mriedem | i'll push a novaclient test change quick just to be sure | |
| 14:14:49 | dansmith | cool | |
| 14:15:10 | openstackgerrit | Matt Riedemann proposed openstack/python-novaclient master: DNM: test conductor fleet with single-tenant job https://review.openstack.org/487091 | |
| 14:43:29 | openstackgerrit | Jan Gutter proposed openstack/nova master: Netronome SmartNIC Enablement https://review.openstack.org/483459 | |
| 14:43:41 | openstackgerrit | Sean Dague proposed openstack/nova master: Clean up second ephemeral disk https://review.openstack.org/341008 | |
| 14:49:14 | openstackgerrit | Jay Pipes proposed openstack/nova master: hardware offload support for openvswitch https://review.openstack.org/398265 | |
| 14:51:10 | openstackgerrit | Jan Gutter proposed openstack/nova master: Netronome SmartNIC Enablement https://review.openstack.org/483459 | |
| 14:52:33 | openstackgerrit | Jan Gutter proposed openstack/nova master: Netronome SmartNIC Enablement https://review.openstack.org/483459 | |
| 14:54:17 | mriedem | jangutter: question in https://review.openstack.org/#/c/486426/3/nova/virt/libvirt/vif.py | |
| 14:55:52 | jangutter | mriedem: let me see if the downstream code checks that. It hadn't occured to me to do input validation there! | |
| 14:56:17 | mriedem | well, one would probably never think to check it if only testing against a single configuratoin | |
| 14:56:22 | mriedem | *configuration | |
| 14:56:46 | sdague | mriedem: we had to revert it once before right? | |
| 14:56:56 | sdague | maybe an ironic issue | |
| 14:57:04 | mriedem | sdague: yes, but for two issues, | |
| 14:57:17 | mriedem | 1. notifications | |
| 14:57:22 | sdague | if you all tell me it's good to go, I'm fine doing it | |
| 14:57:26 | mriedem | 2. it didn't depent on the quotas change | |
| 14:57:34 | mriedem | sdague: i'm just running tests before we do | |
| 14:57:49 | sdague | sfinucan: https://bugs.launchpad.net/nova/+bug/1704458 that bug on your radar? | |
| 14:57:50 | openstack | Launchpad bug 1704458 in OpenStack Compute (nova) "The use_ipv6 flag not only influences nova networking" [Undecided,New] | |
| 14:58:06 | mriedem | sdague: sfinucan is now stephenfin and stephenfin is out this week | |
| 14:58:13 | sdague | ok | |
| 14:58:15 | dansmith | sdague: ironic and notification issues are fixed, and quotas are merged so those are the things we knew about and reverted for before | |
| 14:58:36 | sdague | mriedem: ok, so run the test you want to run, then poke me when it's successful | |
| 14:58:43 | moshele | mriedem: the generic is used in the old way https://github.com/openstack/nova/blob/master/nova/virt/libvirt/vif.py#L341-L343 so we can skip it for now | |
| 14:58:45 | dansmith | ack, thanks | |
| 14:59:30 | mriedem | moshele: what controls that getting called over the other? | |
| 14:59:35 | mriedem | for VIFHostDevice? | |
| 15:00:01 | moshele | mriedem: the vif type that send from neutron | |
| 15:00:18 | jangutter | mriedem, moshele: VIFHostDevice isn't used anywhere _yet_, it's not too difficult to fix though. | |