Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-06-05
14:40:38 sean-k-mooney mriedem: that is done on the compute node but yes
14:41:03 mriedem sean-k-mooney: this is done in the scheduler
14:41:31 sean-k-mooney the pci whitelist is on each the compute nodes and can be different
14:41:37 sean-k-mooney the alias is parsed on the schduler
14:42:22 sean-k-mooney there was a patch to cache this too at one point but i think it was blocked because peopel didnt feel it was enough of a perfomace imporvemnt
14:42:30 openstackgerrit Dan Smith proposed openstack/nova master: Make nova-next archive using --before https://review.opendev.org/661002
14:43:31 sean-k-mooney mriedem: https://review.opendev.org/#/c/427145/
14:44:12 sean-k-mooney mriedem: not sure if stephenfin still feels its not worth it
14:44:45 sean-k-mooney mriedem: but i certenly tried to get it landed 2 years ago :)
14:47:20 sean-k-mooney mriedem: by the way https://github.com/openstack/nova/blob/master/nova/scheduler/host_manager.py#L227 is loading the pci info frmo the database not parsing the whitelist
14:48:46 sean-k-mooney well that is not quite true its contcutiono the PciDeviceStats object form the compute.pci_device_pools
14:49:52 mriedem https://bugs.launchpad.net/nova/+bug/1831758
14:49:53 openstack Launchpad bug 1831758 in OpenStack Compute (nova) "passthrough_whitelist is parsed per compute node per scheduling request even though the whitelist doesn't change between requests" [Low,Confirmed]
14:50:01 mriedem constructing PciDeviceStats parses the whitelist
14:50:49 sean-k-mooney the whitelist does not have to be set on the the schduler node
14:51:29 sean-k-mooney and it is and i can be different per compute node but you are refering to https://github.com/openstack/nova/blob/master/nova/pci/stats.py#L66-L67
14:51:49 openstackgerrit Adrian Chiris proposed openstack/nova stable/rocky: Allow driver to properly unplug VIFs on destination on confirm resize https://review.opendev.org/661500
14:51:49 openstackgerrit Adrian Chiris proposed openstack/nova stable/rocky: Move get_pci_mapping_for_migration to MigrationContext https://review.opendev.org/661499
14:52:58 sean-k-mooney mriedem: i guess that makes sense on the compute node as this class is used in the pci manager but we should not need to do it in the schduler
14:55:15 sean-k-mooney mriedem: we proably should add a use_whitelist kwarg or something so we can disable this on the schduler
14:55:27 openstackgerrit Lee Yarwood proposed openstack/nova master: DNM dump image_meta when attempting to validate image signatures https://review.opendev.org/663348
14:56:44 sean-k-mooney the pci whitelist shoudl only be used in the compute node and the alias should only be used on teh api or schduler nodes.
15:00:44 efried can we not save the date stamp of the file?
15:00:57 efried or use fsnotify (or whatever that thing is called)?
15:07:46 mriedem we're parsing from the config option value which is not mutable
15:08:31 sean-k-mooney mriedem: and as i mention on the bug the config value is only used on teh compute node and not in the schduler
15:08:34 efried oh, sorry, yeah, I was thinking of something quite different. Maybe I need more coffee
15:09:11 mriedem sean-k-mooney: i would not doubt that config mgmt tooling - if it's setting it - is setting it everywhere because people wouldn't know which services care about it
15:09:29 mriedem sean-k-mooney: nor do you know what out of tree filters could be doing with that thing in the HostState object
15:09:51 sean-k-mooney mriedem: well the value can and typicaly iss different per compute node
15:10:05 sean-k-mooney so if they are doing anything with the config value its proably wronge
15:10:11 openstackgerrit Balazs Gibizer proposed openstack/nova master: nova-manage: heal port allocations https://review.opendev.org/637955
15:10:15 mriedem yeah that's true
15:12:46 mriedem tssurya: why do you have a -1 on https://review.opendev.org/#/c/623558/ ?
15:13:21 mriedem it looks like from your numbers the patch makes a noticeable improvement in scheduling time,
15:13:27 mriedem and the services query stuff could be optimized separately later
15:15:47 tssurya mriedem: I don't remember exactly but I think I also didn't understand some parts of the _filter_enabled_cell_computes function
15:16:38 tssurya also I am pretty sure we have a debug per cell for https://review.opendev.org/#/c/623558/2/nova/scheduler/host_manager.py@790 currently in our deployment which should be fixed.
15:16:44 tssurya but yeah could be done in a FUP
15:18:22 mriedem debug per cell?
15:18:30 mriedem so you're running with this patch in production already?
15:18:35 mriedem but changed that warning to a debug level message?
15:18:36 tssurya yeah
15:18:49 mriedem :/
15:18:55 tssurya wait no
15:19:01 tssurya I mean we are running it in production already
15:19:10 tssurya we are still running it as warning
15:19:17 tssurya what I meant was it causes a lot of noise
15:19:19 mriedem that would be good feedback on the change (that you're running it in prod) and since when
15:19:24 tssurya because it prints a message per cell
15:19:32 tssurya per scheduling
15:19:38 tssurya saying it didn't find anything
15:19:43 tssurya so removing that cell
15:19:44 mriedem sure, so -1 for that i guess,
15:20:03 mriedem but to say "not sure what this does but we've been running with it in prod successfully for months now" isn't really justification for a -1 :)
15:20:43 tssurya mriedem: the -1 on the patch was for the extra improvement
15:20:47 tssurya fo sure
15:20:49 tssurya for*
15:21:01 mriedem which extra improvement? the services query?
15:21:07 tssurya and because I wasn't convinced what _filter_enabled_cell_computes was doing
15:21:12 tssurya mriedem: yes
15:21:25 tssurya but I am okay if that needs to be done in a seperate patch
15:21:28 mriedem ok that's FUP worthy, since this change is already big
15:21:39 tssurya plus its ok for us if we don't backport this
15:21:54 mriedem did jay answer your questions about _filter_enabled_cell_computes ?
15:22:01 mriedem because if not, i wouldn't wait for him since he's gone
15:22:06 tssurya mriedem: no
15:22:25 tssurya I can remove the -1
15:22:46 mriedem a todo could be added for the services query
15:22:50 tssurya also fyi this patch works with https://review.opendev.org/#/c/635532/ and both together give us good performance
15:23:00 tssurya belmiro should be putting a blog post soon
15:23:09 tssurya so we will link the post to this review
15:25:54 openstackgerrit Kashyap Chamarthy proposed openstack/nova master: Document mitigation for Intel MDS security flaws https://review.opendev.org/661574
15:26:33 mriedem tssurya: what is your max_placement_results value now?
15:29:00 tssurya mriedem: 10 still
15:29:59 mriedem heh
15:46:00 openstackgerrit Eric Fried proposed openstack/nova master: Raise if flavor and image disagree on hide_hypervisor_id https://review.opendev.org/663365
15:47:01 efried mriedem: Can we +W https://review.opendev.org/#/c/579897/ now?
15:51:36 mriedem jesus look at all of the new review comments i have to parse
15:53:02 efried only since your last +2
15:53:07 efried kashyap: you still around?
15:53:27 mriedem ok i have it in a tab but in the middle of something
15:53:49 dansmith in case you're wondering,
15:53:52 dansmith I hate it.
15:53:56 efried kashyap: I'd be happy to fix up those couple of typos in https://review.opendev.org/#/c/661574/ if you like.
15:54:19 efried dansmith: necessary evil
15:54:28 dansmith not really
15:55:44 openstackgerrit Eric Fried proposed openstack/nova master: Document mitigation for Intel MDS security flaws https://review.opendev.org/661574
15:56:16 efried kashyap: +2 ^ thanks for your patience
15:59:24 openstackgerrit John Garbutt proposed openstack/nova master: WIP: add scope check, see tests catch the change https://review.opendev.org/657823
16:08:37 jaypipes mriedem: for the record, I haven't died or anything. :P
16:09:09 mriedem jaypipes: i know you're not dead, but i also assume you don't want/need to be pinged for stuff you're not going to have the time to work on
16:10:03 jaypipes mriedem: I'd chatted with Surya about that patch series in Denver and stated I'd rebase, fix merge conflicts and push them, which I did. I also stated I wouldn't have time to shepherd them through to completion and she mentioned she would do that.
16:10:45 mdbooth lyarwood: Question in your q35 patch: Any reason we can't be explicit about q35 machine type? Are there an unmanageable number of q35 machine types or something?
16:10:46 jaypipes mriedem: I'm around, but as you say, not really time to spend. I'm still friendly, though. :)
16:11:12 mriedem jaypipes: ok i didn't mean any offense
16:11:21 mriedem just trying to not distract you
16:11:29 jaypipes no offense taken at all!
16:13:51 lyarwood mdbooth: so yeah there are a few and even more downstream iirc
16:14:13 lyarwood mdbooth: $ qemu-kvm --machine help | grep q35 | wc -l

Earlier   Later