Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-05-30
13:12:37 efried shilpasd: Are you implementing it?
13:12:49 shilpasd yes
13:12:50 gibi efried: yes, there was no resource tracking
13:13:33 shilpasd efried: We have added code to sync traits set on aggregates for command "nova-manage placement sync_aggregates".
13:14:24 efried okay
13:15:06 shilpasd efried: further observed that 'syncing of traits set on aggregate is required for the other scenarios' like aggregate-create/delete/aggregate-remove-host/aggregate-set-metadata
13:15:18 efried yes, that makes sense to me.
13:15:52 efried though I'm not sure what we should do on delete or remove-host ...
13:16:52 efried perhaps we need a --flush-traits option of some sort, so the admin can choose.
13:17:09 shilpasd efried: for delete, will remove entry from placement.placement_aggregates and for remove-host will remove host association from aggragte
13:17:43 efried right, but the question is whether you should remove the aggregate's listed traits from the compute node resource providers in that aggregate
13:18:11 efried answer is probably: "it depends" :\
13:18:45 sean-k-mooney efried: gibi three minor nits but yep that looks good to me
13:19:33 efried shilpasd: This would probably be a good topic for the ML.
13:19:54 efried gibi: I +2ed. The spelling nits probably aren't worth fixing, want me to merge?
13:21:17 sean-k-mooney i would eihter fix them and fast approve or just merge it as the content is correct modulo a few typos
13:22:34 openstackgerrit Matt Riedemann proposed openstack/nova master: Add heal_instance_mappings command https://review.opendev.org/655908
13:22:50 gibi efried: merge it please
13:22:58 gibi thanks
13:23:06 efried gibi: done
13:23:14 gibi awesome
13:23:26 efried shilpasd: I suppose another option is to namespace the traits to be used for this feature.
13:24:46 efried shilpasd: e.g. if trait name contains '_AGGREGATE_ISOLATION_', automatically remove it when the agg is deleted or the host is removed. (Unless of course the host belongs to *another* aggregate which still has that trait...)
13:26:13 efried gibi: assume this is for train, yah?
13:26:23 efried (gibi: I'm doing the bp paperwork)
13:26:24 gibi efried: yes, it is for train
13:26:34 efried k
13:26:56 gibi I have couple of patches up already but more work is needed there which i will do in the coming days weeks
13:29:07 efried gibi: are the patches tracked against this bp? The "Implementation" is not showing as started.
13:29:25 gibi efried: I guess the bp rename did not helped
13:29:25 shilpasd efried: aggregate can have any trait and that we need to be sync with placement, irrespective of namespace, right?
13:29:30 gibi efried: let me fix it
13:29:39 efried (I could go look, but I'm being lazy, aka doing a bunch of other things at the same time)
13:30:08 efried shilpasd: When you create an aggregate or add a host, I would say yes, you should add all the traits.
13:30:20 efried shilpasd: I'm just trying to come up with a way to handle *deletion*.
13:30:33 shilpasd efried: understood, okay
13:30:53 efried shilpasd: For example, if you're creating an aggregate to isolate your hosts that have HW_CPU_X86_AVX, it makes zero sense to remove that trait if you destroy that aggregate.
13:30:53 gibi efried: I guess the implementation status will be updated if I push a new patch up targeting the renamed bp
13:31:10 efried gibi: yeah, it's totally no big deal, I really don't care, just happened to notice.
13:31:14 gibi efried: but the patches are here https://review.opendev.org/#/q/topic:bp/support-move-ops-with-qos-ports+(status:open+OR+status:merged)
13:31:19 gibi efried: ack
13:31:34 efried shilpasd: And in fact that trait would be reasserted on the host the next time the periodics run
13:32:14 efried shilpasd: ...but that is definitely not a behavior we should rely on to "heal" such traits.
13:33:53 openstackgerrit Matt Riedemann proposed openstack/nova master: Change InstanceFaultRollback handling in _error_out_instance_on_exception https://review.opendev.org/633212
13:33:54 openstackgerrit Matt Riedemann proposed openstack/nova master: Raise InstanceFaultRollback for UnableToMigrateToSelf from _prep_resize https://review.opendev.org/633227
13:35:17 shilpasd efried: ML is good suggestion to brain storm this, will generate the same, thanks
13:35:42 openstackgerrit Merged openstack/nova-specs master: Server move operations with ports having resource request https://review.opendev.org/652608
13:36:10 efried shilpasd: Cool. To summarize the alternatives we've come up with so far, on delete/remove type operations, we could:
13:36:10 efried - Leave all traits alone. If they need to be removed, it would have to be manually via a separate step.
13:36:10 efried - Support a new option so the caller can dictate whether the operation should remove the traits. (This is all-or-none.)
13:36:10 efried - Define a "namespace" - a trait substring - and remove only traits in that namespace.
13:36:38 efried shilpasd: ...but on create/add type operations, I think we should always add the traits to the hosts.
13:45:12 shilpasd efried: clear, thanks
13:55:37 openstackgerrit Matt Riedemann proposed openstack/nova stable/stein: Fix the server group "policy" field type in api-ref https://review.opendev.org/662224
13:58:23 mriedem gibi: on https://review.opendev.org/#/c/512623/ i guess you re-ordered the series so the functional test, which depends on the fake driver global conf changes, could just be on master and the bug fix can be backported, is that right?
13:58:55 gibi mriedem: yes. lyarwood convinced me that it is better that way
13:59:20 gibi mriedem: this means no functional re-create and functional coverage on stable branches
13:59:27 mriedem i was hoping we could get the functional test backported as well so that's why i was pushing on slimming down the fake driver stuff, but alas i guess it wasn't possible
13:59:41 mriedem if we're good on master i think that's sufficient
13:59:50 gibi mriedem: it was hairy due to the fake driver stuff
13:59:56 mriedem that's a lot better than a lot of bug fixes we've had in the past which are just unit test coverage
14:00:06 efried nova meeting now
14:00:34 mriedem speaking of functional regression recreate tests https://review.opendev.org/#/c/661822/
14:00:42 mriedem i could use some eyes on that since it's going all the way back to ocata
14:01:02 gibi mriedem: added it to my queue
14:01:12 mriedem thanks
14:02:30 gibi mriedem: btw, I'm reading your cross-cell resize review guide and I have to tell it helps me a lot. thanks for writing that up
14:03:04 mriedem gibi: thanks, i was about to ask if anyone had actually read it
14:03:57 efried mriedem: I fully intend to, if I ever get around to reviewing that series.
14:04:16 efried I've been kinda hoping others would have approved it by the time I got around to it.
14:11:44 openstackgerrit Adrian Chiris proposed openstack/nova stable/queens: Move get_pci_mapping_for_migration to MigrationContext https://review.opendev.org/661571
14:11:44 openstackgerrit Adrian Chiris proposed openstack/nova stable/queens: Allow driver to properly unplug VIFs on destination on confirm resize https://review.opendev.org/661572
14:18:45 sean-k-mooney oh ^ are backports. i was confused as i had thought it had merged on master
14:19:04 sean-k-mooney also nova team meeting is not right
14:39:50 gmann johnthetubaguy: on policy spec- base defined role mapping to check_str is enough or you want complete list of policy and their propoed check_str ?
14:47:41 mriedem adrianc_: do you care about this old bug fix? https://review.opendev.org/#/c/361438 if so please ack on it that you'll take it over otherwise i'm going to abandon it
14:47:57 mriedem starlingx probably cares as well but cfriesen isn't around to ask
14:50:14 johnthetubaguy gmann: I was thinking just the base, as examples, maybe add in keypairs too, but the rest we meet those problems as we hit them?
14:50:38 johnthetubaguy gmann: do you think my suggestion works, with changing the base rule based on that config?
14:51:12 gmann johnthetubaguy: +1 on base rule details i will add those.
14:51:55 efried mriedem: dtroyer might be able to answer starlingx care-about questions.
14:52:09 gmann johnthetubaguy: but till enforce_scope is false, scope_type will not have any effect so out old deprecated rule with old roles will work fine right ?
14:52:29 gmann i forgot to reply config things on gerrit.
14:53:00 cdent oh I had an open an missed it, but it's mostly directed at mriedem
14:53:04 mriedem f i suppose they have an irc channel
14:53:15 johnthetubaguy good question... its more than enforce_scope isn't enough, because its basically system or project for almost anything
14:53:35 cdent you remember a couple weeks ago when 10,000 allocation candidates was around 36 seconds? it's down to 6 now after "various changes"
14:53:45 cdent so that's like a
14:53:47 efried \o/
14:53:51 johnthetubaguy gmann: if we add the scope check in the rule by default all the time, boom, all current admins are busted
14:54:00 cdent #success 10000 allocation candidates got faster
14:54:02 openstackstatus cdent: Added success to Success page (https://wiki.openstack.org/wiki/Successes)
14:54:07 johnthetubaguy cdent: nice job
14:54:34 gmann johnthetubaguy: we will keep old check_str as deprecated rule , for example: https://review.opendev.org/#/c/645427/5/nova/policies/services.py
14:54:48 sean-k-mooney cdent: just faster in aggreate or was there one change in partcalar that stood out
14:54:59 gmann ^^ above patch was for granular change but it can be same for check_str change also
14:55:01 sean-k-mooney but also cool
14:55:20 mriedem cdent: cool
14:55:20 cdent sean-k-mooney: i'll take the details to the placement channel
14:59:53 gmann johnthetubaguy: i mean old rule(check_str without scope string) will keep working because we will keep them as deprecated rule. For example, current admin role (without scope thing) will be able to access the host APIs.

Earlier   Later