14:00:11 <edleafe> #startmeeting nova_scheduler
14:00:12 <openstack> Meeting started Mon Sep 25 14:00:11 2017 UTC and is due to finish in 60 minutes.  The chair is edleafe. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:00:13 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
14:00:15 <openstack> The meeting name has been set to 'nova_scheduler'
14:00:22 <edleafe> Good UGT morning! Who's here?
14:01:05 <cdent> o/
14:01:09 <takashin> o/
14:01:50 <jaypipes> o/
14:02:30 <mriedem> o/
14:02:31 <ralonsoh> hi
14:03:02 <bauzas> \o
14:03:18 * edleafe watches people slowly trickle in...
14:03:37 <edleafe> #topic Specs and Reviews
14:03:50 <edleafe> Let's start with the specs
14:03:57 <edleafe> #link WIP - CPU features to be reported as traits https://review.openstack.org/#/c/497733/
14:04:09 <edleafe> alex_xu has that marked as WIP
14:04:54 <edleafe> #link Return Alternate Hosts https://review.openstack.org/504275/
14:05:05 <edleafe> This looks like it's pretty close
14:05:15 <edleafe> #link Return Selection objects https://review.openstack.org/498830/
14:05:29 <edleafe> Still some concerns about naming
14:05:45 <edleafe> The existing code refers to host, nodename, and limits
14:06:19 * bauzas gently reminds he has to run off for 20 mins in about 15 mins
14:06:28 <edleafe> Should 'host' be changed to 'compute_node'?
14:06:50 <bauzas> the Destination object mentions 'host' and 'node'
14:07:12 <bauzas> 'service' could be equivalent to 'host' but not 'compute node'
14:07:20 <edleafe> dansmith had a concern that it might be confused with the service
14:07:46 <bauzas> so, the thing is
14:08:01 <bauzas> (host, node) is an old tuple inherited from our dark past
14:08:06 <jaypipes> edleafe: if we're referring to the service endpoint, use host, if we're referring to the thing that provides compute resources call it compute_node. at least, that's MHO
14:08:20 <bauzas> what jaypipes said
14:08:30 <bauzas> but
14:08:50 <bauzas> keep in mind we can have multiple nodes per service
14:08:52 <edleafe> jaypipes: sure, understood. If this were greenfield I would agree. My concern is changing the name that is currently being used
14:10:02 <edleafe> I'm happy to call it compute_node, but I want to make sure that we agree on the name change
14:10:21 <jaypipes> edleafe: well, agree, it's tricky. especially since, frankly, the name right now is actually referring to *both* the service endpoint and the compute resource provider (because historically they have been tightly coupled)
14:10:46 * alex_xu waves late
14:11:06 <edleafe> jaypipes: I think the problem came when I added the compute_node's uuid, which was requested
14:11:23 <edleafe> calling it 'host_uuid' was the source of the confusion
14:12:10 <jaypipes> edleafe: will add thoughts on the spec.
14:12:12 <edleafe> So maybe: 'compute_node_uuid', 'host', 'node_name' ?
14:12:17 <jaypipes> edleafe: sorry, hadn't got to that one yet :(
14:12:26 <edleafe> no problem
14:12:53 <edleafe> next up
14:12:56 <edleafe> #link POST /allocations https://review.openstack.org/#/c/499259/
14:13:04 <bauzas> edleafe: not sure I understand the problem, but I'll voice in the review
14:13:04 <edleafe> cdent: anything to add about that?
14:13:25 <cdent> just that the current blocker is deciding on which representation (listish or dictish)
14:13:29 <edleafe> bauzas: it's just naming. Naming is hard.
14:13:39 <cdent> (or both)
14:14:00 <bauzas> in my mind, host == service and is *never* equal to node or resource provider
14:14:47 <edleafe> #link Placement Agent https://review.openstack.org/#/c/504803
14:15:14 <edleafe> This proposes to have a separate entity report resources instead of nova compute
14:15:29 <edleafe> I haven't read this one yet
14:16:10 <edleafe> #link Minimal cache headers in Placement https://review.openstack.org/#/c/496853/
14:16:19 <edleafe> This looks pretty straightforward
14:16:36 <jaypipes> yup
14:16:43 <edleafe> None of us want to be bad HTTP citizens, do we?
14:16:58 <jaypipes> I do! I do!
14:17:08 * cdent gives jaypipes a malformed cookie
14:17:19 <jaypipes> lol
14:17:21 <bauzas> do I need a visa for being a good HTTP citizen ?
14:17:34 <bauzas> or an ESTA is enough
14:18:08 <edleafe> Last spec on the list looks pretty dead
14:18:09 <edleafe> #link Detailed error info for Placement API https://review.openstack.org/#/c/418393/
14:18:30 <edleafe> cdent: is this still a thing?
14:19:16 <cdent> I can’t really say. I had moved it to queen to see if anyone wanted to take it up
14:19:43 <cdent> I can, if people think it is worthwile, but if people are meh, then I can be meh too
14:20:25 <cdent> so, are people meh or not meh?
14:20:31 <edleafe> I'm kind of meh - it's nice, but is there a current problem that it would solve?
14:20:32 <mriedem> heh,
14:20:33 <mriedem> https://review.openstack.org/#/c/504803/1/specs/queens/approved/placement-agent.rst
14:20:36 <mriedem> so, resource tracker,
14:20:39 <mriedem> but not in nova-compute
14:20:40 <mriedem> -2
14:21:45 <cdent> that almost silence is a resounding meh, can someone with the power to do so abandon https://review.openstack.org/#/c/418393/ to get it off the radar?
14:21:56 <cdent> we can bring it back later if needed
14:22:11 <edleafe> yeah, it seems like noise right now
14:22:15 * cdent nods
14:22:30 <cdent> mriedem: how would you feel ab out “nova-compute, but not in nova”?
14:22:33 <bauzas> what meh are we weighting now ?
14:23:00 <cdent> bauzas: whether to do https://review.openstack.org/#/c/418393/
14:23:11 <mriedem> cdent: so clint's thing from barcelona?
14:23:11 <cdent> so far the consensus seems to be “worry later, if ever"
14:23:31 <bauzas> cdent: ah, thanks
14:23:46 <cdent> mriedem: sort of yeah, I’m mostly just joshing
14:24:10 <bauzas> cdent: well, MHO is that it could be nice but we're almost already pretty full
14:24:16 <edleafe> Next up are a couple of Traits patches from alex_xu:
14:24:16 <edleafe> #link Add traits to GET /allocation_candidates https://review.openstack.org/479776
14:24:20 <edleafe> #link Add traits to get RPs with shared https://review.openstack.org/478464/
14:24:26 <cdent> bauzas: agreed, so if you can abandon it that would be great (I can't)
14:25:09 <alex_xu> #link nova-spec add traits to /allocation_candidates https://review.openstack.org/497713
14:25:16 <alex_xu> edleafe: ^ the spec
14:25:51 <alex_xu> jaypipes, thanks for the review
14:26:00 <jaypipes> npo
14:26:16 <bauzas> cdent: ack
14:26:20 * bauzas has to leave
14:26:30 <edleafe> let's all take the time to review these, as traits support is critical
14:26:33 * bauzas will be back in 15-20 mins
14:26:37 <alex_xu> #link nova-spec request traits in nova https://review.openstack.org/468797
14:26:40 * edleafe waves
14:28:11 <edleafe> alex_xu: thanks - I missed those
14:28:21 <alex_xu> edleafe: np
14:29:39 <edleafe> moving on
14:29:41 <edleafe> #link Allow _set_allocations to delete allocations https://review.openstack.org/#/c/501051/
14:29:59 <edleafe> cdent: that one looks about ready, no?
14:30:35 <cdent> yes, it and its parent are pretty done. it’s the child that still wanders
14:30:38 <mriedem> cdent: i can abandon https://review.openstack.org/#/c/418393/ - it's basically, no owner, low priority right?
14:30:52 <cdent> mriedem: yes
14:31:33 <mriedem> done
14:31:36 <cdent> danke
14:32:00 <edleafe> cdent: while I have your attention
14:32:04 <edleafe> #link WIP - POST /allocations for >1 consumer https://review.openstack.org/#/c/500073/
14:32:12 <cdent> yes, that’s the wandering child
14:32:22 <cdent> which needs resolution of the asociated spec
14:32:31 <cdent> (the listish, dictish question earlier)
14:32:37 <edleafe> understood
14:33:15 <cdent> conceptually speaking it gets the job done okayish
14:33:50 <edleafe> listish, dictish, okayish
14:34:12 <cdent> ish
14:34:20 <edleafe> Moving onish...
14:34:21 <edleafe> #link Use ksa adapter for placement https://review.openstack.org/#/c/492247/
14:34:35 <edleafe> this looks pretty straightforward
14:34:59 <edleafe> (if your into keystone stuff)
14:35:07 <edleafe> s/your/you're
14:35:16 <jaypipes> hmm... I thought we already used ksa for placement?
14:35:34 <cdent> jaypipes: this is the new style
14:36:02 <jaypipes> oh, for endpoint resolution... gotcha.
14:36:05 <cdent> adapter instead of session
14:36:18 <jaypipes> k
14:36:32 <edleafe> #link Add Alternate Hosts https://review.openstack.org/#/c/486215/
14:36:40 <edleafe> this one's pretty close
14:36:45 <jaypipes> edleafe: just wrapped up a review on that one.
14:36:51 <jaypipes> very close. just some suggested renames
14:37:07 <edleafe> still some issues with how exactly to pull from the pool of hosts to get accepatble alternates
14:37:35 <edleafe> jaypipes: thx
14:37:59 <edleafe> Also:
14:38:00 <edleafe> #link Add Selection objects https://review.openstack.org/#/c/499239/
14:38:11 <jaypipes> edleafe: what issues are you concerned about with pulling from the poolk of hosts?
14:38:17 <edleafe> This is kind of on hold until we can fix the naming issues in the spec
14:38:21 <jaypipes> edleafe: you mean how to choose one or more of them?
14:38:42 <edleafe> jaypipes: it is on the edge case of not many acceptable hosts
14:38:58 <jaypipes> edleafe: sorry, not following you... could you elaborate?
14:39:05 <edleafe> I thought itertools.cycle would help, but that can make an infinite loop
14:39:13 <mriedem> edleafe: do you have functional tests somewhere in the series for https://review.openstack.org/#/c/486215/ ?
14:39:20 <mriedem> unit tests aren't going to cut it for alternate hosts
14:39:34 <edleafe> mriedem: no
14:39:45 <edleafe> mriedem: I guess I'll have to add some
14:40:12 <mriedem> yeah....
14:40:27 <edleafe> mriedem: ok, just WIP'd it
14:40:46 <mriedem> well,
14:40:56 <mriedem> i guess if you're going to add later in the series, yeah
14:41:08 <mriedem> the functional tests will really need to be after the conductor and compute parts are done
14:41:19 <mriedem> so we can test reschedules happening across services
14:41:47 <edleafe> mriedem: I had planned on creating functional tests when it came to the RPC changes
14:42:37 <jaypipes> mriedem: yeah, I think that particular patch doesn't change the signature of select_destinations()...
14:42:49 <jaypipes> mriedem: we really need func tests on the patch after that and the RPC changes.
14:43:02 <edleafe> jaypipes: exactly. Just the _schedule() method
14:43:40 <mriedem> ok
14:43:56 <mriedem> and,
14:44:05 <mriedem> to be clear, the rpc between conductor and scheduler isn't the interesting part so much
14:44:14 <mriedem> it's the interaction between conductor and the computes
14:44:30 <mriedem> like, wherever allocations are posted/removed during rescheduels
14:44:32 <mriedem> *reschedules
14:44:50 <edleafe> mriedem: yeah, that's happening in a much later patch
14:46:48 <edleafe> Moving on to...
14:46:50 <edleafe> #topic Bugs
14:46:53 <edleafe> #link Placement bugs https://bugs.launchpad.net/nova/+bugs?field.tag=placement
14:46:55 <ralonsoh> hi
14:46:59 <ralonsoh> There is an spec
14:47:01 <ralonsoh> missing
14:47:09 <ralonsoh> #link Enable SR-IOV NIC offload feature discovery  https://review.openstack.org/#/c/504895/
14:48:23 <edleafe> ralonsoh: ok, but that's only slightly concerned with the scheduler, right?
14:48:44 <ralonsoh> and placement. So, where do I put this one?
14:50:03 <edleafe> ralonsoh: seems like a mostly nova spec
14:50:22 <edleafe> ralonsoh: do you have questions about it?
14:50:38 <ralonsoh> edleafe: no, just time form you to review it
14:51:02 <edleafe> ralonsoh: ok, thanks!
14:51:36 <cdent> Was dan’s stack of migration uuid and allocation updates listed earlier?
14:51:43 <cdent> #link allocation by migration https://review.openstack.org/#/q/topic:bp/migration-allocations+status:open
14:52:18 <edleafe> cdent: ah, good catch. No, I missed that series
14:52:33 <dansmith> I'm still working on that of course
14:52:40 <dansmith> hence all the -1s
14:53:09 <edleafe> dansmith: ok, just let us know when they are ready for wider review
14:53:25 <dansmith> ack
14:53:53 <edleafe> So... we were on the topic of bugs :)
14:53:59 <edleafe> Only one new one this week
14:54:06 <edleafe> And it's in progress
14:54:22 <edleafe> #link https://bugs.launchpad.net/nova/+bug/1718212
14:54:23 <openstack> Launchpad bug 1718212 in OpenStack Compute (nova) "Compute resource tracker does not report correct information for drivers such as vSphere" [Medium,In progress] - Assigned to Radoslav Gerganov (rgerganov)
14:54:44 <edleafe> Anything else for Bugs?
14:55:19 <mriedem> i've got a fix that needs final +2
14:55:37 <mriedem> https://review.openstack.org/#/c/506458/
14:55:39 <mriedem> and below
14:55:39 * bauzas is back
14:55:44 <mriedem> those will need to go to pike
14:56:11 <bauzas> mriedem: easy peasy
14:56:29 * edleafe admires bauzas' timing
14:56:39 <edleafe> #topic Open discussion
14:56:51 <edleafe> We have 3 minutes or so - anything to discuss?
14:56:59 <takashin> Hi
14:57:02 <takashin> #link Enable cold migration with target host(1/2) https://review.openstack.org/#/c/408955/
14:57:24 <takashin> It is ready for review. Would you review it?
14:57:56 <edleafe> takashin: added to my review list
14:58:08 <takashin> edleafe: Thank you.
14:58:23 <edleafe> Anything else?
14:58:51 <cdent> call it
14:59:12 <edleafe> cdent: sure. What should I call it? :-P
14:59:25 <cdent> over under under done
14:59:38 <edleafe> OK, thanks everyone!
14:59:39 <edleafe> #endmeeting