Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-03-19
15:13:20 cfriesen jroll: thanks
15:13:25 jroll np
15:13:48 diga edleafe: Sure, will go through it
15:14:49 diga jaypipes: Thank you for your help
15:16:43 jaypipes diga: if I come across something for you, I will let you know.
15:17:01 diga jaypipes: Sure
15:21:56 diga jaypipes: I also know NFV technologies like hugepages, numa, dpdk very well. If you have something on that, let me know
15:23:57 jaypipes diga: cool. can you review this spec please? https://review.openstack.org/#/c/552924/
15:24:17 diga jaypipes: sure
15:28:12 Spaz-Home Morning folks
15:29:39 openstackgerrit Chris Dent proposed openstack/nova master: Move placement exceptions into the placement package https://review.openstack.org/549862
15:29:39 openstackgerrit Chris Dent proposed openstack/nova master: Optional separate database for placement API https://review.openstack.org/362766
15:29:40 openstackgerrit Chris Dent proposed openstack/nova master: Isolate placement database config https://review.openstack.org/541435
15:29:40 openstackgerrit Chris Dent proposed openstack/nova master: WIP: Ensure that os-traits sync is attempted only at start of process https://review.openstack.org/553857
15:32:13 Spaz-Home Actually taking a relax weekend.. no coding.. just boiling 55 Habaneros, 2 Ghost Peppers, and trying to make a hot sauce without killing myself. Enjoy your day Cloud Creators :p
15:38:03 cdent jaypipes: what's the status of your "standardized provider descriptor file" spec/idea? Where on the scale or need<->experimental fishing is that?
15:38:23 jaypipes cdent: it's dead.
15:38:29 jaypipes cdent: I'll abandon it.
15:38:32 cdent dead fish
15:38:40 jaypipes cdent: smelly dead fish.
15:39:05 jaypipes cdent: I did too :) but I understand why dansmith doesn't.
15:42:25 mriedem sahid: there are a couple of issues in the tests for https://review.openstack.org/#/c/497457/2
15:42:27 mriedem https://review.openstack.org/#/c/497457/
15:42:39 mriedem sahid: specifically self.assertEqual(0, guest.migrate_configure_max_speed.called)
15:42:44 mriedem .called is a boolean
15:43:05 mriedem it's probably better to use guest.migrate_configure_max_speed.assert_not_called()
15:44:04 mriedem if you're about done for the day i could clean those up and then just approve
15:44:49 sahid oh i was sure of that "called" returns a number
15:44:52 sahid no no, i will
15:44:59 mriedem call_count returns a number
15:45:03 sahid but thanks for asking :)
15:45:12 sahid ack
15:45:29 mriedem thanks for finding that issue in the multinode job config
15:46:31 sahid mriedem: yeah, it's that g-api service which looked very strange
15:47:06 openstackgerrit Jay Pipes proposed openstack/nova master: validate provider summary structure https://review.openstack.org/536516
15:47:07 openstackgerrit Jay Pipes proposed openstack/nova master: WIP - populate alloc cands for trees https://review.openstack.org/534968
15:47:39 kashyap sahid: One more nit there: s/TAPs devices/TAP devices/ :-)
15:47:46 kashyap (In driver.py)
15:48:49 cfriesen has anyone run into problems with setuptools recently? I got a report that 39.0.1 didn't work with our version of pike, wondering if it's a general problem.
15:49:05 sahid kashyap: :) sure i will address that too
15:49:41 kashyap Thx
15:50:42 dansmith cfriesen: I had an issue downstream last week where the parse_version was returning a tuple instead of a string
15:50:56 dansmith cfriesen: migi was looking at it, in case you or he know anything else
15:53:34 cfriesen dansmith: we ended up pinning it to 38.5.1 for now, which seems to work okay
16:00:41 mriedem jaypipes: on that vmware resource pools spec, https://review.openstack.org/#/c/549067/6/specs/rocky/approved/vmware-resource-pool-placement.rst@210
16:01:05 mriedem jaypipes: finally just realized that this is proposing that the admin changes things *in vcenter*, not placement directly, and that's all proxied through the virt driver
16:01:24 mriedem which kind of goes against something we've always said we woudn't do, which is support out of band changes on the hypervisor
16:02:27 cdent mriedem: I think those sorts of changes are pretty much why upt exists?
16:02:42 cdent the hardware underlaying the hypervisor changes in some fashion
16:02:43 jaypipes mriedem: vcenter virt driver is virtually identical (pun intended) to how ironic virt driver wants to do things. i.e. the virt driver (not the compute service worker ala the resource tracker) creates the compute node resource provider records.
16:02:45 cdent upt notices
16:03:10 cdent efried: powervm is much the same, yeah? ^
16:03:38 mriedem jaypipes: this isn't creating the compute node resource provider record,
16:03:59 mriedem it's seeing, oh the admin created some resource pools in this cluster in vcenter directly, and now the virt driver needs to mirror that into placement,
16:04:11 mriedem and what happens when the admin decides to delete those pools in vcenter? nova would also need to mirror that
16:04:43 dansmith cdent: powervm only has one node
16:05:19 mriedem powervm wants to manage shared storage pool aggregates from what i remember
16:05:19 dansmith jaypipes: I read that spec as not exposing more nodes, but exposing a cluster as a very NUMA single node
16:05:28 cdent dansmith: not talking about the node here, just what the node is aware of: efried has mentioned that the powervm driver wants to ... yeah what mriedem said
16:05:36 dansmith mriedem: right, but not multiple nodes from one service as it stands in tree right now
16:05:50 mriedem these would be nested RPs
16:06:02 dansmith yeah, I think jaypipes is saying multiple nodes a-la ironic
16:06:07 dansmith and I don't think that's on the table for either
16:06:16 mriedem compute node RP (created by nova) -> nested tenant-specific resource pool providers, created by the virt driver, proxied from vcenter
16:06:30 mriedem the 1:M compute node thing isn't in this
16:06:30 dansmith yep
16:06:34 dansmith right
16:06:39 cfriesen mriedem: is there anything in particular blocking https://review.openstack.org/#/c/465787/ ? I notice it hasn't really gotten any reviews other than me.
16:06:42 dansmith I think jaypipes was asserting that it was
16:07:23 jaypipes dansmith: sorry, on a video call for 45 mintues
16:07:38 mriedem i guess since we don't model numa as a nested RP for the libvirt driver yet, i'm not sure how much proxy we'll be doing to compare to the vcenter thing
16:07:50 dansmith the difference here I think,
16:08:02 dansmith is that libvirt will report the static nature of the system as nested NUMA RPs,
16:08:11 openstackgerrit Lee Yarwood proposed openstack/nova master: libvirt: Block swapping to an encrypted volume when using QEMU to decrypt https://review.openstack.org/544238
16:08:12 dansmith where vmware would be representing cluster config, which is mutable
16:08:33 dansmith and I agree that sucks, and I don't know what would happen if vmware swizzled things after instances are booted
16:08:35 dansmith but..
16:08:55 dansmith what we have today is also wrong (exposing a cluster as a single uniform set of resources)
16:09:06 elod cfriesen: it's general and there's this patch https://review.openstack.org/#/c/554053/ (which unfortunately ends up in another bug with pip... see depends-on)
16:10:11 mriedem this spec also bakes business logic into the driver about how those resource pools are created in vcenter, i think
16:10:15 cdent dansmith: since I'm not aware of the many bits of history involved with this stuff (either internal or external to vmware) why is that "wrong" instead of "different"?
16:10:25 mriedem basically based on the name of the resource pool, it's encoded into the pool name which tells the virt driver what to do about it
16:10:37 dansmith cdent: because you can't schedule instances to it that span what nova thinks are uniform resources
16:11:02 dansmith cdent: at least, that's my understanding of it at the moment
16:11:05 openstackgerrit Lee Yarwood proposed openstack/nova stable/pike: libvirt: Block swap volume attempts with encrypted volumes prior to Queens https://review.openstack.org/543569
16:11:19 openstackgerrit Eric Berglund proposed openstack/nova master: PowerVM Driver: vSCSI volume driver https://review.openstack.org/526094
16:11:24 dansmith cdent: i.e. if you have a cluster with two 1G nodes, nova thinks you should be able to boot a 2G instance there but you can't
16:12:01 cdent that's recently fixed via inventory handling with max_unit
16:12:08 dansmith s/fixed/hacked
16:12:13 cdent no
16:12:36 cdent you're asserting that targeting a cluster is wrong, and what I'm trying to understand why that's wrong
16:12:40 dansmith it is a hack, because if you have memory available on one and cpu available on another, you can't represent that
16:12:48 dansmith no, that's not what I'm saying
16:13:04 dansmith I'm saying nova thinks the cluster is a single bag of resources that can be combined and that's wrong
16:13:29 dansmith and I'm saying that I think the spec we're discussing is trying to rectify that by using multiple providers with the proper amounts of inventory each
16:14:02 mriedem yeah it's trying to break the single cluster into multiple pools
16:14:03 cdent the spec is trying to allow resource pools (a vmware thing that enables tenanted clusters) to work
16:14:13 mriedem but it's also tying those pools to specific tenants
16:14:24 mriedem like the aggregate multitenenancy isolation filter
16:14:25 mriedem i think

Earlier   Later