Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-12-21
12:19:45 openstackgerrit Matthew Booth proposed openstack/nova master: Fix unit test failures when direct IO not supported https://review.openstack.org/529568
12:20:20 mdbooth If we can get ^^^ in sharpish that would be awesome. It's really annoying.
12:21:13 stephenfin mdbooth: (y)
12:23:41 stephenfin mdbooth: Regarding the preceding patch, do you know if we support out-of-tree volume drivers? https://review.openstack.org/#/c/529556/
12:24:51 stephenfin Also, the one you linked could probably be rebased onto master to get it in sooner. No dependency that I can see
12:27:14 openstackgerrit Lee Yarwood proposed openstack/nova master: libvirt: Refactor encryptor attach and detach calls https://review.openstack.org/460243
12:27:42 openstackgerrit Stephen Finucane proposed openstack/nova master: objects: remove pagesize from __init__ of InstanceNUMATopology https://review.openstack.org/485553
12:27:43 openstackgerrit Stephen Finucane proposed openstack/nova master: objects: remove related pinning from __init__ of InstanceNUMATopology https://review.openstack.org/485554
12:27:43 openstackgerrit Stephen Finucane proposed openstack/nova master: objects: remove cpuset_reserved from __init__ of InstanceNUMATopology https://review.openstack.org/466030
12:29:14 openstackgerrit Stephen Finucane proposed openstack/nova master: setup.cfg: Explicitly set [build_sphinx] builders https://review.openstack.org/508483
12:32:53 kashyap lyarwood: On this - https://review.openstack.org/#/c/464008/1/nova/virt/libvirt/config.py. Did you forget to add the comment describing the relation b/n classes & tags?
12:33:24 kashyap In response to Andrew Volkov's comments
12:35:08 openstackgerrit Stephen Finucane proposed openstack/nova master: Convert floating IP tests to ddt https://review.openstack.org/471360
12:36:58 lyarwood kashyap: I did, I'm not sure if it's really required tbh
12:40:36 kashyap lyarwood: Yeah, it's fine for me
13:05:56 openstackgerrit Merged openstack/nova master: Pass bdms to versioned notifications during finish_revert_resize https://review.openstack.org/527446
13:15:03 openstackgerrit Stephen Finucane proposed openstack/nova master: tests: Use correct response type in tests https://review.openstack.org/527182
13:15:13 openstackgerrit Stephen Finucane proposed openstack/nova master: Handle images with no data https://review.openstack.org/526329
13:16:15 openstackgerrit Stephen Finucane proposed openstack/nova master: Remove dead parameter from '_create_domain_and_network' https://review.openstack.org/512352
13:19:58 jaypipes stephenfin: lol, sorry about the force=True thing...
13:20:13 jaypipes stephenfin: just goes to show the importance of kwargs ;)
13:20:27 stephenfin jaypipes: I'll be honest, you had me doubting myself :D
13:20:51 jaypipes stephenfin: well, sorry about that. :) but it was not obvious...
13:21:17 stephenfin No no, it's fine. Turned out the parameter wasn't necessary anyway
13:21:23 stephenfin So something good came out of it
13:23:01 stephenfin Think you might have called this one wrong too ;) https://review.openstack.org/#/c/527472/7/nova/pci/stats.py@265
13:23:27 jaypipes stephenfin: still looking at that one...
13:23:50 jaypipes stephenfin: you're probably right. I'm just getting old. but I still want to look into that further.
13:24:52 stephenfin jaypipes: Sounds good to me. I can rework in any case if it makes it a little easier to parse
13:25:29 mdbooth Waiting for my unit tests to run is the pythonistas waiting for a compile
13:25:51 jaypipes indeed
13:26:43 mdbooth Unfortunately I don't have a foosball table
13:28:54 openstackgerrit Matthew Booth proposed openstack/nova master: Fix unit test failures when direct IO not supported https://review.openstack.org/529568
13:28:54 openstackgerrit Matthew Booth proposed openstack/nova master: Removed unused argument from LibvirtDriver._connect_volume https://review.openstack.org/529556
13:28:55 openstackgerrit Matthew Booth proposed openstack/nova master: Remove unused argument from LibvirtDriver._disconnect_volume https://review.openstack.org/529585
13:29:06 mdbooth I blame lyarwood ^^^
13:32:08 jaypipes stephenfin: to be honest, the whole stats.py module makes me want to throw up a bit in my mouth.
13:32:11 kashyap mdbooth: Removing cruft is good. Thank you
13:32:24 openstackgerrit Lee Yarwood proposed openstack/nova master: conf: Do not inherit image signature props with snapshots https://review.openstack.org/527046
13:32:57 mdbooth kashyap: There's also also a lesson above in how a misunderstanding in one bit of code can be copied around and mutate.
13:33:00 stephenfin jaypipes: Agreed. The docstrings are a start. Next steps is a big de-dupe
13:33:18 stephenfin But I wasn't mixing that in here. Would like it to get in before the new year :)
13:33:20 kashyap mdbooth: Yeah, /me nods
13:34:56 stephenfin jaypipes: Is the move to resource providers going to get swept up in efried's generic device thing'y or is it separate? If latter, dibs
13:37:04 jaypipes stephenfin: the generic device manager ideas revolve around using the ProviderTree for managing PCI (and other) devices just like any other resource provider (like a compute node for VCPU/MEMORY_MB or a pGPU group for VGPU resources). the idea is to get rid of the whole PciDevicePoolList (stats) object entirely.
13:37:45 jaypipes stephenfin: and the generic device manager would *only* gather inventory and do the binding with the guest. it wouldn't do anything with scheduling/placement logic
13:38:22 stephenfin Right-o, that's what I suspected. So we need to do the reporting aspect and then build upon that to use what's been reported?
13:42:59 jaypipes stephenfin: not entirely sure what you mean by "build upon that to use what's been reported"
13:44:45 jaypipes stephenfin: the placement service will be the thing that allocates a device to a guest instance and the generic device manager will just be responsible for plumbing it through to the guest. for example, the GDM will get the local address for a provider's UUID and stuff like that. what the GDM *won't* do is track available/allocated/claimed crap for devices, since that information will already be in the placement service.
13:45:45 jaypipes stephenfin: in addition, the GDM will manage the mapping of device address or local name to the resource provider's UUID.
13:46:23 jaypipes stephenfin: and the GDM will coordinate work with the hypervisor (if necessary) to inventory the collection of devices available on a compute node for guests to use (i.e. the whitelist stuff)
13:46:32 stephenfin jaypipes: Oh, I thought GDM was only responsible for the "reporting device (in a generic way) to placement" piece. Looks like I need to re-read that spec
13:46:42 openstackgerrit Lee Yarwood proposed openstack/nova master: conf: Do not inherit image signature props with snapshots https://review.openstack.org/527046
13:47:22 jaypipes stephenfin: yes, that's the majority of its responsibility. that's what that "maintaining a mapping of local device name/address to resource provider UUID" is all about. :)
13:47:40 stephenfin So "build upon that to use what's been reported" referred to wiring up what gdm gave us with the various drivers
13:47:49 stephenfin Which I thought would be a separate thing. Guess that doesn't make sense though
13:47:56 stephenfin to separate it, that is
13:49:03 jaypipes stephenfin: some of that I believe will depend on the hypervisor driver.
13:49:23 jaypipes stephenfin: libvirt does a lot of that work. but drivers like powervm don't necessarily.
13:51:31 mdbooth Eurgh. My local disk serials series is hitting merge conflict hell :(
13:51:46 mdbooth Prepare for another rebase of the whole stack
13:51:47 dtantsur hi mriedem! could you please take a look at https://bugs.launchpad.net/nova/+bug/1739440 and tell me if it's okay as a bug, or should be a blueprint, or a spec, or a bad idea at all
13:51:48 openstack Launchpad bug 1739440 in OpenStack Compute (nova) "Ironic virt driver is hardcoding Ironic API microversion" [Undecided,New]
13:52:30 mriedem not a spec,
13:52:41 mriedem not a bad idea
13:53:01 jaypipes dtantsur: ^ answer.
13:53:11 mriedem i don't know that it's a bug, or a blueprint
13:53:48 dtantsur okay, thanks! I'll leave it as a bug for now then, we can reconsider if the actual change starts looking more like a feature
13:53:49 mriedem i guess it's no different than having different computes with different versions of libvirt that can support some capability
13:53:56 mriedem dtantsur: yeah that's a good route
13:54:22 stephenfin jaypipes: Yup, that all makes sense. Hopefully swapping the driver stuff that uses PciDevicePoolList in with whatever gdm offers us will be painless enough
13:54:45 dtantsur TheJulia: re version negotiation bug ^^^^
14:04:34 mnaser so I've been thinking about this lately - with cinder, volume types can have a specific iops quota to them. in an environment like ceph where performance scales (almost) linearly will capacity, being able to tie the # of iops to the volume size is used by some other cloud providers which is very interesting for this use case
14:05:17 mnaser i think the implementation details are pretty easy because we just need a "iops per gb" and then instead of using the number provided by cinder
14:05:18 mriedem edleafe: i'm going to update the alternate hosts patch - i extended a reschedule functoinal test to use caching scheduler which exposes a bug in the scheduler manager
14:05:23 mnaser just take the volume size * iops per gb and thats it
14:06:05 edleafe mriedem: you already wrote it? I was just starting on it
14:10:06 mriedem edleafe: yup it's just a simple functional test extend
14:10:33 edleafe mriedem: ok, I'll hold off
14:13:50 openstackgerrit Matt Riedemann proposed openstack/nova master: Make conductor pass and use host_lists https://review.openstack.org/511358
14:13:51 mriedem edleafe: ^
14:14:10 mriedem so with that we get the test coverage and don't need a big new unit test, plus a unit test wouldn't have caught the bug in the scheduler manager
14:15:17 mriedem +2 now
14:15:26 mriedem and now jay is gone
14:20:38 mnaser does the attach_volume() driver function at any point know the size of the disk?
14:29:36 mdbooth mnaser: If this patch lands it would https://review.openstack.org/#/c/528363/6
14:30:30 mnaser mdbooth: i see, well thankfully, it was all just noise and this was all i needed - https://review.openstack.org/#/c/447127/ :D
14:30:48 mdbooth mnaser: Are you writing a new volume driver, btw?
14:31:35 mnaser mdbooth: oh nope, just was going to try and make the existing rbd one do qos per-gb so the qos scales with the volume size
14:32:14 mdbooth mnaser: i.e. bigger disks are permitted more bandwidth, linearly?
14:32:32 mnaser mdbooth: correct, bandwidth and iops, because we're using ceph so those two are related
14:32:53 mdbooth Makes sense on the face of it.
14:34:16 openstackgerrit Takashi NATSUME proposed openstack/nova master: api-ref: Verify parameters in servers.inc https://review.openstack.org/528201
14:37:08 openstackgerrit rahul bardia proposed openstack/python-novaclient master: nova limits ERROR (Exception): Field names must be unique https://review.openstack.org/527922
14:44:54 mdbooth Is there any chance we could chuck this one into the gate: https://review.openstack.org/#/c/529568/ ?
14:45:19 mdbooth It's a bug affecting execution of unit tests on a filesystem which doesn't support direct io
14:45:36 mdbooth Which apparently my workstation's filesystem doesn't
14:46:22 mriedem gmann: alex_xu: don't we have something in the docs that talks about not doing API extensions anymore?
14:46:25 mriedem looking for that here https://docs.openstack.org/nova/latest/contributor/index.html#the-nova-api
14:46:29 mdbooth Incoming... (sorry: merge conflict)

Earlier   Later