Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-04-11
19:47:19 mordred but that's not helpful for gate testing :)
19:47:48 openstackgerrit Matt Riedemann proposed openstack/nova master: Add uuid to Service model https://review.openstack.org/453025
19:47:49 openstackgerrit Matt Riedemann proposed openstack/nova master: Add online data migration for populating services.uuid https://review.openstack.org/454899
19:48:51 openstackgerrit Jay Pipes proposed openstack/nova master: Read Neutron port 'binding_profile' during boot https://review.openstack.org/449257
19:49:09 openstackgerrit Jay Pipes proposed openstack/nova master: Modify PciPassthroughFilter to accept lists https://review.openstack.org/444946
19:49:21 openstackgerrit Jay Pipes proposed openstack/nova master: Read Neutron port 'binding_profile' during boot https://review.openstack.org/449257
19:53:27 mriedem cfriesen: you've got a little unit test failure to fix up in https://review.openstack.org/#/c/447055/
19:59:24 cfriesen mriedem: okay, will sort it out. I'm putting out customer fires.
20:00:22 cfriesen mriedem: mordred: apparently libvirt/qemu will boot a vmdk-format glance image.
20:01:31 mriedem https://en.wikibooks.org/wiki/QEMU/Images#Image_types
20:02:01 mriedem i don't see ploop in that list
20:02:16 mordred cfriesen: fascinating
20:03:04 cfriesen I have a report that it boots up, but then fails when we try to live-migrate the instance.
20:03:26 cfriesen I'll raise a bug once we have the info nailed down a bit more securely
20:03:50 mriedem please include customer credit card numbers in the bug report thanks
20:04:00 cfriesen wouldn't that be nice. :)
20:04:26 cfriesen not actually expecting it to get fixed, but at least someone else can see the bug report and know it's broken. :)
20:07:21 jaypipes wait, mriedem, your customers pay you?
20:07:53 mriedem i don't have customers
20:08:47 d9k is there any code for different components to share locks?, i dont see anything
20:09:29 d9k also oslo, which uses fasteners does not have any code for shared locks either
20:10:26 d9k i posted earlier about a race nova and neutron introduce with iptables and ipsets, both need to acquire shared locks from each other
20:11:06 sdague d9k: there is no mechanism for that today, is there an open bug on that issue?
20:16:54 d9k no not yet, acquiring neutron-ipset lock in the iptables-save/restore bit fixes https://ptpb.pw/16S8, but https://ptpb.pw/aukQ needs locks too
20:19:59 mriedem http://forumtopics.openstack.org/cfp/details/11
20:21:51 d9k also this happens fairly often when running tempest compute.servers
20:30:13 sdague d9k: a bug with details would be good, sounds like you've done a lot of the debugging, and it would be great to get all those details down
20:38:56 swebster Has anybody run the latest py27 unit tests on a Centos system? I'm getting a failure on the nova.tests.unit.objects.test_objects.TestObjectVersions.test_versions test which seems to be related to the recent "Use cursive for signature verification" change: https://review.openstack.org/#/c/351232/. The test passes on my Ubuntu Xenial box, with cursive versions both at 0.1.2.
20:44:57 mriedem swebster: it's likely something in the cursive library, what is the actual error? (in a paste)
20:46:10 swebster mriedem: Here's a paste of the error(s) seen https://pastebin.com/1rqJaTPs
20:48:55 mriedem my guess would be something with this https://github.com/openstack/cursive/blob/master/cursive/signature_utils.py#L177
20:49:08 mriedem that changes what's in REGISTERED_TYPES
20:49:14 mriedem which changes our enum list in nova
20:50:48 mriedem this https://github.com/openstack/nova/blob/master/nova/objects/fields.py#L430
20:51:17 mriedem swebster: to find out, import cursive 0.1.2 in centos and xenial and compare signature_utils.SignatureKeyType.REGISTERED_TYPES.keys()
20:54:32 openstackgerrit Matt Riedemann proposed openstack/nova-specs master: Deprecate multinic, os-virtual-interfaces, and floating IP action APIs https://review.openstack.org/384261
20:54:41 mriedem sdague: jaypipes: mordred: alex_xu: johnthetubaguy: ^ there you go
20:54:45 mriedem big bag of deprecation
20:55:09 jaypipes mriedem: cool.
20:55:29 swebster mriedem: Very right. Centos: ['RSA-PSS', 'ECC_SECP521R1', 'ECC_SECP384R1', 'DSA']. Ubuntu: ['RSA-PSS', 'ECC_SECP521R1', 'ECC_SECT409R1', 'ECC_SECT409K1', 'ECC_SECT571R1', 'ECC_SECP384R1', 'ECC_SECT571K1', 'DSA']
20:56:46 mriedem swebster: that's super fun
20:56:51 mriedem they used to be hardcoded https://review.openstack.org/#/c/351232/16/nova/objects/fields.py
20:56:56 mriedem before we were using cursive
20:57:40 mriedem swebster: please report a bug
20:57:48 mriedem https://bugs.launchpad.net/nova/
20:57:51 swebster mriedem: And now it looks like they could be different system to system. Will do, and thanks
20:58:25 mriedem yeah that sucks, and i was afraid of something like that happening
20:58:33 mriedem we should probably just hard-code the superset
21:01:46 mriedem melwitt: so you hit "sphinx.errors.SphinxWarning: WARNING: no "man_pages" config value found; no manual pages will be written" right?
21:01:49 mriedem did you just recheck it?
21:02:04 melwitt mriedem: cdent rechecked it and it failed again
21:03:07 mriedem well that's just great
21:03:12 mriedem i'm going to go eat my feelings
21:13:22 dansmith last time I tasted feelings they had a distinct "like butt" quality
21:13:33 dansmith but, I'm not very adventurous with my mouth
21:13:39 dansmith er, tastes
21:25:02 mriedem these feelings tasted like pretzels
21:36:47 mriedem ok, on to writing a spec to deprecate os-hosts now?!
21:48:42 dansmith mriedem: uuid should be nullable on the model but not on the object right? https://review.openstack.org/#/c/453025/6/nova/objects/service.py
21:49:20 dansmith AFAICT, there's no need for it to actually be None anywhere
21:49:28 dansmith but if it's nullable everything needs to be defensive
21:52:15 openstackgerrit Matt Riedemann proposed openstack/nova master: Add uuid to Service model https://review.openstack.org/453025
21:52:18 mriedem yes
21:52:20 mriedem you're right
21:52:48 mriedem compute nodes uuid is read-only https://github.com/openstack/nova/blob/master/nova/objects/compute_node.py#L55
21:52:50 mriedem but not nullable
21:53:14 mriedem aggregate uuid is nullable=False, but not read-only https://github.com/openstack/nova/blob/master/nova/objects/aggregate.py#L244
21:53:34 dansmith yeah
21:53:48 dansmith there was a reason for the agg one not being read-only
21:53:56 mriedem so should be nullable=False and read_only=True?
21:54:17 dansmith tbh, just make it nullable=False (i.e. don't specify it)
21:54:22 dansmith the readonly has been problematic recently,
21:54:25 mriedem ok
21:54:37 dansmith so I'd leave that off, which is why it is in aggregate
21:57:11 openstackgerrit Eric Fried proposed openstack/nova master: PowerVM Driver: spawn/destroy #2: functional https://review.openstack.org/438598
21:57:35 efried mriedem ^^ back atcha.
21:58:01 efried Still a couple uncertainties inline.
21:59:26 openstackgerrit Matt Riedemann proposed openstack/nova master: Add online data migration for populating services.uuid https://review.openstack.org/454899
21:59:26 openstackgerrit Matt Riedemann proposed openstack/nova master: Add uuid to Service model https://review.openstack.org/453025
22:36:03 mriedem mgagne: http://lists.openstack.org/pipermail/openstack-dev/2017-April/115292.html
22:37:03 mriedem browne: efried: BobBall: would be good to check out https://review.openstack.org/#/c/453272/ and see if vmware, powervm and xen drivers can support this API eventually
22:37:08 mgagne mriedem: Thanks for starting the conversation and doing proper follow up
22:37:11 mriedem claudiub: you too but i think hyperv already does
22:38:09 mriedem mgagne: does the cinder extend volume api have a confirm / reject state like nova resize does?
22:38:39 mgagne mriedem: cinder doesn't have this confirmation system
22:39:27 mgagne mriedem: as you mention, there are pending questions: capabilities from virt driver. But also error handling in Nova since event system is async, Cinder can't query back for status.
22:39:56 mriedem mgagne: the nova api could fail fast in one case,
22:39:59 mriedem that the computes aren't new enough
22:40:06 claudiub mriedem: yeah, that's something i was going to comment on.
22:40:17 mriedem i.e. if we know the instance is running on an ocata compute, we know it's going to fail, and we could fail the request to cinder with a 409 or something
22:40:22 mgagne mriedem: this means a sync call to compute node? or could capabilities be cached somewhere in database?
22:40:29 claudiub it seems that it only talks about iscsi volumes.
22:40:37 mriedem mgagne: we have the nova-compute service version in the database
22:40:51 claudiub while for hyper-v we support iscsi, smb, and fibre channel volumes.
22:40:51 mriedem mgagne: so we know if the compute host is running ocata or pike code,
22:40:58 mgagne cool, didn't know about that
22:41:01 mriedem or what the minimum nova-compute is in the deployment
22:41:31 claudiub for smb, afaik, the hyper-v driver might have to do the volume resize. didn't read the whole spec to see if that's possible in nova
22:41:34 mgagne claudiub: yes, the proposed implementation is very "optimist" by assuming only libvirt+iscsi exists in its presumed perfect world =)
22:41:39 mriedem claudiub: yeah good point - mgagne: how do we know which brick connector to get? do we have to get the connection_info dict from cinder for the given volume?

Earlier   Later