Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-01-25
09:48:22 sfinucan Actually, do we still bother with those 'backport' lines? I figured every bug should be backported, if it applied to the stable code too and didn't introduce a functional change
09:48:29 sfinucan lyarwood: (y)
09:48:33 lyarwood sfinucan: http://docs.openstack.org/project-team-guide/stable-branches.html#proposing-fixes for the actual workflow
09:48:36 openstackgerrit Duong Ha-Quang proposed openstack/nova: Remove mox in nova/tests/unit/compute/test_shelve.py (3) https://review.openstack.org/342648
09:49:52 sfinucan lyarwood: Excellent. Thank you. Think I've done it now (https://review.openstack.org/425087)
09:50:10 sfinucan Hmm, commits to stable branches don't appear in IRC. Interesting...
09:51:38 sfinucan duonghq: Thanks for respinning that. Appreciate it :)
09:52:00 duonghq sfinucan, thank you very much
10:23:13 openstackgerrit Stephen Finucane proposed openstack/nova: adding debug info for pinning calculation https://review.openstack.org/355150
10:36:56 openstackgerrit Stephen Finucane proposed openstack/nova: Add PCIWeigher https://review.openstack.org/379524
10:37:53 openstackgerrit Stephen Finucane proposed openstack/nova-specs: Remove templates from approved/implemented specs https://review.openstack.org/405550
10:44:30 openstackgerrit Huan Xie proposed openstack/nova: Fix live migrate with XenServer https://review.openstack.org/424428
10:58:09 bhagyashris Hi all, I have one question, In the multinode environment , Is there a need to enable the placement api service on compute node also?
10:58:35 openstackgerrit Vasyl Saienko proposed openstack/nova: Implement interface attach/detach in ironic virt driver https://review.openstack.org/419975
10:59:11 johnthetubaguy bhagyashris: everything talks to a single placement API endpoint (as defined in the keystone service catalog), over HTTP, if that helps.
11:01:06 BobBall johnthetubaguy: Regarding https://review.openstack.org/#/c/424428/3/nova/objects/migrate_data.py is the "legacy dictionary" used anywhere now? i.e. should the to_legacy_dict method care about new fields added to the object?
11:01:42 johnthetubaguy BobBall: I suspect now both old a new nodes are sending objects, we can drop all that code
11:02:11 BobBall Rather than dropping in this patch I was assuming not bothering updating the legacy dict?
11:02:13 johnthetubaguy BobBall: although technically speaking, we should bump the compute API major version to make that explicit
11:02:42 BobBall I presume there should be a more global legacy-dictionary cleanup effort? rather than a per-object piecemeal approach?
11:03:27 BobBall bah, you know what, the easiest path is just to update the dict anyway. Makes sure that if there is any dodgy code anywhere that it'll all work...
11:03:33 johnthetubaguy BobBall: I like being consistent and updating that, if I am honest, but I have no functional reason, just prettyness
11:04:52 ildikov johnthetubaguy: hi. can you look into this Doodle poll please: http://doodle.com/poll/9per237agrdy7rqz ?
11:06:00 ildikov johnthetubaguy: I know you have collisions around the time slot I hope to keep on other days, but Monday is unfortunately not working for a few Cinder guys :(
11:07:34 johnthetubaguy ildikov: so its not too bad until we high daylight savings, when many of the other days are tricky
11:07:58 johnthetubaguy ildikov: but hopefully we will have complete all our work by then :)
11:08:02 ildikov johnthetubaguy: we can reconsider when we switch again
11:08:09 johnthetubaguy that too
11:08:17 ildikov johnthetubaguy: hehe, I like your optimism! :)
11:08:35 ildikov johnthetubaguy: I just would like to have a slot that works for most so we can prep for the PTG
11:09:12 ildikov johnthetubaguy: so that we can complete all our work before the next daylight savings switch ;)
11:09:49 ildikov johnthetubaguy: great, thank you!
11:10:31 ildikov johnthetubaguy: I will harass a few Cinder folks and it seems we will most prolly switch back to Thursdays for a bit, will announce this week I hope
11:16:40 bauzas cdent: mmm, there is a bug with https://github.com/openstack/nova/blob/master/nova/scheduler/client/report.py#L129-L131
11:16:48 bauzas cdent: swap is counted in MB
11:16:57 bauzas cdent: https://github.com/openstack/nova/blob/master/nova/scheduler/filters/disk_filter.py#L36-L38
11:17:14 bauzas so we don't report correctly the right size
11:17:20 cdent whoops
11:17:30 bauzas just creating a bug report
11:17:44 cdent nice find
11:20:45 mdbooth bauzas: I've hit that a bunch of times, btw. In general, I'd prefer to see a coding style guideline that all storage sizes are in bytes.
11:21:06 mdbooth We variously use B, MB and GB throughout the code, and it's super-easy to make mistakes.
11:21:34 cdent it's good ux for it be other things for humans, but yeah, as soon as the data passes into internal code it ought to be all the same
11:22:06 bauzas mdbooth: sure but changing the Flavors attributes seems a bit long to do
11:22:08 mdbooth Yep
11:22:31 mdbooth bauzas: Oh, yeah. I think we need to live with our existing mistakes :) Just not make new ones.
11:23:26 mdbooth i.e. Any new code which represents a storage size should use bytes exclusively.
11:27:27 openstackgerrit Matthew Booth proposed openstack/nova: libvirt: Mock is_job_complete in test_driver https://review.openstack.org/425140
11:27:29 mdbooth lyarwood: ^^^
11:28:04 mdbooth lyarwood: Could you take a quick look at the above change. If you approve, I would rebase your mirror attribute fix on top of it and remove all the changes in test_driver.py
11:29:46 bauzas mdbooth: agreed
11:29:55 bauzas mdbooth: it should count on bytes
11:30:00 bauzas cdent: FYI https://bugs.launchpad.net/nova/+bug/1659266
11:30:01 openstack Launchpad bug 1659266 in OpenStack Compute (nova) "Disk allocation for instance is not good with swap" [High,Confirmed]
11:30:08 cdent ✔
11:30:26 lyarwood mdbooth: thanks, yeah that looks good if people are happy for us to mix mox and mock, I assumed there was an unwritten rule stopping us from doing this
11:31:05 mdbooth lyarwood: I think we don't write new mox. Partial mox->mock is acceptable, I believe?
11:31:08 bauzas cdent: in theory, we would need to data migrate the existing allocations
11:31:39 mdbooth I mean going the whole hog is great. The main change is from functional->unity there, rather than mox->mock
11:31:42 bauzas cdent: are we wiping off the existing allocations somewhere when we auto-heal ?
11:32:16 lyarwood mdbooth: yeah, I might stash what I have and follow up with the full move to mock after this
11:32:17 mdbooth lyarwood: As I mention in the commit, we should either have a test library for this, or mock at the function boundary.
11:33:07 mdbooth The real change is mocking is_job_complete, rather than modelling the corresponding behaviour of the libvirt daemon.
11:33:25 mdbooth I prefer the latter, but we don't have that in place, and doing it piecemeal isn't maintainable.
11:33:53 cdent bauzas: we update allocations when is_new_instance or is_removed_instance
11:34:09 mdbooth lyarwood: Anyway, would you mind if I rebased your mirror tag change on to it?
11:34:24 mdbooth It would make your change a lot smaller. No changes required in test_driver.py.
11:34:26 cdent bauzas: we update inventory frequently
11:34:30 lyarwood mdbooth: not at all if you have the time to work on it
11:34:38 mdbooth Yeah, it'll only take me a second.
11:34:44 lyarwood mdbooth: awesome thanks
11:35:40 bauzas cdent: my concerns are if we have wrong allocations when a new instance comes in
11:36:16 cdent bauzas: yeah. I haven't looked closely enough. Is the bug such that the allocations will be over or under where they should be?
11:36:38 bauzas yup
11:36:50 bauzas but I need to bail out for lunch
11:37:04 bauzas if I have a swap of 1MB
11:37:23 bauzas root to 1GB and eph to 0
11:37:46 bauzas then I'll express an allocation of 2GB with that bug instead of 1.001
11:38:38 cdent When nova-compute restarts does it rewrite all allocations?
11:39:19 openstackgerrit Matthew Booth proposed openstack/nova: libvirt: Use the mirror element to detect job completion https://review.openstack.org/399735
11:39:22 mdbooth lyarwood: ^^^
11:39:37 mdbooth 75 insertions(+), 9 deletions(-) Much smaller :)
11:39:47 cdent bauzas: I'll do some experimenting in devstack today (because I was planning to do that anyway)
11:40:31 lyarwood mdbooth: nice!
11:41:00 mdbooth lyarwood: It's identical to your change, just test_driver changes removed
11:41:10 bhagyashris johnthetubaguy: i am new to placement api feature, Does it mean, the placement api only needs to be enabled on controller node ?
11:41:54 johnthetubaguy bhagyashris: I think so
11:44:12 lyarwood mdbooth: that's great, I would say thanks but my LOC changed stats just took a hit :D
11:44:21 cdent bhagyashris: yes, it goes on a controller and needs to be in the service catalog
11:44:35 mdbooth lyarwood: lol
11:45:17 cdent mdbooth++
11:45:21 lyarwood mdbooth: I agree https://review.openstack.org/#/c/391597/
11:45:30 lyarwood ^_^
11:47:43 mdbooth lyarwood: Oh, wow
11:47:55 mdbooth That's fantastic :)
11:51:49 mdbooth Gerrit is a bit 500-y this morning
11:53:31 bhagyashris cdent: thank you for information, however if I am having two node setup then how many resource_providers will be created? two or one? In my case it is creating single resource_provider only
11:54:00 cdent bhagyashris: you should get as many resource providers as you have compute nodes
11:54:38 cdent if you have two compute nodes but are only seeing one, then it may be that one of the compute nodes does not have the right stuff in the [placement] section of nova.conf
11:56:18 bhagyashris cdent: ok, so in both compute nodes I need to add placement section in nova.conf? actually I am using devstack for multinode setup so one one node it does not create placement section

Earlier   Later