Earlier  
Posted Nick Remark
#openstack-nova - 2017-11-22
12:41:18 openstackgerrit Rodolfo Alonso Hernandez proposed openstack/nova master: Format NIC features using os-traits definitions https://review.openstack.org/466051
12:42:05 openstackgerrit Rodolfo Alonso Hernandez proposed openstack/nova master: Format NIC features using os-traits definitions https://review.openstack.org/466051
12:58:22 gmann nova API meeting on openstack-meeting-4
13:10:19 FoXMaN hello
13:11:06 FoXMaN i need some advice on how to determine what slows down nova-api-os-compute service as i am out of ideas
13:11:25 FoXMaN request takes sometimes more than 50 seconds
13:12:23 alex_xu stephenfin: bauzas I'm in the bugsmash, there are new contributors, and there is the list of patches people are looking for review https://etherpad.openstack.org/p/OpenStack-Bug-Smash-Queens-Wuhan-Bugs-List, appreciate if you can review them!
13:13:34 alex_xu or any core is still working, but I guess not
13:18:04 mdbooth stephenfin: Will try to look at that one, too. Found an issue with the first, btw, but haven't reviewed it properly yet.
13:35:27 openstackgerrit Yikun Jiang proposed openstack/python-novaclient master: Use utils.prepare_query_string instead of duplicated code https://review.openstack.org/515640
13:35:45 mdbooth stephenfin: Just looking at that second patch. os-brick's remotefs connector avoids a whole class of races because it *never unmounts anything*.
13:36:14 openstackgerrit Silvan Kaiser proposed openstack/nova master: Added mount fstype based validation of Quobyte mounts https://review.openstack.org/522245
13:37:04 jaypipes morning supernovas
13:37:05 mdbooth Ah, until recently it seems this driver was VZStorageRemoteFSClient
13:37:34 mdbooth morning jaypipes
13:40:27 stephenfin mdbooth: So the manager that's there only cares for connections?
13:40:33 stephenfin *connecting
13:40:42 stephenfin on account of the "never disconnecting" anything
13:40:50 mdbooth Yep
13:40:50 stephenfin *anything" bit
13:40:57 stephenfin Ah, that makes sense
13:41:04 mdbooth Well, not really
13:41:17 mdbooth But it does make the locking simpler
13:41:55 mdbooth The complexity of my patch is in ensuring we never disconnect anything that's still in use.
13:42:20 mdbooth If you never disconnect anything ever, that's a resource consumption issue, but you aren't going to have that kind of race
13:45:10 stephenfin mdbooth: I'm a little confused here so let's recap
13:45:18 stephenfin That driver is using os-brick, and disconnect for that is a no-op for os-brick
13:46:36 mdbooth Yeah, it's using remotefs in os-brick
13:46:48 mdbooth remotefs is basically mount/unmount
13:46:49 stephenfin But I'm guessing the NFS driver does not use that, otherwise you wouldn't have had to add the manager in commit 4aa39c44a4b08ee4e05548d5c258e795089b2bdd
13:46:54 mdbooth Except it doesn't have unmount
13:47:10 stephenfin Well, you would for connect but not disconnect
13:47:13 mdbooth stephenfin: Indeed, NFS doesn't use it.
13:47:29 mdbooth And tbh, I don't think anything should use it, but it's their driver.
13:47:42 mdbooth Not without unmount, anyway
13:48:34 mdbooth The nova mount manager stuff is really generic. If it fit into the os-brick model it should probably live there.
13:48:36 stephenfin OK. So instead of them doing what the do there, would it make more sense for them to change LibvirtVZStorageVolumeDriver to subclass from LibvirtMountedFileSystemVolumeDriver
13:48:45 stephenfin *what they do
13:48:51 jianghuaw bauzas, I created a new patch to enable only one vGPU type on each compute node. So that we will have same behavior for both libvirt and XenAPI befre we switch to use n-rp. https://review.openstack.org/#/c/521769/
13:49:04 mdbooth I recommended that in a review comment. I won't leave a -1 for that, though.
13:49:12 stephenfin Like you did for the NFS driver in 4aa39c44a4b08ee4e05548d5c258e795089b2bdd
13:49:23 stephenfin OK. To be honest, that sounds like a viable -1 to me
13:49:42 bauzas jianghuaw: ack
13:49:43 stephenfin I mean, we have a good pattern for how to do this. Why not insist we use the same pattern everywhere
13:49:43 mdbooth Well the problem already exists
13:49:49 mdbooth And the patch they've proposed makes it better
13:49:55 bauzas jianghuaw: maybe stacking that with the other changes could help us reviewing ?
13:50:02 mdbooth The time to -1 was a while back
13:50:42 jianghuaw bauzas, That's in the PS stack. I'd like you help to reivew it.
13:51:17 jianghuaw bauzas, note I kept the original logic which will handle multiple gpu group, so it's easy for me to switch to multiple types once n-rp is ready.
13:51:34 bauzas jianghuaw: sure, but I mean squashing
13:51:58 stephenfin mdbooth: I assume it would require significant rework?
13:52:30 jianghuaw bauzas, ok. I see. Will squash it.
13:52:57 mdbooth stephenfin: Well I don't know enough about how it mounts, but if it's basically just a wrapper round 'mount -t vzstorage' or whatever, it should be very simple.
13:53:26 stephenfin Oh, so we'd be doing it ourselves instead of relying on os-brick?
13:53:52 mdbooth Yeah, the existing code doesn't use os-brick because it was replacing code which didn't use os-brick.
13:54:00 mdbooth There's no reason it couldn't, though.
13:54:09 mdbooth That would require a bit of rework, though.
13:54:43 stephenfin Ah, that's where I was confused. I thought it was a case of removing that new code, changing what we subclassed, and one or two other minor things
13:54:53 mdbooth Still, the os-brick connector doesn't implement umount, so I'm surprised anybody would use it.
13:55:20 stephenfin If there's other rework there then it makes sense to not ask for that. It's a nice-to-have for future
13:55:26 stephenfin Agreed. That seems...odd
13:56:35 openstackgerrit Matt Riedemann proposed openstack/nova master: Adding NVMEoF for libvirt driver https://review.openstack.org/482640
14:06:45 stephenfin mdbooth: Overall though, are you happy with that patch?
14:08:09 bauzas jianghuaw: I plan to write some functional tests for verifying our implementations
14:08:18 bauzas jianghuaw: it takes me a long time for that tho
14:08:37 bauzas but once I'll upload them, it could be nice to use them
14:09:06 mdbooth stephenfin: It serialises calls to connect, so it probably solves the reported issue.
14:09:19 mdbooth However, if they implement umount, which they should, it's insufficient.
14:10:22 mdbooth stephenfin: So I think my +0 accurately reflects my feelings.
14:10:47 jianghuaw bauzas, do you mean adding tests in tempest?
14:11:00 stephenfin mdbooth: That sounds fair. I'd edge towards +2 because it _does_ resolve the bug
14:11:07 openstackgerrit Radoslav Gerganov proposed openstack/nova master: VMware: expose max vCPUs and max memory per ESX host https://review.openstack.org/516635
14:11:07 openstackgerrit Radoslav Gerganov proposed openstack/nova master: VMware: fix memory stats https://review.openstack.org/516634
14:11:08 stephenfin Though I would much rather they implement disconnect also
14:11:08 openstackgerrit Radoslav Gerganov proposed openstack/nova master: VMware: implement get_inventory() driver method https://review.openstack.org/506175
14:11:22 stephenfin ...allowing them to use that superclass in the process
14:11:34 bauzas jianghuaw: nope, just in nova.tests.functional
14:11:48 mdbooth stephenfin: Ok.
14:14:00 jianghuaw bauzas, ah, got it. Thanks for letting me know of it.
14:14:11 jianghuaw bauzas, yes. that's needed.
14:14:15 openstackgerrit Merged openstack/nova master: Allow shuffling hosts with the same best weight https://review.openstack.org/494136
14:15:58 openstackgerrit Matt Riedemann proposed openstack/python-novaclient master: Add support for microversion 2.55 - flavor description https://review.openstack.org/520819
14:33:00 openstackgerrit Rodolfo Alonso Hernandez proposed openstack/nova master: Format NIC features using os-traits definitions https://review.openstack.org/466051
14:35:14 ralonsoh stephenfin: why the -1 in https://review.openstack.org/#/c/466051/?
14:35:20 stephenfin bauzas: I think what ralonsoh was doing here is correct https://review.openstack.org/#/c/466051/
14:35:26 stephenfin ralonsoh: Just left the comment
14:35:26 openstackgerrit Balazs Gibizer proposed openstack/nova master: Moving more utils to ProviderUsageBaseTestCase https://review.openstack.org/499539
14:35:43 stephenfin Quick -1 to stop it getting merged :)
14:36:30 bauzas stephenfin: well, there are 2 possibilities
14:37:02 bauzas stephenfin: #1 either we wait for a specific nova global requirements update and then we don't need to ask ralonsoh to modify directly the reqs.txt file
14:37:38 bauzas stephenfin: #2 or ralonsoh depends-on his change to https://review.openstack.org/#/c/521883/ and amending directly reqs.txt
14:38:43 stephenfin bauzas, ralonsoh: Hmm, yeah I was thinking the requirements job might fail
14:39:33 bauzas stephenfin: the reqs job failed because the g-r patch is not yet merged
14:39:56 bauzas so ralonsoh can modify reqs.txt directly if he provides a Depends-On
14:40:10 stephenfin ralonsoh: That job checks to make sure requirements are synchronized between projects so, for example, nova doesn't depend on requirements '==0.1.0', while neutron depends on '==0.2.0'
14:40:21 stephenfin Because if we did that, then we couldn't install them side-by-side
14:40:35 openstackgerrit Balazs Gibizer proposed openstack/nova master: factor out compute service start in ServerMovingTest https://review.openstack.org/503037
14:40:45 openstackgerrit Balazs Gibizer proposed openstack/nova master: Test resource allocation during soft delete https://review.openstack.org/495159
14:40:50 stephenfin At least, not using system packages (vs. virtualenvs or installing things in containers)

Earlier   Later