Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-01-17
15:29:55 mriedem although let's say we completed cell1, set a marker, and then a new instance was created in cell1, we would check that one again
15:30:49 maciejjozefczyk you're right above, but we basically fill vif list for instances, that are created in releases older than newton, so for this use-case we shouldn't care
15:31:02 mriedem ok, but you understand my concern for cern right? they have 70+ cells, and if they run online data migrations per cell (which i'm assuming they have to do given some migrations are per-cell), then this would churn a lot of data needlessly on the other cells
15:31:12 mriedem belmoreira: can you confirm how you run online_data_migrations? ^
15:31:17 sean-k-mooney mriedem: at the point at wich the online data migration is running have you upgraded to the new code or are teh cells still running the old code
15:31:47 sean-k-mooney mriedem: if they are using the new code the the new instance would not need the data migration right?
15:31:47 maciejjozefczyk mriedem: I understand your point, and yeah, I can do it because its easy :)
15:31:48 mriedem sean-k-mooney: either
15:32:07 mriedem maciejjozefczyk: sorry, do what exactly?
15:32:12 maciejjozefczyk but I don't think its needed anyway
15:32:24 maciejjozefczyk mriedem: do a marker per cell, to track new instances in cell after migrate
15:33:50 maciejjozefczyk cern shouldn't have anything to migrate by this script, due they don't have instances older than newton I think
15:34:18 maciejjozefczyk only OVH has cattle-like instances :)
15:34:37 mriedem you mean pet-like
15:34:42 mriedem i know huawei public cloud has a lot of pets
15:34:45 maciejjozefczyk yup
15:34:46 mriedem and old instances
15:35:36 maciejjozefczyk so this will 'scan' all the instances in all cells sequentially and after it one new code nothing will be migrated
15:37:52 openstackgerrit Merged openstack/nova master: api-ref: Body verification for the lock action https://review.openstack.org/622835
15:38:57 openstackgerrit Merged openstack/os-vif master: Apply workaround to host_info serialization test https://review.openstack.org/631173
15:47:15 mriedem maciejjozefczyk: ok i certainly don't want to hold this up. i think we need a test for the marker adjustment across cells (i can help with that if needed) and add more context to the release note (i left a suggestion), and it would be great to get some buy-in from belmoreira but again i don't want to hold it up
15:47:35 mriedem i appreciate you working on this
15:48:14 mriedem i sent an email to one of our operators internally just to get him to take a look also and make sure it's ok with him
15:48:20 mriedem (i might not get a response...)
15:48:25 maciejjozefczyk mriedem: Thanks for your time :) I'll do like you suggest.
15:50:05 maciejjozefczyk btw I did migration on one of my db with 40k instances, I did about 7k migrations, so a lot of pets
15:53:22 mriedem maciejjozefczyk: cool. ovh has multiple regions but not multiple cells yet right?
15:53:35 mriedem and you're working on collapsing some regions into multiple cells?
15:54:09 belmoreira mriedem maciejjozefczyk what's the patch that you are discussing?
15:54:37 belmoreira mriedem yes, data migrations are per cell
15:54:48 maciejjozefczyk mriedem: we don't use cells because neutron is not cell-aware :(
15:54:52 gryf belmoreira, https://review.openstack.org/#/c/614167
15:55:12 maciejjozefczyk mriedem: so wi'll have more issues because of neutron
15:55:29 mriedem neutron has routed networks if that helps
15:55:43 mriedem but nova isn't really aware of routed networks for scheduling...
15:56:09 maciejjozefczyk oh, so as you see :D
15:57:28 maciejjozefczyk but its mostly related to our network architecture, we created own neutron ml2 plugin along with network_types that are different… and more complicated
16:02:28 mriedem maciejjozefczyk: heh vif get by instance uuid will never return deleted instances regardless of the context :) https://github.com/openstack/nova/blob/4f5cb30a2f88965772bcbff0a19a1dcfcd3836c6/nova/db/sqlalchemy/api.py#L1542
16:02:31 mriedem jaypipes: ^
16:02:39 mriedem b/c the db api hard-codes it
16:02:43 mriedem \o/
16:05:04 maciejjozefczyk ^^
16:05:40 maciejjozefczyk ;D
16:06:44 efried jaypipes, cdent, mriedem: Pretty sure I've got the CERN stack down to only spurious timeout failures at this point. Will y'all be able to take a look today?
16:07:10 cdent efried: aye
16:07:39 efried thx
16:08:45 jaypipes mriedem: lol
16:08:50 mriedem idk, got some other stuff i need to deal with first
16:09:00 mriedem including the tasks i signed up for yesterday
16:09:08 jaypipes efried: sorry, the CERN stack?
16:09:16 mriedem the pile seems to grow faster than i can dig out of it
16:09:29 mriedem jaypipes: the reduce placement api calls series
16:09:43 mriedem provider tree hullabaloo 2019
16:09:51 jaypipes oh, that's the CERN stack? :) yes, efried I will be reviewing it all again today
16:14:28 maciejjozefczyk bb guys, grabbing a beer
16:46:11 openstackgerrit Balazs Gibizer proposed openstack/nova master: Add functional test for libvirt vgpu reshape https://review.openstack.org/631559
16:46:17 gibi mriedem: ^^
16:48:35 gibi mriedem: there are some raw edges but I thin it is not horrific
16:49:01 mriedem gibi: great, thanks for doing that
16:49:03 mriedem <3
16:50:00 gibi mriedem: it was a nice challenge :)
16:50:34 mriedem nice to do something other than bw provider i imagine :)
16:53:30 gibi mriedem: yeah, I'm pretty much waiting for the decision which direction to go with the bwm
17:10:29 mriedem watch out
17:10:43 openstackgerrit Matt Riedemann proposed openstack/nova master: Isolate cell-targeting code in MigrationTask https://review.openstack.org/621310
17:10:44 openstackgerrit Matt Riedemann proposed openstack/nova master: Extract compute API _create_image to compute.utils https://review.openstack.org/621311
17:10:44 openstackgerrit Matt Riedemann proposed openstack/nova master: Add Migration.cross_cell_move and get_by_uuid https://review.openstack.org/614012
17:10:45 openstackgerrit Matt Riedemann proposed openstack/nova master: Add Destination.allow_cross_cell_move field https://review.openstack.org/614035
17:10:45 openstackgerrit Matt Riedemann proposed openstack/nova master: Add InstanceAction/Event create() method https://review.openstack.org/614036
17:10:46 openstackgerrit Matt Riedemann proposed openstack/nova master: Change HostManager to allow scheduling to other cells https://review.openstack.org/614037
17:10:46 openstackgerrit Matt Riedemann proposed openstack/nova master: Add CrossCellWeigher https://review.openstack.org/614353
17:10:47 openstackgerrit Matt Riedemann proposed openstack/nova master: Add can_connect_volume() compute driver method https://review.openstack.org/621313
17:10:47 openstackgerrit Matt Riedemann proposed openstack/nova master: Add Instance.hidden field https://review.openstack.org/631123
17:10:48 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Add TargetDBSetupTask https://review.openstack.org/627892
17:10:48 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Add PrepResizeAtDestTask https://review.openstack.org/627890
17:10:49 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Add PrepResizeAtSourceTask https://review.openstack.org/627891
17:10:49 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Add CrossCellMigrationTask https://review.openstack.org/631581
17:10:56 cdent jeebus
17:11:16 mriedem i believe gibi's is more impressive
17:20:06 mriedem dansmith: melwitt: are you aware of any test utilities to compare to objects and ignore some fields? I found nova.tests.objects.test_objects.compare_obj but that takes an object and a primitive and doesn't take a blacklist method for fields to ignore
17:20:34 melwitt I thought we do. /me looks
17:20:36 dansmith hmm, I thought there was one
17:20:53 dansmith you could just del the fields you don't want of course
17:22:50 adrianc sean-k-mooney: ^
17:23:11 adrianc whoops :) disregard
17:23:36 melwitt mriedem: this is the thing I have used before https://github.com/openstack/nova/blob/master/nova/tests/unit/db/test_db_api.py#L1395
17:23:53 mriedem yeah just found this https://github.com/openstack/nova/blob/master/nova/tests/functional/db/test_console_auth_token.py#L47
17:24:13 mriedem that db api util is more like what i'm looking for
17:57:14 cdent efried: lgtm
18:15:13 openstackgerrit Maxim Nestratov proposed openstack/nova master: libvirt: Virtuozzo containers config drive support https://review.openstack.org/449818
18:20:05 openstackgerrit Rodolfo Alonso Hernandez proposed openstack/os-vif master: Import IP implementation modules outside privsep context https://review.openstack.org/629025
19:06:14 cfriesen I'm trying to work out the logic in the image-fetching and caching code on the compute node but it's not easy to follow. If the glance image is stored in ceph and nova-compute is using local storage, will the image data be pulled down via the glance API or can nova short-circuit it and pull it in directly from ceph?
19:10:22 mriedem cfriesen: https://review.openstack.org/#/c/572805/ might be related
19:11:11 cfriesen mriedem: how apropos
19:43:18 efried mriedem: So close to using TaskFlow, remind me why we aren't doing that?
19:44:17 mriedem efried: b/c i've never taskflowed?
19:45:04 mriedem learning a new toolset is kind of real low priority when i have a mountain to climb first, but i know
19:45:12 efried I just mean, looking at TaskBase, it's essentially the same as a TaskFlow Task - execute and rollback methods - but we had to write the plumbing for things like doing the actual rollbacks or whatever.
19:45:39 mriedem task flow also has a lot of threading stuff in it though doesn't it?
19:45:41 efried more just wondering why it wasn't used in the first place. I get following momentum

Earlier   Later