Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2022-07-27
13:32:39 gibi in the past we said we would at least like to see the client patches proposed (not merged) before we approve a microversion
13:32:45 gibi Uggla: awesome
13:32:49 sean-k-mooney ill be dropping soon but ill see if i have time to review before i do
13:33:32 opendevreview Kashyap Chamarthy proposed openstack/nova stable/ussuri: Add a workaround to skip hypervisor version check on LM https://review.opendev.org/c/openstack/nova/+/851224
13:34:22 Uggla gibi: https://review.opendev.org/c/openstack/python-novaclient/+/831651 and https://review.opendev.org/c/openstack/python-openstackclient/+/831902
13:34:33 gibi Uggla: thanks
13:35:03 Uggla gibi, there is also a related tempest test: 841088: Tempest test for checking unshelve to host | https://review.opendev.org/c/openstack/tempest/+/841088
13:35:16 gibi ack
13:38:26 opendevreview Kashyap Chamarthy proposed openstack/nova stable/train: Add a workaround to skip hypervisor version check on LM https://review.opendev.org/c/openstack/nova/+/851226
13:40:21 kashyap bauzas: gibi: If you have the stomach for it today still (other backports): https://review.opendev.org/q/Iec387dcbc49ddb91ebf5cfd188224eaf6021c0e1
13:40:41 bauzas gibi: kashyap: was working on the novaclient change
13:40:49 kashyap No worries
13:46:31 opendevreview Merged openstack/nova master: Add a workaround to skip hypervisor version check on LM https://review.opendev.org/c/openstack/nova/+/851034
14:04:35 gibi Uggla: I left feedback in the python-novaclient patch and +1d the osc patch. nice work!
14:05:14 Uggla gibi, ok I'll have a look ASAP.
14:12:57 opendevreview Sylvain Bauza proposed openstack/python-novaclient master: Add support for 2.92 : keypair import mandatory https://review.opendev.org/c/openstack/python-novaclient/+/851231
14:13:02 bauzas gibi: ^
14:13:09 gibi bauzas: thanks
14:13:13 bauzas now working on the OSC patch
14:14:14 gibi bauzas: you probably want to put your novaclient patch top of Uggla's
14:14:33 bauzas gibi: there is a depends-on my api patch
14:14:39 bauzas gibi: but I can rebase it
14:15:00 gibi yeah, keep the depends on, but rebase your client patch on Uggla's client patch
14:15:08 gibi otherwise you will be in merge conflict
14:16:39 bauzas gibi: cool, yeah I know about the merge conflict but the fix will be simple
14:16:51 gibi yepp
14:16:59 gibi othervise your client patch looks good to me
14:20:00 opendevreview Sylvain Bauza proposed openstack/python-novaclient master: Add support for 2.92 : keypair import mandatory https://review.opendev.org/c/openstack/python-novaclient/+/851231
14:20:19 bauzas gibi: done
14:20:55 gibi bauzas: thanks, reapplied my +@
14:20:56 gibi +2
14:21:11 bauzas thanks
14:21:44 gibi your home has hair? ;)
14:45:49 bauzas gibi: sorry, haircut *at* home :p
14:46:25 bauzas my wife and I have a hairdresser visiting uds
14:46:30 gibi ohh, nice
14:46:39 gibi my wife tend to cut mine
14:46:39 bauzas cheaper and better
14:46:55 gibi that is the cheapest :)
14:47:28 bauzas yeah but if I use a hair mower, then my hair will have issues :)
14:48:19 bauzas I prefer to have a pro using scissors
15:48:14 artom gibi, so rollback_live_migration_at_destination() is the cast that contains the mutated migrated context, right?
15:48:26 artom So what if we just change it to a call instead of cast to make it blocking?
15:48:41 artom That way, we're sure that by the time we drop the claim, we're out of the mutated context?
15:49:16 gibi that is a dirty thing that will mean we will have a short time window where the db contains the wrong info, then then the second call will return the db content to the correct state
15:49:23 gibi so we will have a smaller race window
15:49:26 gibi but we still have the race
15:50:02 artom You're assuming that mutating the context saves to the DB?
15:50:05 artom I don't think that's the case
15:50:16 gibi we mutate the context and then save the instance today
15:50:25 gibi just the mutate is not an issue
15:50:57 gibi the instance.save happens in _cleanup in the driver
15:51:06 gibi and that is run under the mutated context
15:52:55 gibi links to the code are here https://bugs.launchpad.net/nova/+bug/1982497/comments/2
15:53:08 artom gibi, oh, right
15:53:32 artom We mutate the context in the compute manager, and in that context we eventually call down to instance.save() in the libvirt driver
15:53:34 gibi hence my second patch that forbids instance.save under the mutated context to avoid this in the fututre
15:53:41 gibi artom: yes
15:54:00 artom Wait, so why did you jump through the sync loops in your reproducer test?
15:54:05 artom The Condition(), for example?
15:54:56 gibi I needed a special serialization order to make sure that we avoid the case when we save the wrong state, then right after we save the good state.
15:55:25 artom Ah
15:55:53 gibi _rollback_live_migration makes the DB invalid then drop_move_claim_at_destination fixes it.
15:56:31 gibi so in that order it is just a small race window (same as if we would make _rollback_live_migration a call instead of a cast)
15:57:24 gibi interestingly in a real environment _rollback_live_migration tend to take longer and finishing _after_ drop_move_claim_at_destination probably due to _rollback_live_migration doing disk IO that can be slow
15:58:34 artom I mean we could just remove the instance.save() from the cleanrup()...
15:58:41 artom And count on the compute manager doing it later...
15:58:43 artom Sounds fragile
15:59:03 gibi artom: yepp that is one way, we can do that and then figure out the fallout
16:00:04 gibi it seems we have that instance.save as delete_instance_files() can fail and we want to retry
16:00:11 gibi later in a periodic if that fails
16:02:06 gibi unfortunately _cleanup not just set instance.cleaned but also init instance.system_metadata['clean_attempts'] that is used in _run_pending_deletes
16:02:56 gibi and honestly not 100% sure if that is the only instance update we do during the full destroy codepath
16:03:34 gibi I think that is the only instance.save but maybe other pieces of code also changes the instance that we would loose if we remove the instance.save
16:08:07 melwitt gmann: fyi the tempest tests for the volume extend coverage have merged, so I think the nova patch to start running the tests is ready for review https://review.opendev.org/c/openstack/nova/+/843700
16:08:29 gmann melwitt: checking
16:08:37 melwitt thanks!
16:57:42 sean-k-mooney im going to go nap soon i think but i might review the keypair change before or after
16:57:53 sean-k-mooney is that the next microverion in the queue
17:14:30 gibi t is
17:14:31 gibi it is
17:14:49 bauzas I'm just about to upload the OSC patch in 5 mins
17:14:57 bauzas after creating the relnote
17:15:10 bauzas so we'll have all the meat for 2.92 except the tempest tests
17:15:27 sean-k-mooney[m] ok cool
17:21:57 bauzas WTF
17:22:20 bauzas [sbauza@sbauza python-openstackclient]$ git review -s
17:22:20 bauzas Problems encountered installing commit-msg hook
17:22:20 bauzas The following command failed with exit code 255
17:22:20 bauzas "scp -P29418 sbauza@review.opendev.org:hooks/commit-msg .git/hooks/commit-msg"
17:26:13 bauzas ok, fixed by following https://www.mediawiki.org/wiki/Gerrit/Alternatives_to_git-review
17:27:28 bauzas there it goes :
17:27:31 bauzas gibi: sean-k-mooney: https://review.opendev.org/q/topic:bp%252Fkeypair-generation-removal
17:27:44 bauzas all the changes are up
17:42:29 sean-k-mooney[m] gmann: holding +w to leave bauzas respond to your comments
17:42:49 sean-k-mooney[m] i think they could be adressed in a followup
#openstack-nova - 2022-07-28
01:39:59 opendevreview melanie witt proposed openstack/nova master: imagebackend: Cache the key manager when disk is encrypted https://review.opendev.org/c/openstack/nova/+/826756
04:34:51 opendevreview OpenStack Proposal Bot proposed openstack/nova master: Imported Translations from Zanata https://review.opendev.org/c/openstack/nova/+/851337
06:09:11 gibi good morning
08:39:31 bauzas sean-k-mooney: gmann: gibi: thanks for your comments on https://review.opendev.org/c/openstack/nova/+/849133 will respin

Earlier   Later