Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2020-11-05
10:15:52 bauzas stephenfin: we have non-positional arguments that are unrelated to RPC versions
10:16:10 bauzas we just keep them optional
10:16:49 bauzas lyarwood: oh, TIL this page was existing
10:17:02 bauzas gibi: we fixed it by code reviews
10:17:49 bauzas ah, this is already documented https://docs.openstack.org/nova/latest/contributor/code-review.html#rpc-api-versions
10:17:53 gibi bauzas: sure, code review is the fallaback, human intelligence is king, but if we can automate it then we could avoid failing humans like me at the original code rview
10:18:12 lyarwood bauzas: ah cool
10:18:20 bauzas but I guess "The manager-side method needs to tolerate older calls as well as newer calls" is maybe too much overall, and we need to explain it more
10:18:47 bauzas gibi: we could enforce owners to propose functional tests
10:18:53 bauzas for testing the RPC pins
10:19:06 bauzas like I did in my regression test
10:19:29 bauzas this would be a simpliest approach
10:19:52 gibi I guess enforce by code review
10:20:01 bauzas that, yeah
10:20:06 gibi I agree
10:20:13 bauzas but from what I've seen, nobody is really doing it
10:20:16 gibi still I want to automate it if possible :D
10:20:39 lyarwood shouldn't we cover mixed compute upgrades in the multinode grenade job?
10:20:46 bauzas gibi: well, we don't really set new versions a lot right?
10:20:50 gibi becuase all are code review rules are as good as the way we enforce them
10:21:00 gibi bauzas: we do it less and less, I agree
10:21:01 bauzas like, we only had one rpc minor bump per release since a while
10:21:16 bauzas gibi: well, we have a code review documentation
10:21:24 bauzas and I expect cores to know it at least
10:21:39 bauzas I mean, that's a breaking change to accept a RPC change
10:22:43 bauzas maybe we rushed over accepting some feature that was long overdue, but maybe considering to require a functest would ensure that we would put the burden on code owners
10:22:45 gibi bauzas: you are correct that we assume that core reviews catch these kind of problems, but they don't as you found
10:23:16 gibi if we do these thing less and less it means that we will easier to forget what to look at in these changes
10:24:07 gibi as we don't excersize this knowledge
10:26:26 gibi so I agree that one thing is to raise awerness for this issue as you did.
10:27:16 gibi but also I will think about some kind of automation as I cannot promise I won't forget this rule again 6 months from now when we bump the next
10:44:06 bauzas gibi: ahah lol, i had to rush off home because I forgot my kids at the school :whoops:
10:44:38 bauzas gibi: fwiw, the change we merged was a bit hairy, so I do understand that it was difficult to find the problem
10:45:02 bauzas gibi: that's why I said we should at least ask to provide a functional test, that's it
10:45:41 gibi yeah, I should not forget to ask a functional test pining to old RPC version when a new RPC version is proposed
12:05:33 brinzhang0 gibi: hi good morning
12:05:59 brinzhang0 gibi: Hope you can review
12:06:00 brinzhang0 Cyborg shelve/unshelve support patch https://review.opendev.org/#/c/729563/ :D
12:06:33 gibi brinzhang0: add to my queue
12:06:42 gibi added
12:06:51 brinzhang0 gibi: thanks
13:36:01 bauzas brinzhang0: gibi: hah, this time the new argument is nullable :p
13:36:23 bauzas but maybe it's time to ask for a functional testclass verifying the RPC API ? :)
14:11:28 openstackgerrit Balazs Gibizer proposed openstack/nova master: Bump the lowest eventlet version to 0.26.1 https://review.opendev.org/761427
15:12:05 iurygregory Hi nova folks, a friend of mine using openstack queens asked me if it's possible to update the config drive of an instance?
15:12:31 bauzas gibi: the next nova meeting is in 45 mins, right?
15:12:36 bauzas tz change
15:14:35 gibi bauzas: yes, the meeting is at 16:00 UTC which is 17:00 CET
15:14:43 bauzas cool cool
15:14:48 bauzas nicer for us :)
15:15:23 gibi :)
15:18:40 sean-k-mooney openstack meetings are alwasy utc and never move
15:18:46 sean-k-mooney its other that do
15:19:57 sean-k-mooney fortunetlly DLS will not be a thing in europe after 2021
15:20:27 dansmith hopefully not on the west coast either, but it's not set yet
15:20:56 sean-k-mooney it was ment to happen this year but got delayed so this was ment to be the last switch
15:21:34 sean-k-mooney the current plan is contries adopting permenatn summer time will swap for the last time in the spring
15:21:51 sean-k-mooney and the rest will swap for the last time in the fall
15:32:15 gibi stephenfin, bauzas: spent some time thinking automating to catch bugs like https://bugs.launchpad.net/nova/+bug/1902925 . Besides code review (that fails some time like in this case) what we can do is to extend the grenade testing.
15:32:15 openstack Launchpad bug 1902925 in OpenStack Compute (nova) "Upgrades to compute RPC API 5.12 are broken" [Critical,In progress] - Assigned to Sylvain Bauza (sylvain-bauza)
15:32:36 gibi As far as I understand it run livemigration between mixed computes
15:32:41 bauzas sean-k-mooney: I'm against summer time
15:33:01 bauzas gibi: you need then two compute services
15:33:09 gibi bauzas: we have multinode grenade
15:33:10 bauzas and a rolling upgrade scenario
15:33:33 bauzas because the rpc pins will automatically set the version to the oldest compute one
15:33:42 bauzas (if set to 'auto')
15:34:01 gibi I think nova-grenade-multinode does what we need
15:34:11 bauzas and again, tbh, I wonder whether it's just a code review usage
15:34:46 gibi as per https://github.com/openstack/nova/blob/d25bc07d26212408211b64953af7ef6047ca3d9d/playbooks/legacy/nova-grenade-multinode/run.yaml#L47-L50
15:34:49 bauzas dansmith: your thoughts on it ? tl;dr: automatical uprade testing vs. asking for functional tests that would verify a RPC version minor bump
15:35:37 bauzas gibi: if we run two computes, then okay, we don't need them to be on separate nodes but the other services
15:35:43 bauzas ie. aio+compute
15:35:53 bauzas which is what grenade-multinode is doing AFAIR
15:35:58 bauzas so you're right
15:36:13 bauzas gibi: but then we need to test all the RPC calls in tempest
15:36:18 bauzas good luck with this
15:36:42 bauzas I just feel the simpliest is just to ask for functests
15:36:53 bauzas I wrote them yesterday night and it took me 20 mins
15:37:27 gibi bauzas: never said that we should not ask for a func test. I'm saying that we tend to forget about it as the current bug shows
15:37:28 dansmith yeah, so we could always pin the version to .0,
15:37:37 dansmith but coverage in tempest will be hard,
15:37:57 dansmith plus tempest needs to be graceful as some api calls will fail expectedly if the version doesn't support the new feature
15:38:00 bauzas technically, we need to set the pin to the previous release version
15:38:09 sean-k-mooney bauzas: same i want to stick on utc in my case
15:38:18 dansmith I'd prefer some test that ensures we've hit all the versions for each call in unit/func or something
15:38:40 bauzas dansmith: that's my thoughts
15:38:55 gibi dansmith: I thinked about that angle as that would be a good thing in my eys
15:38:57 bauzas checking it thru tempest is something I'd love, but I'm pragmativ
15:39:05 bauzas pragmatic
15:39:06 dansmith gibi: yeah
15:39:21 bauzas we honestly have the pattern to ask with my functest
15:39:28 bauzas it's just a simple request
15:39:39 bauzas and I guess (or I hope) none of the cores to miss this
15:39:52 gibi I have my doubt about my memory
15:40:01 sean-k-mooney testing rpc versions? trying to catch up on the converstation
15:40:03 gibi so I won't promise I will always remember
15:40:26 gibi sean-k-mooney: basically avoiding the bug https://bugs.launchpad.net/nova/+bug/1902925
15:40:26 openstack Launchpad bug 1902925 in OpenStack Compute (nova) "Upgrades to compute RPC API 5.12 are broken" [Critical,In progress] - Assigned to Sylvain Bauza (sylvain-bauza)
15:40:32 bauzas gibi: well, RPC and DB upgrades are possibly the hugest changes we could review, right?

Earlier   Later