Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-11-04
15:39:41 efried okay, s/fast/reliable/
15:39:43 donnyd Just better at IO bound jobs
15:40:06 efried Yeah, I would rather my patch take 2h and succeed than 1.5h and fail.
15:40:17 donnyd I'm sure ironic and tripleo hate FN because my CPUs are old
15:40:41 donnyd And they are mostly CPU bound
15:40:54 donnyd Give it a swing with the label
15:41:05 donnyd And if it works, you have a bandaid
15:42:21 openstackgerrit Eric Fried proposed openstack/nova master: Add cyborg tempest job. https://review.opendev.org/670999
16:14:29 efried donnyd: I wasn't serious about the label. That would be pretty publicly greedy of us.
16:19:19 dansmith mriedem: question for you in here while I look at the test: https://review.opendev.org/#/c/635646/48/nova/conductor/tasks/cross_cell_migrate.py
16:19:50 dansmith holy hell, everything in the check/gate are running tests and only minutes old
16:19:58 dansmith we should send the whole community to china more often
16:24:15 donnyd Well not really efried
16:24:45 donnyd If a job has requirements to run and a specific label meets the requirements then I don't see the issue
16:25:02 donnyd But it's your call... just pointing out the options
16:26:44 mriedem dansmith: thanks, replied
16:27:18 mriedem dansmith: is my sarcasm detector picking this up correctly? https://review.opendev.org/#/c/635080/48/nova/tests/unit/compute/test_compute_mgr.py@10545
16:27:32 dansmith mriedem: yes.
16:28:02 mriedem heh, verbose commentage is how i keep from feeling lonely
16:34:55 dansmith mriedem: oops, my "...yes I see" was supposed to go into the fault_clone complaint nit
16:35:13 dansmith went back to note that and picked the wrong comment to edit
16:35:48 dansmith mriedem: one more question
16:40:30 openstackgerrit Merged openstack/nova master: Use long_rpc_timeout in conductor migrate_server RPC API call https://review.opendev.org/692550
16:41:03 mriedem replied
16:43:04 mriedem agree about the mapping == add a unit test wrinkle for that?
16:43:23 mriedem will do - and the variable name nits
16:45:58 dansmith cool
16:48:27 efried dansmith: I just put a bunch more words in the 'flavor groups' spec. Would you still be -1 if the answer to the backward compat question were "you can ignore flavor groups and keep doing extra specs forever"? IMO there's still too much complexity (most of which still isn't addressed) to make it worthwhile.
16:49:08 efried whoah, that timeout patch merged already?? Didn't we approve it like an hour ago??
16:49:13 dansmith efried: you mean would I still be effectively -2 I assume
16:49:21 efried yeah
16:50:22 efried pretty sure we're in agreement on it, I'm just wondering how aggressively I/we should say "despite feeding you paths to address all the holes, don't bother doing that, because this is going to die anyway"
16:50:22 dansmith efried: and without reading, you mean keeping extra_specs for the hard flavor case and only allowing the groups for compose-ability? I still think it's practically useless without some way to define which things can be composed, and agree it's far too much change to too many fundamental things to be worth it or viable
16:50:41 efried yeah, that's more or less what I said.
16:50:44 efried in the review
16:51:11 dansmith I think I'm still effectively -2 even with keeping extra_specs for those reasons, yeah
16:51:46 dansmith I think that it's legit to -2 something like this on "this isn't the direction we want to go" and if you and I both -2 it for that reason, I think that is a fairly tight case
16:51:54 dansmith it's not like we haven't considered it or reviewed it in detail
16:52:23 dansmith certainly don't want to just flip him the bird, but you wanted to be more upfront with people (as do I) so... that'd be pretty upfront
16:52:29 efried yeah, right now we're both -1.9 I think, but I don't want the author to take our -1s as a message that he should go try to fill it in.
16:52:38 dansmith yeah, fair point
16:52:54 dansmith I tried to call out my effectively -2ness for that reason, but maybe -2 would be better
16:53:11 efried so if you wouldn't mind reading my latest comments (when you get a chance) and if you agree with what I'm saying, go ahead and -2 and I'll follow suit.
16:53:21 dansmith sure
16:53:22 efried thx
16:53:46 mriedem i have run into "I don't want the author to take our -1s as a message that he should go try to fill it in." on a couple of specs for ussuri even though i've said a few times, "i don't think we need to do this, the existing alternatives or sufficient or this doesn't fit with the project IMO" but the specs continue to be updated
16:54:07 mriedem *are sufficient
16:54:29 mriedem i probably haven't -2ed because they are mostly trivial things to do, just plumbing,
16:54:40 mriedem not anywhere near the complexity of this flavor groups tihng
16:54:42 mriedem *thing
16:56:10 cdent Explicit words might be the way to go.
16:57:27 dansmith cdent: my words were explicit, but I hedged on the vote
16:57:44 dansmith because people sometimes accuse me of closing the door to further reviews with a -2 or even a -1
16:58:02 cdent well closing the door is what's wanted here, yeah?
16:58:22 dansmith well,
16:58:38 dansmith yes, but I want to do that in concert with other reviewers, like efried and I just did
16:59:20 dansmith so which do I need to kill? there is public, private, and shared
16:59:25 dansmith oops, wrong window
16:59:42 cdent all three /me is all for early kills
16:59:45 cdent !
17:02:48 mriedem https://review.opendev.org/#/c/682302/ and https://review.opendev.org/#/c/580336/ are the ones i'm struggling with fwiw
17:02:54 mriedem the latter has been coming up since berlin
17:03:17 mriedem and we have decent alternatives already available, though like i said the change itself would probably be pretty minimal
17:06:21 dansmith efried: done
17:06:28 efried thanks dansmith, following
17:09:50 dansmith huh, devstack defaults to configuring us for unversioned notifications apparently?
17:10:40 mriedem yeah, because downstream projects don't all use versioned
17:10:50 mriedem i seem to remember this coming up recently(ish) too when debugging something for watcher
17:11:08 mriedem i think b/c we (nova) changed our default
17:11:33 mriedem yup https://review.opendev.org/#/q/Ied9d50b07c368d5c2be658c744f340a8d1ee41e0
17:21:45 openstackgerrit Matt Riedemann proposed openstack/nova master: Follow up to I3e28c0163dc14dacf847c5a69730ba2e29650370 https://review.opendev.org/692856
17:23:54 mriedem heh that long_rpc_timeout change just merge conflicted most of the cross-cell-resize series, yay
17:25:12 dansmith red pill or blue pill...
17:26:02 mriedem oh well, needed to work on adding unit tests to https://review.opendev.org/#/c/637630/47 anyway so might as well rebsae
17:26:04 mriedem *rebase
17:27:49 mriedem the gate is empty due to the summit so it's actually the best time
17:31:45 mriedem that reminds me,
17:31:55 mriedem efried: https://review.opendev.org/#/q/topic:bp/support-move-ops-with-qos-ports-ussuri+status:open if you have some time - i can probably answer questions
17:32:15 mriedem the bottom one is mostly only big b/c of tests
17:33:53 efried mriedem: oh, yeah, lost on my list...
18:10:32 openstackgerrit Merged openstack/nova stable/stein: Avoid error 500 on shelve task_state race https://review.opendev.org/692630
18:10:37 openstackgerrit Merged openstack/nova master: Add finish_snapshot_based_resize_at_dest compute method https://review.opendev.org/635080
18:35:10 dansmith teehee
18:35:46 dansmith starting up 100 fake compute services on one machine... load of over 40 as they all slam the ever-loving crap out of conductor trying to create their db entries
18:36:15 openstackgerrit Merged openstack/nova master: Add FinishResizeAtDestTask https://review.opendev.org/635646
18:49:46 mriedem dansmith: how big is the host?
18:49:57 dansmith how big?
18:50:07 dansmith about yay big
18:50:09 mriedem i do the fake compute thing in a 8vcpu/8gb ram vm sometimes and can't really go over 30 computes
18:50:10 dansmith about 19" wide
18:50:30 mriedem cpu/ram
18:50:44 dansmith oh, 32G memory, 8 cores if that's what you mean
18:50:54 mriedem yeah
18:51:10 mriedem did you adjust API_WORKERS or let it do ncpu/2?
18:51:31 dansmith just defaults
18:52:25 mriedem ah. fake computes huh. so all that time trying to figure out which network to delete and you found out you can't create a fake driver vm with networking anyway right?
18:52:48 mriedem i always forget that when i do a devstack with the fake driver
18:53:36 mriedem i had written up some of that here at one point https://docs.openstack.org/devstack/latest/guides/nova.html#fake-virt-driver
18:53:37 dansmith one real compute and then a bunch of extra fakes on the same machine
18:53:41 mriedem ah ok

Earlier   Later