Thursday, 2019-05-09

« Wednesday, 2019-05-08 Index Friday, 2019-05-10 »
*** ttsiouts has quit IRC00:12
*** mriedem has quit IRC00:15
*** artom has joined #openstack-placement00:30
*** Sundar has quit IRC00:45
*** cdent has quit IRC01:06
*** sean-k-mooney has quit IRC01:42
*** ttsiouts has joined #openstack-placement02:09
*** ttsiouts has quit IRC02:42
*** e0ne has joined #openstack-placement03:47
*** ttsiouts has joined #openstack-placement04:39
*** e0ne has quit IRC04:46
*** ttsiouts has quit IRC05:12
*** belmoreira has joined #openstack-placement06:06
*** ttsiouts has joined #openstack-placement07:09
*** helenafm has joined #openstack-placement07:41
*** ttsiouts has quit IRC07:42
*** ttsiouts has joined #openstack-placement07:46
*** belmoreira has quit IRC07:55
*** belmoreira has joined #openstack-placement07:58
*** e0ne has joined #openstack-placement08:14
*** e0ne has quit IRC08:17
*** ttsiouts has quit IRC09:15
*** tssurya has joined #openstack-placement09:18
*** belmoreira has quit IRC09:33
*** e0ne has joined #openstack-placement09:38
*** e0ne has quit IRC09:40
*** belmoreira has joined #openstack-placement09:48
*** e0ne has joined #openstack-placement09:56
*** tssurya has quit IRC09:57
*** e0ne has quit IRC09:59
*** helenafm has quit IRC10:10
*** mugsie has quit IRC10:25
*** ttsiouts has joined #openstack-placement10:31
*** mugsie has joined #openstack-placement10:35
*** mugsie has quit IRC10:35
*** mugsie has joined #openstack-placement10:36
*** ttsiouts has quit IRC10:36
*** mugsie has quit IRC10:38
*** mugsie has joined #openstack-placement10:39
*** tssurya has joined #openstack-placement10:42
*** belmoreira has quit IRC10:55
*** belmoreira has joined #openstack-placement10:56
*** belmoreira has quit IRC11:15
*** helenafm has joined #openstack-placement11:38
*** belmoreira has joined #openstack-placement11:48
*** belmoreira has quit IRC12:03
*** belmoreira has joined #openstack-placement12:49
*** mriedem has joined #openstack-placement12:51
*** artom has quit IRC13:04
*** ttsiouts has joined #openstack-placement13:24
*** altlogbot_0 has quit IRC13:43
*** altlogbot_2 has joined #openstack-placement13:45
*** ttsiouts has quit IRC14:08
*** ttsiouts has joined #openstack-placement14:23
*** ttsiouts has quit IRC14:28
*** ttsiouts has joined #openstack-placement14:59
*** ttsiouts has quit IRC15:04
*** helenafm has quit IRC15:43
*** belmoreira has quit IRC15:49
*** cdent has joined #openstack-placement15:52
* cdent waves from the enterprise15:53
efriedo/ cdent15:56
* edleafe sends cdent an enterprise-to-enterprise wave16:03
cdentwe need to spend 5 years writing a standard that no one will follow before we can send that wave16:04
aspiersX-D16:07
*** ttsiouts has joined #openstack-placement16:09
*** stephenfin has quit IRC16:30
*** tssurya has quit IRC16:38
cdentefried, edleafe, mriedem: the numbers on https://review.opendev.org/#/c/657423/ at http://logs.openstack.org/23/657423/9/check/placement-perfload/422d143/logs/placement-perf.txt are about what's expected. any thoughts?16:42
*** irclogbot_2 has quit IRC16:46
*** irclogbot_3 has joined #openstack-placement16:47
mriedemumm, is it, idk16:52
mriedem~33 seconds for GET /a_c with 10K providers and no limit16:53
mriedemif i'm reading that correctly, right?16:53
mriedemGET /allocation_candidates?resources=VCPU:1,DISK_GB:10,MEMORY_MB:256&member_of=14a5c8a3-5a99-4e8f-88be-00d85fcb1c17&required=HW_CPU_X86_AVX216:53
mriedemare all providers in that same aggregate and have that trait?16:53
*** cdent_ has joined #openstack-placement16:56
cdent_sorry mriedem I'm on a corp network so it likes to fail16:56
cdent_by expected I mean that it seems to have scaled something other than exponentially16:57
mriedemi guess the interesting thing to me would be if there is a linear graph of the time it takes to do the GET /a_c queries as the number of providers increases16:57
*** cdent has quit IRC16:57
*** cdent_ is now known as cdent16:57
mriedemok yeah if it's not exponential then great16:57
cdentit's somewhere between linear and geometric16:58
edleafecdent: that's 32sec for 2 runs, right?16:58
cdenteach run was 32 seconds, separately16:59
cdentmriedem: this would save some time: https://review.opendev.org/#/c/657716/16:59
cdentbut tetsuro accurately pointed out that we're stuck with the protections for a while16:59
edleafecdent: ack16:59
cdentIn the absolute, 32 seconds is too slow17:00
cdentthe profiling i've done is very noisy so will need closer analysis than I've been able to do this week17:00
*** stephenfin has joined #openstack-placement17:01
mriedemon https://review.opendev.org/#/c/657716/ i'd think we could do a blocker migration of sorts in train placement such that you can't upgrade until you've done that online data migration in stein17:03
mriedemseems like a simple "SELECT COUNT(id) FROM resource_providers WHERE root_provider_id IS NULL;" would do the trick yeah?17:04
mriedemif that's more than 0 you didn't do your homework17:05
mriedemwe already have a placement-status upgrade check for it, we could change that to an error and drop the code in train17:06
mriedemdrop the compat code i mean17:07
cdenton [t fYB9]17:07
purplerbot<mriedem> are all providers in that same aggregate and have that trait? [2019-05-09 16:53:47.351364] [n fYB9]17:07
cdentthere are three aggregates each rp is in one, two or three aggs, but every one is in the one queries by member_of17:08
cdentsame situation for traits17:08
cdentmriedem: can you leave a version of that "drop the compat code" comment on the review, responding to tetsuro please and thank you17:09
efriedcdent: gabbi response_json_paths boggle: I want to do GET /a_c calls knowing I'm going to get a bunch of results, but not knowing what order. Is there any way to describe that I expect a certain set of results in any order?17:22
cdentif they are dicts/objects without lists it ought to do dict style comparisons17:23
cdentif there are lists involved you can sort17:24
efriedcdent: It's list, but sorting would be a bear even in a language I was familiar with.17:25
cdentefried: If you've got a concrete example, I'll likely be able to provide a more concrete response17:25
cdentbut generally: doing spot checks of pieces of the response, rather than the entire response is often easier17:26
efriedI'm writing tests for can_split, so I want to be able to show that I get results like17:26
efriednuma0: 0; numa1: 517:26
efriednuma0: 1; numa1: 417:26
efriednuma0: 2; numa1: 317:26
efried...17:26
efriedAnd yeah, I can write spot checks, like to make sure that sorted(...numa0) = [0, 1, 2, 3, 4, 5] and sorted(...numa1) = $same17:27
efriedbut that doesn't really prove that the total is 5 for each entry in the list, ya know?17:27
cdenta) again seeing a thing you've tried (in gabbi) that has failed to work will make it easier to talk about this, b) sometimes (rarely ;) gabbi is not going to be the right tool for the job. It sounds like you're trying to write a functional test for db calls, not test api resonses17:29
efriedwell, I'm not going to be able to write a test that fails for reasons other than "we haven't implemented can_split yet", but I can throw you what I'm thinking...17:30
cdentIf I didn't have enterprisey brain dissolution17:30
cdentand I've got a meeting now, back in about 30 min17:30
*** artom has joined #openstack-placement17:41
*** mriedem has left #openstack-placement17:49
*** mriedem has joined #openstack-placement17:49
openstackgerritMatt Riedemann proposed openstack/placement master: Change "Missing Root Provider IDs" upgrade check to a failure  https://review.opendev.org/65816417:55
mriedemso, it's not obvious to me how to best get a connection to the db from an alembic migration script to run queries, or if that's even something one should do18:09
mriedemlooks like the alembic operations stuff is all about writing, but i just want to query18:10
mriedemmaybe that's just the alembic.context?18:11
cdentmriedem: doing a bit of a scan around: conn = op.get_bind()18:20
cdent    res = conn.execute("select secret, key, id from client")18:20
cdentfrom https://stackoverflow.com/questions/18726527/fetch-table-values-using-alembic-and-update-to-another-table18:21
mriedemyeah i got to the context.get_bind() but don't want to run straight sql18:21
mriedemmaybe a blocker migration is overboard - the online data migration is there, and the upgrade cehck18:23
mriedem*check18:23
cdentI _think_ once you've got a 'conn' you can do whatever you like with it18:24
*** ttsiouts has quit IRC18:25
mriedemyeah i think i see something in the sql docs18:25
*** jaypipes has quit IRC18:37
mriedemok got it18:39
*** ttsiouts has joined #openstack-placement18:40
mriedemnow how to write a test18:40
*** ttsiouts has quit IRC18:45
openstackgerritMatt Riedemann proposed openstack/placement master: WIP: Add blocker alembic migration for null root_provider_ids  https://review.opendev.org/65818218:49
openstackgerritMatt Riedemann proposed openstack/placement master: WIP: Remove null root provider protections  https://review.opendev.org/65771618:54
cdentthanks for stacking that mriedem.19:02
mriedemyw19:03
cdentIf I were conscious, I'd better understand why it's okay, and tetsuro's concerns are moot. I'm going to take everyone's word for it and look more closely later19:04
cdentfor now, lunch19:04
*** ttsiouts has joined #openstack-placement19:11
openstackgerritEric Fried proposed openstack/placement master: Add NUMANetworkFixture for gabbits  https://review.opendev.org/65746319:18
openstackgerritEric Fried proposed openstack/placement master: Gabbi test cases for can_split  https://review.opendev.org/65819219:18
efriedcdent: ^19:18
*** e0ne has joined #openstack-placement20:09
*** e0ne has quit IRC20:10
*** ttsiouts has quit IRC20:14
cdentthanks efried, looking at that and trying to tease out some ideas20:22
*** ttsiouts has joined #openstack-placement20:29
*** tssurya has joined #openstack-placement20:31
*** ttsiouts has quit IRC20:32
*** ttsiouts has joined #openstack-placement20:33
*** ttsiouts has quit IRC20:47
*** takashin has joined #openstack-placement20:52
*** cdent has quit IRC21:11
*** cdent has joined #openstack-placement21:26
*** ttsiouts has joined #openstack-placement21:29
*** ttsiouts has quit IRC21:34
openstackgerritChris Dent proposed openstack/placement master: Remove null root provider protections  https://review.opendev.org/65771621:45
openstackgerritChris Dent proposed openstack/placement master: Remove null root provider protections  https://review.opendev.org/65771621:53
openstackgerritChris Dent proposed openstack/placement master: DNM: See what happens with 10000 resource providers  https://review.opendev.org/65742321:53
openstackgerritChris Dent proposed openstack/placement master: WIP: Optionally run a wsgi profiler when asked  https://review.opendev.org/64326921:53
openstackgerritChris Dent proposed openstack/placement master: Remove null root provider protections  https://review.opendev.org/65771621:54
openstackgerritChris Dent proposed openstack/placement master: DNM: See what happens with 10000 resource providers  https://review.opendev.org/65742321:54
openstackgerritChris Dent proposed openstack/placement master: WIP: Optionally run a wsgi profiler when asked  https://review.opendev.org/64326921:54
cdentI am unsafe at any speed. Probably best take away my keys to gerrit.21:55
*** ttsiouts has joined #openstack-placement22:46
*** ttsiouts has quit IRC22:51
*** ttsiouts has joined #openstack-placement23:17
*** ttsiouts has quit IRC23:21
mriedemjesus think of the children23:51
* cdent starts works on a nested perfload23:58
« Wednesday, 2019-05-08 Index Friday, 2019-05-10 »

Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!