| *** tetsuro has quit IRC | 00:23 | |
| *** spatel has joined #openstack-placement | 00:41 | |
| *** spatel has quit IRC | 01:23 | |
| *** openstackgerrit has quit IRC | 02:37 | |
| *** e0ne has joined #openstack-placement | 08:11 | |
| *** cdent has joined #openstack-placement | 08:18 | |
| *** e0ne has quit IRC | 08:20 | |
| *** e0ne has joined #openstack-placement | 08:34 | |
| *** openstackgerrit has joined #openstack-placement | 08:36 | |
| openstackgerrit | Merged openstack/osc-placement master: Cap sphinx for py2 to match global requirements https://review.opendev.org/679137 | 08:36 |
|---|---|---|
| openstackgerrit | Jens Harbott (frickler) proposed openstack/placement master: Fix links to migration scripts https://review.opendev.org/679205 | 08:45 |
| *** e0ne has quit IRC | 08:59 | |
| *** e0ne has joined #openstack-placement | 09:27 | |
| *** e0ne has quit IRC | 09:27 | |
| cdent | gibi: those rally scale tests are basically designed in "let's push it until it breaks" kind of way: break something fix it, break the next thing fix it, repeat | 09:41 |
| cdent | allocation generation collisions is just the latest in a series of breaks and fixes | 09:41 |
| cdent | there will likely be more | 09:41 |
| cdent | in this case I think it points out a legit flaw in placement: concurrent writes to the same resource provider are harder to manage than they maybe should be | 09:42 |
| cdent | it's not surprising because it wasn't designed to expect that | 09:42 |
| openstackgerrit | Chris Dent proposed openstack/placement master: Add allocation_conflict_retry_count conf setting https://review.opendev.org/679108 | 09:48 |
| openstackgerrit | Chris Dent proposed openstack/placement master: Introduce jitter during allocations replace_all retry https://review.opendev.org/679109 | 09:48 |
| *** e0ne has joined #openstack-placement | 10:21 | |
| *** e0ne has quit IRC | 10:27 | |
| gibi | cdent: thanks for the background info. I think I more like the config based solution than the random sleep. | 10:32 |
| cdent | gibi: yeah, me too. I stuck the jitter in there too out of concern that maybe even if the retry is really high, things would synchronize in a way that made it continue to always conflict | 10:34 |
| cdent | in which case a lot of time is wasted to accomplish nothing | 10:34 |
| cdent | unfortunately that's hard to test... | 10:35 |
| gibi | cdent: I think the eventual client side retry injects some jitter anyhow | 10:35 |
| cdent | right, but ideally we'd avoid a client side retry after, say, 100 retries server side | 10:37 |
| cdent | and its 100 retries somehow getting synchronized to fail is what I was worried about avoiding with a server-side jitter | 10:37 |
| cdent | but mostly I think I was watching some video about distributed systems and some slide was "never forget jitter" | 10:38 |
| cdent | so it got in my head | 10:38 |
| cdent | if it was easier to test... | 10:38 |
| gibi | cdent: OK, I'm close to be convinced that both change is needed for a more optimal solution.. | 10:39 |
| cdent | I'm trying to get the scale tests run internally again, using those patches | 10:40 |
| cdent | so we can confirm their efficacy | 10:40 |
| cdent | but first i have to port them in to nova :( | 10:40 |
| gibi | I've just ported a rocky feature back to pike internally in nova, so I share your pain | 10:42 |
| *** e0ne has joined #openstack-placement | 11:00 | |
| *** e0ne has quit IRC | 11:20 | |
| *** e0ne has joined #openstack-placement | 11:23 | |
| *** e0ne has quit IRC | 11:32 | |
| cdent | argh. nova's copy of placement is so icky | 11:38 |
| *** e0ne has joined #openstack-placement | 11:40 | |
| * cdent locates lunch | 11:47 | |
| *** cdent has quit IRC | 11:47 | |
| *** cdent has joined #openstack-placement | 12:00 | |
| *** e0ne has quit IRC | 12:11 | |
| cdent | sean-k-mooney: you might have input on this trait addition discussion: https://storyboard.openstack.org/#!/story/2006456 | 12:20 |
| sean-k-mooney | thanks ill take a look. | 12:21 |
| sean-k-mooney | speed slelect | 12:21 |
| sean-k-mooney | well hw/cpu/x86/intel.py would be the right place to put it | 12:21 |
| sean-k-mooney | its a hardware feature rather then a hypervior feature | 12:22 |
| sean-k-mooney | i see efried feels differently | 12:22 |
| sean-k-mooney | am im being ping direclty on 2 other threads right now ill loop back to this when i have enofugh time to read it properly | 12:23 |
| sean-k-mooney | cdent: thanks for the heads up | 12:23 |
| cdent | thanks for looking, there's no rush as far as I can tell | 12:24 |
| sean-k-mooney | this is for U correct | 12:27 |
| sean-k-mooney | we dont have speed select support in nova today | 12:28 |
| cdent | sean-k-mooney: I would guess that any use on the nova side is for U | 12:28 |
| sean-k-mooney | i mean it can be merged soone but i dont think we will report it | 12:28 |
| cdent | os-traits itself doesn't really care one way or another about cycles | 12:28 |
| sean-k-mooney | ya makes sense | 12:28 |
| *** cdent has quit IRC | 12:35 | |
| *** cdent has joined #openstack-placement | 12:45 | |
| *** mriedem has joined #openstack-placement | 13:07 | |
| efried | hw/cpu/x86/intel is arguably the right place for the feature that says "this host has speed-selectable CPUs". My contention is that we don't need that trait to exist. | 13:19 |
| *** dklyle has quit IRC | 13:35 | |
| *** dklyle has joined #openstack-placement | 13:35 | |
| *** e0ne has joined #openstack-placement | 13:40 | |
| *** e0ne has quit IRC | 13:43 | |
| *** efried is now known as efried_afk | 13:47 | |
| mriedem | melwitt is out today and tomorrow but the review changes to the osc-placement resource provider inventory set --amend patch are pretty small if someone wants to take a crack at them https://review.opendev.org/#/c/675520/6 | 14:08 |
| mriedem | mostly just test tweaks | 14:08 |
| * cdent is in the midst of trying to finish consumer types | 14:09 | |
| *** e0ne has joined #openstack-placement | 14:22 | |
| *** efried_afk is now known as efried | 15:13 | |
| *** e0ne has quit IRC | 15:14 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: Add consumer_types table and consumer_type_id column to consumers https://review.opendev.org/669170 | 17:00 |
| openstackgerrit | Chris Dent proposed openstack/placement master: [WIP, DNR] Microversion 1.37: Adds support for consumer_types https://review.opendev.org/672476 | 17:00 |
| openstackgerrit | Chris Dent proposed openstack/placement master: WIP: Microversion 1.37: Adds support for consumer_types https://review.opendev.org/672476 | 17:00 |
| *** e0ne has joined #openstack-placement | 17:02 | |
| *** cdent has quit IRC | 17:03 | |
| *** e0ne has quit IRC | 17:04 | |
| *** e0ne has joined #openstack-placement | 17:04 | |
| *** e0ne has quit IRC | 17:04 | |
| *** e0ne has joined #openstack-placement | 17:35 | |
| *** e0ne has quit IRC | 17:35 | |
| *** N3l1x has joined #openstack-placement | 18:09 | |
| *** e0ne has joined #openstack-placement | 18:29 | |
| *** e0ne has quit IRC | 18:32 | |
| openstackgerrit | Merged openstack/placement master: Fix links to migration scripts https://review.opendev.org/679205 | 18:45 |
| *** e0ne has joined #openstack-placement | 18:55 | |
| *** e0ne has quit IRC | 19:01 | |
| *** e0ne has joined #openstack-placement | 19:03 | |
| *** e0ne has quit IRC | 19:04 | |
| *** e0ne has joined #openstack-placement | 19:05 | |
| *** e0ne has quit IRC | 19:05 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!