| *** artom has quit IRC | 00:38 | |
| *** openstackgerrit has joined #openstack-placement | 06:10 | |
| openstackgerrit | Michael Davies proposed openstack/os-traits master: Add support for ppc64le platforms https://review.opendev.org/680580 | 06:10 |
|---|---|---|
| *** cdent has joined #openstack-placement | 08:19 | |
| *** e0ne has joined #openstack-placement | 08:24 | |
| *** cdent has quit IRC | 08:44 | |
| openstackgerrit | Qitao proposed openstack/placement master: Fix misspell word https://review.opendev.org/680626 | 08:51 |
| *** zigo has quit IRC | 08:57 | |
| *** zigo has joined #openstack-placement | 09:05 | |
| *** cdent has joined #openstack-placement | 09:25 | |
| *** spatel has joined #openstack-placement | 09:31 | |
| *** spatel has quit IRC | 09:35 | |
| *** artom has joined #openstack-placement | 10:07 | |
| *** spatel has joined #openstack-placement | 11:01 | |
| *** spatel has quit IRC | 11:05 | |
| *** cdent has quit IRC | 11:32 | |
| *** cdent has joined #openstack-placement | 11:43 | |
| *** mriedem has joined #openstack-placement | 12:46 | |
| cdent | biab | 14:28 |
| *** cdent has quit IRC | 14:28 | |
| *** cdent has joined #openstack-placement | 15:22 | |
| *** johnthetubaguy has quit IRC | 15:37 | |
| *** mriedem is now known as mriedem_lunch | 16:11 | |
| sean-k-mooney | cdent: by the regarding placement feature for you there is 1 maybe too that im vagely aware of on the other hand there are several feature nova has not used yet so we might not get to a point where we need the new feature | 16:23 |
| cdent | sean-k-mooney: what's that? | 16:25 |
| sean-k-mooney | inplace rebuild without duplicate allocations | 16:28 |
| sean-k-mooney | if you rebuild normally you dont change resouce requiremetns | 16:28 |
| sean-k-mooney | however if you have the numa toplogy or cpu pinning set in the image you could | 16:28 |
| cdent | that sounds like a nova feature, not a placement feature? | 16:29 |
| cdent | that is: what allocations you write is up to nova | 16:29 |
| sean-k-mooney | yes but the allocation could chagne | 16:29 |
| cdent | again: the client is entirely in charge of what the allocations are for any consumer id | 16:29 |
| sean-k-mooney | so the feature we were toying with was could we ask placemetn is this existing allocation valid for this new request | 16:29 |
| cdent | as in "is this mine"? that sounds like a client side memory of associated consumer ids | 16:30 |
| cdent | emphasis on the client side | 16:30 |
| cdent | might be able to hack consumer types (if they merge) | 16:30 |
| sean-k-mooney | yes i think we might be able to do it client side however the image can also ask for things like traits | 16:31 |
| sean-k-mooney | soe changing the image might invalidate the host | 16:31 |
| cdent | I'm still not really understanding how placement would be able to help in this situation | 16:31 |
| cdent | I think I don't have enough context | 16:32 |
| cdent | but I would hope/guess that what you said about nova needing to catch up other places first is probalby right | 16:32 |
| sean-k-mooney | well basicaly i am saying i have already have allocation x. i need resouces y and assing on the current resouce provider can placmeent satify the request using the current allcoation or the remaing free resouces | 16:32 |
| cdent | oh! you want to be able to say "can I extend this allocation in the same resource provider" | 16:33 |
| cdent | which needs to be somewhat atomic | 16:33 |
| cdent | now I get you | 16:33 |
| sean-k-mooney | yes kind of | 16:33 |
| sean-k-mooney | so instead of doubleing the allcoation | 16:33 |
| sean-k-mooney | which is what we do on resize to same host | 16:34 |
| sean-k-mooney | can i in place atomicaly update theam | 16:34 |
| sean-k-mooney | for rebuild we cant move the instace unlike resize | 16:34 |
| sean-k-mooney | so it much more likely for the double allcoation to fail | 16:34 |
| * cdent nods | 16:34 | |
| sean-k-mooney | today i dont think we go back to placement on rebuild | 16:35 |
| sean-k-mooney | but that means we are ignoreing tratis on the new image if that is correct | 16:35 |
| sean-k-mooney | i know we rerun some fo the filter | 16:35 |
| sean-k-mooney | if we do go back to placemnt then we have the double allocation problem | 16:35 |
| cdent | i suppose you could ask for resources in GET /a_c that are the (positive) difference between what you have and what you want, and use the in_tree param | 16:36 |
| cdent | but you'll race on that | 16:36 |
| sean-k-mooney | yes possibley | 16:36 |
| cdent | the thing we want to absolutely avoid is placement having to manage any more state than it already does | 16:37 |
| cdent | s/state/types of state/ | 16:38 |
| *** e0ne has quit IRC | 16:39 | |
| sean-k-mooney | ya | 16:39 |
| sean-k-mooney | i get that | 16:39 |
| sean-k-mooney | how i would expect it to work if it was added is i say this is my exsting allocation and the desitred state | 16:40 |
| sean-k-mooney | and placmeent either rturns an error 409?/nothing or the update allcoation | 16:40 |
| sean-k-mooney | but i would like to see if we can do it client side first | 16:40 |
| sean-k-mooney | or if its neeeded at all | 16:40 |
| cdent | I kind of get the impression that there's going to need to be quite a bit of hard deciding about which features actually get attention in nova in U | 16:43 |
| cdent | and since there's quite the backlog... | 16:43 |
| *** efried is now known as efried_afk | 16:51 | |
| *** cdent has quit IRC | 16:58 | |
| *** mriedem_lunch is now known as mriedem | 17:38 | |
| *** e0ne has joined #openstack-placement | 17:58 | |
| *** e0ne has quit IRC | 18:15 | |
| *** efried_afk is now known as efried | 18:54 | |
| *** efried is now known as efried_afk | 20:19 | |
| *** efried_afk is now known as efried | 21:22 | |
| *** mriedem has quit IRC | 21:41 | |
| *** e0ne has joined #openstack-placement | 22:02 | |
| *** e0ne has quit IRC | 22:34 | |
| openstackgerrit | Merged openstack/placement master: Fix misspell word https://review.opendev.org/680626 | 23:28 |
| *** efried has quit IRC | 23:51 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!