| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-05-18 | |||
| 18:54:42 | sean-k-mooney | so the AggregateInstanceExtraSpecsFilter requires you to modify all flavor to get teh behavior you want | |
| 18:57:36 | josephillips | and is requiered for the host aggregate have the extra filtes for ram allocation cpu allocation? | |
| 18:57:47 | josephillips | or just with the normal filters this functionality works too | |
| 18:57:48 | josephillips | ? | |
| 18:58:03 | sean-k-mooney | josephillips: we do not support setting it via the aggreate anymore | |
| 18:58:15 | sean-k-mooney | we removed that a few release ago | |
| 18:59:02 | sean-k-mooney | josephillips: http://lists.openstack.org/pipermail/openstack-discuss/2019-July/008118.html | |
| 18:59:04 | openstackgerrit | Andreas Jaeger proposed openstack/os-vif master: Switch to newer openstackdocstheme and reno versions https://review.opendev.org/728612 | |
| 18:59:40 | sean-k-mooney | josephillips: actully i ment to link http://lists.openstack.org/pipermail/openstack-dev/2018-January/126283.html | |
| 18:59:59 | josephillips | great | |
| 19:00:16 | sean-k-mooney | we deprecated the Aggregate[Core|Ram|Disk]Filter in ocata and remvoed them in train i think | |
| 19:00:30 | sean-k-mooney | josephillips: what release of openstack are you using | |
| 19:00:33 | josephillips | train | |
| 19:01:07 | sean-k-mooney | josephillips: ok in train you have two options. you can manage this via config file by setting the allcaotion ratios in the compute node config or via the api via placment | |
| 19:02:16 | sean-k-mooney | josephillips: this is the spec that cover why we made this change https://specs.openstack.org/openstack/nova-specs/specs/stein/implemented/initial-allocation-ratios.html let me see if i can get the docs link | |
| 19:02:59 | melwitt | note that the --aggregate and --amend options were added to osc-placement cli to help with this https://docs.openstack.org/osc-placement/latest/cli/index.html#cmdoption-openstack-resource-provider-inventory-set-aggregate | |
| 19:03:10 | sean-k-mooney | josephillips: https://docs.openstack.org/nova/latest/admin/configuration/schedulers.html#scheduling-considerations | |
| 19:03:43 | sean-k-mooney | melwitt: the were added in stien right | |
| 19:03:55 | openstackgerrit | Merged openstack/nova master: compute: Allow snapshots to be created from PAUSED volume backed instances https://review.opendev.org/728011 | |
| 19:03:59 | sean-k-mooney | so it should be avilable in train | |
| 19:04:24 | melwitt | looks like train https://docs.openstack.org/releasenotes/osc-placement/train.html#relnotes-1-7-0-stable-train | |
| 19:06:25 | openstackgerrit | Merged openstack/nova stable/ussuri: Update scheduler instance info at confirm resize https://review.opendev.org/728782 | |
| 19:06:32 | openstackgerrit | Merged openstack/nova master: compute: Remove snapshot quiesce tests for STOPPED and SUSPENDED instances https://review.opendev.org/728012 | |
| 19:17:06 | josephillips | sean-k-mooney: looks like work with the filterAggregateTypeAffinityFilter | |
| 19:17:34 | josephillips | so for live migrations this host aggregate respect this? | |
| 19:17:50 | sean-k-mooney | yes it will | |
| 19:18:15 | josephillips | so i can move only on nodes that meets the host aggregate specs | |
| 19:18:28 | sean-k-mooney | i cant recall if we support --force on live migrtion if we do that might bypass it | |
| 19:18:45 | josephillips | i will try on my lab | |
| 19:18:54 | sean-k-mooney | so yes and no | |
| 19:19:45 | sean-k-mooney | the limitation of the AggregateTypeAffinityFilter is that if flavor X is the only flavor that you allow for host aggreate X the Flavor y can land in host aggreate X | |
| 19:20:05 | josephillips | now im just have to check how ubuntu openstack is handling the ratio of memory or cpu if is on placement | |
| 19:20:05 | sean-k-mooney | but flavor X can still land on a host not in host aggreate X | |
| 19:20:33 | sean-k-mooney | josephillips: that is not disto specfic | |
| 19:21:09 | sean-k-mooney | although the charms may or may not configure the plament aggreats for you i have not really looked at how they work in a few years | |
| 19:21:15 | openstackgerrit | Elod Illes proposed openstack/nova stable/train: [train-only] Remove pycodestyle from test requirements https://review.opendev.org/729004 | |
| 19:22:05 | sean-k-mooney | josephillips: i would expect that you can either configure it in the compute node cofnig per host using the juju chram for nova or you have to do it manually via placement after the fact | |
| 19:22:40 | josephillips | sean-k-mooney: i lost in your explanation of the AggregateTypeAffinityFilter limitations | |
| 19:23:00 | josephillips | if i have the flavor m1 for the m1 servers they dont have any host aggregate | |
| 19:23:09 | josephillips | m2 flavors only for m2 servers with host aggregate | |
| 19:23:20 | josephillips | m2 flavors can land on m1 servers? | |
| 19:23:28 | sean-k-mooney | yes | |
| 19:23:39 | josephillips | under what circustances? | |
| 19:23:45 | sean-k-mooney | but m1 instance cannot land on m2 servers | |
| 19:24:00 | sean-k-mooney | josephillips: it depend on the weighers | |
| 19:24:29 | josephillips | so the correct way is be specific with nodes m1 and nodes m2 | |
| 19:24:37 | sean-k-mooney | if the weigher consider the m1 server to be a better choice because they have more free ram,disk ectra then it will select them | |
| 19:24:46 | sean-k-mooney | josephillips: yes | |
| 19:25:25 | sean-k-mooney | if you want to force it to only land on the specifc set you need to create a general host aggreate with all the m1 host and an m2 host aggreate with just the m2 hosts | |
| 19:25:46 | sean-k-mooney | then in the genral host aggreate list all the flavor excpet the m2 flavor and in the m2 host aggreate only list the m2 flavor | |
| 19:26:36 | sean-k-mooney | josephillips: tl;dr the AggregateTypeAffinityFilter prevent flavor that are not in the list form landing in the aggreate | |
| 19:27:02 | sean-k-mooney | josephillips: it does not force flavor to land on an aggreate | |
| 19:28:26 | sean-k-mooney | three is one thing you could do since your on train | |
| 19:28:34 | sean-k-mooney | to make it slightly simpler. | |
| 19:29:04 | sean-k-mooney | you could create only the aggreate for the special m2 hosts and use teh AggregateTypeAffinityFilter as you are doing now | |
| 19:29:48 | sean-k-mooney | but you caould also add a custom trait to all those hosts in plamcnet and then add trait:CUSTOM_MY_TRAIT=required | |
| 19:30:03 | sean-k-mooney | to the flavor | |
| 19:30:26 | sean-k-mooney | the custom required trait will force plamcent to only slect the hosts where you applied trait | |
| 19:30:43 | sean-k-mooney | and the filter will prevent instnace that dont have the correct flavor form landing there | |
| 19:30:56 | sean-k-mooney | if you put those two things togeter it will do what you want | |
| 19:31:25 | sean-k-mooney | your m2 instance will only land on the m2 host and the m1 instance cant land on the m2 hosts | |
| 19:32:03 | sean-k-mooney | josephillips: im about to leave for the night but one question before i do. your are not trying to use this for tenant isolation are you? | |
| 19:33:43 | sean-k-mooney | josephillips: this feature https://docs.openstack.org/nova/latest/reference/isolate-aggregates.html might be of use to you | |
| 19:35:07 | sean-k-mooney | josephillips: as might https://docs.openstack.org/nova/latest/admin/aggregates.html#tenant-isolation-with-placement if you are interested in the teant isolation case | |
| 19:35:42 | sean-k-mooney | the isolate aggreates feature is what was descibing as the hybrid approch | |
| 19:40:09 | sean-k-mooney | the isolate-aggregates featuer which is new in train might be the only thing you require to achive your goal. it was intended for more or less the same usecase but im not sure as you did not clearly state what you were trying to achive. | |
| 19:41:08 | josephillips | no | |
| 19:41:11 | josephillips | sean-k-mooney: im just looking use a second family of procesors and | |
| 19:41:23 | josephillips | dedicate some nodes for SaaS applicatios | |
| 19:42:06 | sean-k-mooney | https://docs.openstack.org/nova/latest/reference/isolate-aggregates.html shoudl allow you to do that | |
| 19:42:54 | sean-k-mooney | openstack --os-placement-api-version 1.6 trait create CUSTOM_SAS_HOSTS | |
| 19:43:22 | sean-k-mooney | then you can add CUSTOM_SAS_HOSTS to each host that you want to use | |
| 19:43:42 | sean-k-mooney | for sas applcaipton | |
| 19:43:54 | josephillips | and i add these meta to the flavor | |
| 19:43:55 | josephillips | ? | |
| 19:44:02 | josephillips | CUSTOM_SAS_HOSTS | |
| 19:44:26 | sean-k-mooney | in the flavor you set trait:CUSTOM_SAS_HOSTS=required | |
| 19:44:32 | sean-k-mooney | and you set the same in a host aggreate | |
| 19:44:59 | sean-k-mooney | openstack --os-compute-api-version 2.53 aggregate set --property trait:CUSTOM_SAS_HOSTS=required MY_SAS_HOST_AGGREATE | |
| 19:45:09 | josephillips | so im still have to use host aggregate | |
| 19:45:10 | josephillips | ? | |
| 19:45:17 | sean-k-mooney | yes | |
| 19:45:26 | josephillips | without any filter? | |
| 19:45:32 | sean-k-mooney | yes | |
| 19:45:43 | josephillips | mhmmm. i will try this | |
| 19:45:59 | sean-k-mooney | so you enable the featrure in the scudler config scheduler.enable_isolated_aggregate_filtering=true | |
| 19:46:44 | josephillips | just for ask and sorry for ask this but what is exactly the difference doing the filtering by nova or by placement | |
| 19:47:07 | sean-k-mooney | placemnt is faster and the way we want to do this going forward | |
| 19:47:28 | sean-k-mooney | plamcent returns a subset of host which is then processed by the filters in nova | |
| 19:47:39 | sean-k-mooney | so if you do it in placment the filter have much less work to do | |
| 19:47:54 | josephillips | ok | |
| 19:48:08 | josephillips | one las question what is the recommendation | |
| 19:48:19 | josephillips | if i will add more nodes with m2 processors | |
| 19:48:23 | josephillips | for general use | |
| 19:48:33 | josephillips | i create another trait:M2_hosts | |
| 19:48:36 | josephillips | and the flavor | |
| 19:48:59 | josephillips | for these hosts and thats it? is the correct way when you adding new servers with different specs | |
| 19:49:05 | sean-k-mooney | if you take the https://docs.openstack.org/nova/latest/reference/isolate-aggregates.html apporch you just dont add teh custom trait to the m2 hosts you want for general use | |
| 19:51:17 | sean-k-mooney | josephillips: what i would suggest it take a few minutes to read teh spec https://specs.openstack.org/openstack/nova-specs/specs/train/implemented/placement-req-filter-forbidden-aggregates.html and understand teh feature an see if your probelme is adress by it. i think it is and then feel free to come back and ask some quetions about it after you play awound with it a bit in a day or two | |
| 19:52:10 | josephillips | thank you so much | |