Add GPU Numbers Predicates #1692
Conversation
|
Welcome @peiniliu! |
volcano-sh-bot
added
the
size/L
|
Hi @peiniliu , thanks for your contribution. Please fix the DCO check. |
Thor-wl
requested review from
Thor-wl,
jasonliu747 and
william-wang
and removed request for
hudson741 and
hzxuzhonghu
|
@jasonliu747 Really hope this feature can be involved into v1.4. I have not got a gpu env. do you have gpu environment for verification? |
|
@william-wang let me double check and get back to you! |
volcano-sh-bot
added
size/XXL
Hi Jason, thx! I've fixed the DCO. Pls let me know if further things are needed. Best, Peini |
|
@peiniliu Rebase and submit the pr again. |
volcano-sh-bot
added
size/L
|
@jasonliu747 @william-wang |
peiniliu
force-pushed
the
dev
branch
3 times, most recently
from
edf8719 to
6db3361
Compare
|
This PR will fix #1686 ? |
|
this PR supports users to define GPU number predicates via container level resource requirements using the 'volcano.sh/gpu-numbers'. For provided issue, you may look at the queue and preempt or reclaim actions. |
|
|
||
| The main architecture is similar as the previous, but the gpu-index results of each pod will be a list of gpu cards index. | ||
|
|
||
|  |
There was a problem hiding this comment.
Kebe-API-Server => Kube-API-Server
|
I fixed those comments. |
| @@ -46,6 +46,8 @@ Same as above, after installed, update the scheduler configuration in `volcano-s | |||
|
|
|||
| Please refer to [volcano device plugin](https://github.com/volcano-sh/devices/blob/master/README.md#quick-start) | |||
|
|
|||
| * By default volcano device plugin supports shared GPUs, users do not need to config volcano device plugin. Default setting is the same as setting --gpu-strategy=number. For more information [volcano device plugin configuration](https://github.com/volcano-sh/devices/blob/dev/doc/config.md) | |||
|
|
|||
There was a problem hiding this comment.
Generally LGTM.
/cc @Thor-wl @william-wang Please take another look
|
@shinytang6: GitHub didn't allow me to request PR reviews from the following users: another, look, Please, take. Note that only volcano-sh members and repo collaborators can review this PR, and authors cannot review their own PRs. In response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
as explained, currently device plugin does not fit multiple containers. as well as the initContainer is designed for some start scripts which may not request GPUs. Well, I think the important part is the device plugin, once it supports, the GPU usage for initContainer can be added. |
IC, that's OK for me. |
|
/lgtm |
|
I removed some unrelated info in the doc after the presentation. |
|
/lgtm |
Signed-off-by: peiniliu <[email protected]>
volcano-sh-bot
added
lgtm
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: shinytang6, Thor-wl The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
|
|
||
| id := predicateGPU(pod, nodeInfo) | ||
| if id < 0 { | ||
| ids := predicateGPUbyMemory(pod, nodeInfo) |
There was a problem hiding this comment.
share gpu pod only need to one gpu id, but predicateGPUbyMemory reture all gpu id which are suitable for the pod.
Support specify GPU numbers for pod resource requests issue#1440
Currently, Volcano only supports specified GPU share memory. Specified GPU number is not supported. This pr supports defining GPU numbers for pod resource requests. You can check the design doc https://github.com/peiniliu/volcano/blob/dev/docs/user-guide/how_to_use_gpu_number.md for more details.