[Question] Is multi-tenant implemented by Namespace + ResuorceQuota?

[zoux86]

Hi, !
I've been be interested in multi-tenant, so I was wondering if you can share how to implement multi-tenant an Resource sharing between multi-tenant .Thanks.

[k82cn]

The draft idea will use Queue + Namespace; we'll propose the design later :)

[zoux86]

@k82cn Thanks. By the way , is kube-batch support The management of GPU resources now?
#227

[k82cn]

Yes, we support GPU; but only the counting part; there's an issue in the pipeline for topology (#236).

[zoux86]

@k82cn Thanks~

[hex108]

@k82cn Is there minResource for namespace in your plan? minResource means that least resource that namespace could get. When the namespace does not need so much resource, it could share it to other namespaces by weight. When it needs these resource, it could preempt them.

[k82cn]

Is there minResource for namespace in your plan?

hm... that should be part of this; but the design is not finilized.

When the namespace does not need so much resource, it could share it to other namespaces by weight. When it needs these resource, it could preempt them.

There'll be several thresholds there, e.g. how many resource will be reserved, how many resource it can borrow, how many resource it would like to share :)

[hex108]

There'll be several thresholds there, e.g. how many resource will be reserved, how many resource it can borrow, how many resource it would like to share :)

Yes, are minResource, maxResource and fair share same as Hadoop YARN? That's what we want exactly.

[k82cn]

Yes, are minResource, maxResource and fair share same as Hadoop YARN? That's what we want exactly.

They're similar :)

[hex108]

Could we start work on this part now?

[k82cn]

I think we can start design doc & discuss right now. we're discussing the scope of this repo; if we can not add lifecycle management part, we have to implement algorithm here and lifecycle management in volcano :)

[k82cn]

/sig scheduling
/sig apps

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale