KEP-9734: Historical Borrowing based DRS Fair Sharing#9737
KEP-9734: Historical Borrowing based DRS Fair Sharing#9737mukund-wayve wants to merge 2 commits intokubernetes-sigs:mainfrom
Conversation
Change from historical usage to historical borrowing Cut down the length as much as possible
k8s-ci-robot
added
release-note-none
✅ Deploy Preview for kubernetes-sigs-kueue canceled.
|
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: mukund-wayve The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
|
Hi @mukund-wayve. Thanks for your PR. I'm waiting for a kubernetes-sigs member to verify that this patch is reasonable to test. If it is, they should reply with Tip We noticed you've done this a few times! Consider joining the org to skip this step and gain Once the patch is verified, the new status will be reflected by the I understand the commands that are listed here. DetailsInstructions 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-sigs/prow repository. |
k8s-ci-robot
added
the
needs-ok-to-test
k8s-ci-robot
added
size/L
k8s-ci-robot
removed
the
do-not-merge/work-in-progress
|
/ok-to-test |
k8s-ci-robot
added
ok-to-test
|
cc @mwielgus |
|
|
||
| ``` | ||
| borrowing(r) = max(0, usage(r) - quota(r)) | ||
| borrowing_sum(r) = (1 - α) × previous_borrowing_sum(r) + α × borrowing(r) |
There was a problem hiding this comment.
For α = 1 this is exactly the current behavior?
| preemption target selection, so that the two mechanisms are consistent. | ||
|
|
||
| * Preserve the preemption loop-free property that the current DRS-based fair | ||
| sharing guarantees. |
There was a problem hiding this comment.
For the current DRS-based fair sharing we have a proof that it is loop-free https://kueue.sigs.k8s.io/docs/concepts/fair_sharing/#proof-that-two-workloads-wont-preempt-each-other
Here it'll be a little more tricky as the DRS will change over time without any new admission. So the proof will not hold (I think).
Lets say we have a setup like this:
- cohort (1 quota):
- CQ1 (no quota), workload
wl1requiring 1 quota - CQ2 (no quota), workload
wl2requiring 1 quota
Lets say these are workloads that take a long time to finish.
I think in your proposal wl1 and wl2 will keep preempting each other, right? With wl1 admitted the new DRS will keep increasing until eventually it will be preempted by wl2 and so on.
Its not necessarily a bad thing if the preemptions will not be too frequent -- it might be even desirable, as this is more fair to do such "multiplexing" instead of allowing wl1 to run until completion. But this is a different property than for the current algorithm.
4 participants
What type of PR is this?
/kind kep
What this PR does / why we need it:
Creates a KEP for #9734
Which issue(s) this PR fixes:
Part of #9734
Special notes for your reviewer:
Does this PR introduce a user-facing change?