Dominik Rabij comments

Results 21 comments of


                                            Dominik Rabij

Support flexStart capacity with non-queued provisioning

What was the use-case exactly? Does https://github.com/AI-Hypercomputer/xpk/pull/631 and https://github.com/AI-Hypercomputer/xpk/pull/686 solve the issue?

xpk should fail fast if a JobSet or PathwaysJob will result in an invalid value for the `jobset.sigs.k8s.io/coordinator` label

Created a task: http://b/456100304.

feat: Pathways use single resource group

I think it's set in https://github.com/AI-Hypercomputer/xpk/blob/f0626b93a0a7d3530e2daafeeb553e9a8168a0a3/src/xpk/core/kueue_manager.py#L324. @samos123 could you verify?

XPK should set CPU and memory unlimited when configuring Kueue

Is this still a problem with the current code?

JobSet syntax error on workload creation

I think this has been fixed already. Could you verify?

Support setting node auto-provisioning cpu and memory parameters

http://b/456420340

Utilize tier 1 networking for higher bandwidth

Thanks, I'll create a feature request: http://b/456067605.

Consider vendoring kjob and other binary dependencies

Related task: http://b/441035007.

refactor(multiple): eliminate usages of any type batch #4

@andrewseguin PTAL

refactor(multiple): eliminate usages of any type batch #4

@andrewseguin apparently narrowing the `MatTableDataSource` type is a breaking change. WDYT how should we proceed with this?