- Additional time needed (in calendar days, due end of day AoE): 3 days
- Reason this enhancement is critical for this milestone: This KEP fixes a kubelet/workload contract
breach where Guaranteed pods requesting hugepages are admitted by the Memory Manager but fail at runtime
because Burstable pods have consumed the hugepages without being tracked. This is a known production
issue (
https://github.com/kubernetes/kubernetes/issues/134395). The KEP has received review from
@johnbelamaric (PRR) and @ffromani (SIG Node), and all feedback has been addressed. We are awaiting
final SIG Node /lgtm to merge.
- Risks from adding code late (to k8s stability, testing, etc.): Minimal. This is an alpha feature
behind the MemoryManagerHugepagesVerification feature gate, scoped to a single sysfs read during pod
admission in the Memory Manager's Static policy. No API changes, no changes to default behavior.
- Risks from cutting enhancement (partial implementation, critical customer usecase, etc.): Delays the
fix for a known production issue affecting hugepage workloads. Users running mixed Guaranteed/Burstable
hugepage workloads will continue to experience silent runtime failures.
Thank you,
-Srikalyan