mediumScheduling

MIG configuration problem

MIG instances don't appear as allocatable, or the wrong profile is exposed.

Scope
NVIDIA MIG partitioning on A100/H100-class parts.
Check recent changes
  • ·MIG strategy/profile change
  • ·Operator upgrade
Diagnostics

What's allocatable?

bash
kubectl get node <node> -o json | jq '.status.allocatable' | grep mig

Expect: Entries like nvidia.com/mig-1g.10gb matching the intended profile.

Likely causes & fixes
  • high likelihoodMIG mode/profile not applied

    Fix · Set nvidia.com/mig.config on the node label; wait for reconfigure.

  • medium likelihoodStrategy mismatch (single vs. mixed)

    Fix · Align the operator MIG strategy with your profiles.

  • low likelihoodPods request a whole GPU, not a MIG slice

    Fix · Request the specific MIG resource name.

Validation
  • ✓Intended MIG instances appear allocatable and schedule
Escalation

Reconfigure repeatedly fails → operator/driver logs, vendor support.

Data to collect first
  • ·ClusterPolicy MIG settings
  • ·Node MIG labels
  • ·Allocatable MIG resources