Skip to content

docs(gpu-operator): Correct single and mixed MIG strategy definitions - #529

Merged
mikemckiernan merged 4 commits into
mainfrom
devin/1793124937-mig-strategy-definitions
Oct 29, 2026
Merged

mikemckiernan merged 4 commits into
mainfrom
devin/1793124937-mig-strategy-definitions

Conversation

@mikemckiernan

@mikemckiernan mikemckiernan commented Oct 27, 2026 •

Copy link
Copy Markdown
Member

Summary

gpu-operator/gpu-operator-mig.rst defined the MIG strategies by MIG mode (single: MIG enabled on all GPUs; mixed: MIG not enabled on all GPUs). That contradicts the same page: the "Example: Mixed MIG Strategy" section fully partitions a GPU with all-balanced, and the custom-configuration steps say to use mixed for more than one instance profile.

This PR redefines each strategy by what the device plugin and GPU Feature Discovery present to Kubernetes, and states that the strategy does not enable MIG mode. It also links once to the device plugin README MIG_STRATEGY section for resource and label names.

Both custom MIG configuration examples now say to use mixed when the configuration uses more than one MIG profile or leaves MIG mode disabled on any GPU, and state what single supports. The installation example's custom-mig config (manifests/input/mig-cm-values.yaml) disables MIG on GPU 0, so a sentence after the YAML points that out.

The wording is backed by operand source (k8s-device-plugin bfc6655ff, gpu-operator ff1e0171b):

  • single — internal/rm/device_map.go: requireUniformMIGDevices = true; assertAllMigDevicesAreValid(true) errors when a MIG-enabled GPU has no MIG devices or when MIG device attributes differ ("more than one MIG device type present on node"); the build errors when full GPUs and MIG devices coexist. internal/lm/mig-strategy.go newMigStrategySingleLabeler applies the same rules and emits <model>-MIG-INVALID on failure. internal/rm/rm.go AddDefaultResourcesToConfig maps MIG devices to gpu.
  • mixed — buildGPUDeviceMap skips only MIG-enabled GPUs, so MIG-disabled GPUs remain nvidia.com/gpu; MIG devices register as mig-<profile>.
  • Strategy vs. MIG mode — gpu-operator applyMIGConfiguration sets MIG_STRATEGY only for GFD, the device plugin, and the MPS control daemon; MIG Manager does not read it.

Built locally with ./repo docs -p gpu-operator -p openshift. The page renders; the only build errors are the existing tab-set errors in dra-intro-install.rst and gpu-driver-upgrades.rst. The first commit also includes trailing-whitespace fixes from the pre-commit hook on three existing lines of the same file.

Link to Devin session: https://nvidia-cloud.devinenterprise.com/sessions/72ec3e92fc204828b501584860d0faad
Open in Devin Desktop: https://nvidia-cloud.devinenterprise.com/desktop/session/72ec3e92fc204828b501584860d0faad?variant=devin
Requested by: @mikemckiernan

Signed-off-by: Mike McKiernan <mmckiernan@nvidia.com>
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
@devin-ai-integration

Copy link
Copy Markdown
Contributor

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

  • Disable automatic comment, CI, and merge conflict monitoring

@github-actions

Copy link
Copy Markdown

Documentation preview

https://nvidia.github.io/cloud-native-docs/review/pr-529

…ixed strategy

Signed-off-by: Mike McKiernan <mmckiernan@nvidia.com>
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Comment thread gpu-operator/gpu-operator-mig.rst Resolved
Comment thread gpu-operator/gpu-operator-mig.rst Resolved
Comment thread gpu-operator/gpu-operator-mig.rst Resolved
mikemckiernan and others added 2 commits October 28, 2026 19:05
…ery GPU

Signed-off-by: Mike McKiernan <mmckiernan@nvidia.com>
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: Rajath Agasthya <rajathagasthya@gmail.com>
@mikemckiernan
mikemckiernan merged commit 227ccd5 into main Oct 29, 2026
1 of 2 checks passed
@mikemckiernan
mikemckiernan deleted the devin/1793124937-mig-strategy-definitions branch October 29, 2026 15:12
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants