[runzhen]

What this PR does / why we need it:

• when MIG is enabled, start the DRA after MIG service
• add the unit test case

Which issue(s) this PR fixes:

Fixes #

[github-actions]

Windows Unit Test Results

  3 files   18 suites   1m 10s ⏱️
563 tests 563 ✅ 0 💤 0 ❌
585 runs  585 ✅ 0 💤 0 ❌

Results for commit 1c8918c.

♻️ This comment has been updated with latest results.

[Copilot]

🟡 Changes recommended

The condition checks the wrong layout variable, leaving layout-only MIG nodes without the intended DRA startup ordering.

1 open issue

• Handle layout-only MIG configurations in startup dependency logic · New

What changed in this PR

This PR updates Linux GPU provisioning to start the NVIDIA DRA driver after MIG partitioning when MIG is configured.

Changes:

• Adds conditional systemd startup ordering for the DRA driver.
• Adds an Ubuntu 24.04 E2E scenario covering MIG Mixed mode, DRA service startup, and workload scheduling.

File	Description
parts/​linux/​cloud-init/​artifacts/​cse_config_gpu.sh	Adds MIG-dependent ordering to the DRA service override.
e2e/​scenario/​scenario_gpu_managed_experience.go	Adds a managed DRA scenario using an A100 GPU with MIG Mixed mode.

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

The scoped ordering change has no identified blocking defects; deterministic ShellSpec coverage remains a non-blocking improvement.

No open issues 🎉

1 previously missed issue

In code that hasn't changed since last review

Add tests for DRA drop-in ordering across GPU profile inputs

parts/​linux/​cloud-init/​artifacts/​cse_config_gpu.sh:451

🟡 Medium Risk — 🧪 Test Coverage: Please extend the existing startNvidiaManagedExpServices ShellSpec cases in spec/parts/linux/cloud-init/artifacts/cse_config_gpu_spec.sh to assert the generated DRA drop-in's ordering. None currently checks this directive, and the new E2E scenario sets only GPUInstanceProfile and checks eventual health, so it cannot reliably catch removal of the ordering or a broken layout-only condition. Capture the DRA drop-in and assert After=mig-partition.service kubelet.service for profile-only and layout-only inputs, and After=kubelet.service when both inputs are empty.

1 resolved since last review

• Handle layout-only MIG configurations in startup dependency logic

_{Review effort: Balanced}

[github-actions]

Package	Line Rate	Health
test	0%	❌
windows	27%	❌
staging_cse_windows Package 1	65%	➖
debug	0%	❌
provisioningscripts	2%	❌
parts_windows Package 1	73%	➖
Summary	38% (1652 / 6834)	❌

[github-actions]

Package	Line Rate	Health
github.com/Azure/agentbaker/aks-node-controller	72%	➖
github.com/Azure/agentbaker/aks-node-controller/common	100%	✔
github.com/Azure/agentbaker/aks-node-controller/helpers	68%	➖
github.com/Azure/agentbaker/aks-node-controller/parser	88%	✔
github.com/Azure/agentbaker/aks-node-controller/pkg/gpu	100%	✔
github.com/Azure/agentbaker/aks-node-controller/pkg/nodeconfigutils	89%	✔
github.com/Azure/agentbaker/aks-node-controller/utils	0%	❌
github.com/Azure/agentbaker/apiserver	24%	❌
github.com/Azure/agentbaker/cmd	0%	❌
github.com/Azure/agentbaker/cmd/starter	0%	❌
github.com/Azure/agentbaker/fuzz/api	0%	❌
github.com/Azure/agentbaker/hotfix/render-nodecustomdata	0%	❌
github.com/Azure/agentbaker/pkg/agent	78%	➖
github.com/Azure/agentbaker/pkg/agent/datamodel	77%	➖
github.com/Azure/agentbaker/pkg/agent/toggles	0%	❌
github.com/Azure/agentbaker/pkg/vhdbuilder/datamodel	87%	✔
Summary	75% (5873 / 7861)	➖

[github-actions]

Package	Line Rate	Health
shellspec spec	28%	❌
Summary	28% (4557 / 15973)	❌

[Copilot]

🟡 Changes recommended

The provisioning change still needs the required ShellSpec regression coverage for profile-only, layout-only, and non-MIG service ordering.

1 open issue

• Add ShellSpec coverage for GPU service ordering · New

_{Review effort: Balanced}