NRI Plugin Refactoring

[roma-glushko]

What This PR Does

Refactoring NRI plugin to:

• move it's codebase to interna/nri package
• use urfave/cli as CLI framework
• use a unified, common logging setup
• separate containerd machinery from the actual injection logic
• break down injection logic into a clear multi-step pipeline that is easy to unit test

Since we have three components that behave like servers (Node Agent, Control Plane, NRI), I have promoted the health server logic into its own package reused in all three.

Why

It's stemmed from MEP0003. I'm keeping NRI in line with the recent restructuring and revisiting it's codebase to make sure it's clearly organized.

Checklist

• Commits are signed off (git commit -s)
• Tests pass (go test -v -race ./...)
• Linter passes (make lint-fix)
• New code has SPDX license headers
• Documentation updated (if applicable)
• CHANGELOG.md updated (if user-facing change)

---

_{Stack created with GitHub Stacks CLI • Give Feedback 💬}

[roma-glushko]

Too many flags, had to group them like this 😃

[greptile-apps]

Greptile Summary

The refactor separates NRI runtime integration from injection logic and consolidates health-probe handling.

• Moves the NRI implementation into internal/nri and introduces a staged injection pipeline.
• Replaces the NRI command-line parsing with urfave/cli.
• Reuses the shared health server across the node agent, control plane, and NRI plugin.
• Updates the image and Helm chart for the renamed nri-plugin binary and logging options.

_{Reviews (11): Last reviewed commit: "Updated changelog" | Re-trigger Greptile}

[ArangoGutierrez]

Thanks for this. Splitting adjust.go into one file per injection step makes the pipeline much easier to follow, and the step order, the three-arm GPU switch and every fail-open branch come through the move unchanged.

The behaviour moved with the code; what is left to tidy sits around the refactor rather than inside it.

• The shipped binary changes name (nvml-mock-nri to nri-plugin), all fourteen env-var bindings move to the MOKKA_NRI_ prefix, the --plugin-name default becomes mokka-nri-plugin, and pkg/nri/nvmlmock stops being importable, with nothing recorded under [Unreleased]. Anyone configuring the plugin through the environment falls back to defaults after this with no signal. (CHANGELOG.md:8)

  k8s-test-infra/CHANGELOG.md

  Lines 6 to 10 in e5b2a4c

  	and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
  	## [Unreleased]
  	### Fixed

• The DefaultConfig comment says it returns the identity the chart deploys with, but the chart passes --plugin-name=nvml-mock from values.yaml while the default here is mokka-nri-plugin. Rewording it keeps the next reader from trusting the default to match a real deployment. (internal/nri/config.go:27)

[ArangoGutierrez]

nodeAgent.logging and controlPlane.logging both declare level and format enums in values.schema.json, but nri.logging does not, so --set nri.logging.level=verbose templates cleanly and then the plugin exits 1 on ParseLevel. Copying that schema block under properties.nri keeps a typo failing at template time instead of on every node.

[roma-glushko]

This should help a5d8a51 ✅

[ArangoGutierrez]

Deleting these two lines leaves the agent and health suites green, so /readyz would answer 200 for a node whose simulator is not ready. TestReadyz_OneNotReady covered exactly this rule before the move; a case in agent_test.go driving Readiness with one simulator down would put it back.

[roma-glushko]

Covered the test gap in 419e547 ✅