DCGM-Exporter 4.6.0-4.8.3 (#711)

Filtered, squashed public import of the internal release snapshot.

Source tag: 4.6.0-4.8.3
Source commit: 6e9668062deb36d7315f69731ad22eccd4019401
Source branch: origin/rel_exp_4_8_3
GitHub base: d5e5f51

Excluded paths:
- .gitlab-ci.yml
- .gitlab/**
- .coderabbit.yaml
- .greptile/**
- internal/pkg/counters/llms_contract_test.go

Attribution audit: carried work from #489, #510, #654, #658, #660,
#663, #664, #665, #667, #671, #678, #687, #697, #699, and #704.
Issues #641 and #703 are acknowledged through their corresponding
implementers. Daphne's #655 remains directly attributed on GitHub main.

Signed-off-by: Nick Curry <ncurry@nvidia.com>
Co-authored-by: Yang Tong <25810794+tongyangyeyue@users.noreply.github.com>
Co-authored-by: Hairui Yang <hairuiyang@deeproute.ai>
Co-authored-by: Aditya Singh <aditya.chouhan@neevcloud.com>
Co-authored-by: Allen Zhou <loveinjavac@gmail.com>
Co-authored-by: Yufeng Huang <91405051+Raid57@users.noreply.github.com>
Co-authored-by: Rocker Zhang <zhang.rocker.liyuan@gmail.com>
Co-authored-by: Swati Gupta <swatig@nvidia.com>
Co-authored-by: Asutosh Panda <asutosh.pda@gmail.com>
Co-authored-by: Shravya Vorugallu <shravyavorugallu@gmail.com>
Co-authored-by: randomizedcoder <64496590+randomizedcoder@users.noreply.github.com>
Co-authored-by: skartikey <1942366+skartikey@users.noreply.github.com>
Co-authored-by: Dave Young <dave.andrew.young@gmail.com>
Co-authored-by: rvatkar <263779040+rvatkar@users.noreply.github.com>
Co-authored-by: dshaiknvidia <152336361+dshaiknvidia@users.noreply.github.com>

Author: 15 people

.codex/README.md

@@ -0,0 +1,19 @@
+# Codex Project Notes
+
+This directory contains Codex-facing guidance for DCGM Exporter. The
+repository-level `AGENTS.md` remains authoritative; these files are supporting
+notes for implementation, validation, and review sessions.
+
+## Safety Boundary
+
+Do not add Codex permission overrides, auto-approval config, automatic mutating
+hooks, or local environment secrets to this repository. Follow the active
+session sandbox and approval policy.
+
+## Files
+
+- `project-instructions.md` - compact implementation rules.
+- `validation.md` - explicit validation recipes.
+- `agent-profiles.md` - role guidance for focused review and verification.
+- `review-guardrails.md` - MR review checks.
+- `coderabbit-prevention.md` - recurring review issue prevention checklist.

.codex/agent-profiles.md

@@ -0,0 +1,38 @@
+# Codex Role Profiles
+
+Use these as role guidance, not as permission grants.
+
+## Debugger
+
+- Capture command, input, environment, exact error, and reproduction path.
+- Read the owning package before proposing a fix.
+- Verify with the narrow failing test first, then the broader relevant gate.
+
+## Verifier
+
+- Check that claimed files and behavior actually changed.
+- Run the relevant commands from `.codex/validation.md`.
+- Separate current-change failures from pre-existing or prerequisite failures.
+
+## Test Runner
+
+- Use package tests for pure logic.
+- Use GPU integration, Docker image, or Kubernetes e2e tests only when the
+  behavior needs that layer.
+- Do not weaken assertions to pass environmental tests.
+
+## Security Reviewer
+
+- Check secrets, auth material, token logging, web config handling, container
+  privilege, RBAC, and fail-open behavior.
+
+## Performance Reviewer
+
+- Measure before optimizing.
+- Check scrape latency, allocation-heavy rendering, lock contention, duplicate
+  parsing, and polling/reload behavior.
+
+## Documentation Specialist
+
+- Keep examples current and command outputs realistic.
+- Prefer direct links to repository files and official docs over memory.

.codex/coderabbit-prevention.md

@@ -0,0 +1,31 @@
+# CodeRabbit Prevention Checklist
+
+Apply before handing off a change.
+
+## Contracts
+
+- Do not invent metric labels, CLI flags, env vars, Helm values, or DCGM fields.
+- Keep README, Helm values/templates, raw YAML, and tests aligned with code.
+- Broad terminology changes need `rg` sweeps across Go, YAML, CSV, docs, tests,
+  metrics, and CI.
+
+## Go
+
+- Prefer explicit errors with context.
+- Avoid silent fallback for failed parsing, file I/O, metric registration,
+  serialization, or scrape rendering.
+- Do not add global mutable state without reset/restore tests.
+- Keep mock seams in wrapper packages and test helpers.
+
+## Tests
+
+- Assert the behavior named by the test.
+- Prefer typed field assertions over string/debug output matching.
+- Use package tests for logic and reserve GPU/e2e tests for hardware or
+  deployment contracts.
+
+## Docs
+
+- Comments and docs describe current behavior, not planned future cleanup.
+- Examples use current commands and real paths.
+- Security-sensitive examples avoid real-looking credentials.

.codex/project-instructions.md

@@ -0,0 +1,29 @@
+# DCGM Exporter Codex Instructions
+
+Use `AGENTS.md` and scoped `AGENTS.md` files as the source of truth.
+
+## Implementation Rules
+
+- Start feature work with the root `AGENTS.md` Feature Planning checklist and
+  Repository Surface Map.
+- Read the owning package, tests, Makefile target, and CI job before editing.
+- Keep docs and examples tied to real flags, env vars, Helm values, metrics,
+  and file paths from the repo.
+- Prefer small package-local helpers. Do not add broad agent-only abstractions.
+- Do not mutate generated artifacts unless the change requires regeneration and
+  the matching command is run.
+- Do not include unrelated worktree files, local settings, or coverage output in
+  commits.
+
+## Security
+
+- Never log or commit secrets, auth headers, kubeconfig material, web config
+  credentials, registry tokens, or secret-manager values.
+
+## Metrics
+
+- Metric names, labels, HELP text, and TYPE values are user-visible contracts.
+- Exporter-owned counters must stay aligned across code, CSV examples, docs,
+  and `llms.txt`.
+- Hardware-backed behavior should be tested with fakes where possible and with
+  GPU integration/e2e only when necessary and available.

.codex/review-guardrails.md

@@ -0,0 +1,32 @@
+# Review Guardrails
+
+Use these before MR handoff.
+
+## Planning
+
+- The change is mapped to an owning surface from root `AGENTS.md`.
+- User-visible contracts, docs, and prerequisite-dependent gates are accounted
+  for without duplicating root guidance.
+
+## Metrics And CLI
+
+- Metric CSV rows load through `internal/pkg/counters`.
+- Exporter-owned counters are documented in `llms.txt`.
+- CLI flags, env vars, defaults, Helm values, raw YAML, and README examples are
+  aligned.
+- Remote hostengine URI examples match `pkg/cmd/app.go`.
+
+## Tests
+
+- New behavior has the lowest useful test layer.
+- GPU/Docker/Kubernetes gates are run when prerequisites exist, or the skip is
+  documented.
+- Tests avoid tautological assertions and fixed sleeps when readiness can be
+  observed.
+
+## CI, Docker, And Scripts
+
+- Verification commands are not masked with `|| true` outside cleanup.
+- Shell scripts keep strict mode and quote path variables.
+- Version pins go through `hack/versions.env`, `make sync-versions`, and
+  `make validate-versions`.

.codex/validation.md

@@ -0,0 +1,54 @@
+# Validation Recipes
+
+Run commands from the repository root.
+
+## Common Docs And Go Gates
+
+```bash
+go test ./internal/pkg/counters
+make test-main
+make lint
+make check-fmt
+make validate
+```
+
+## Conditional Gates
+
+Generated mocks or go:generate inputs:
+
+```bash
+make generate
+git diff --exit-code
+```
+
+Version-derived files:
+
+```bash
+make sync-versions
+make validate-versions
+```
+
+Docker image behavior:
+
+```bash
+make local
+make test-images
+```
+
+GPU/DCGM integration behavior:
+
+```bash
+make test-integration
+```
+
+GPU Kubernetes behavior:
+
+```bash
+make e2e-local-check
+make e2e-local-up
+make e2e-local-deploy
+make test-e2e
+```
+
+If GPU, Docker, Kubernetes, DCGM, or network prerequisites are unavailable,
+record the skip and reason in the MR.

.cursor/README.md

@@ -0,0 +1,30 @@
+# Cursor Agent Configuration
+
+This directory contains project guidance for Cursor. It mirrors the intent of
+`AGENTS.md` without granting permissions or installing hooks.
+
+## Structure
+
+```text
+.cursor/
+  README.md
+  commands/
+  rules/
+  skills/
+```
+
+No `hooks.json` is provided. Formatting, validation, and security checks are
+explicit commands that agents or developers run intentionally.
+
+## Quick Reference
+
+| Area | Primary guidance |
+|---|---|
+| Go implementation | `rules/go-standards.mdc` |
+| Metrics and CSV contracts | `rules/metrics-contracts.mdc` |
+| Helm/Kubernetes | `rules/helm-kubernetes.mdc` |
+| Shell, CI, security | `rules/shell-ci-security.mdc` |
+| Tests | `rules/testing.mdc` |
+
+Use `commands/plan-feature.mdc` before broad feature work and
+`commands/validate-after-change.mdc` before handoff.

.cursor/commands/README.md

@@ -0,0 +1,10 @@
+# Cursor Commands
+
+These command files describe explicit workflows. They do not install hooks or
+change permissions.
+
+- `run-quality-checks`
+- `plan-feature`
+- `validate-after-change`
+- `code-review`
+- `show-dev-guide`

.cursor/commands/code-review.mdc

@@ -0,0 +1,13 @@
+---
+description: Review a DCGM Exporter change before MR handoff
+globs: ["**/*"]
+---
+
+# Code Review
+
+Check:
+
+- Metric, CLI, Helm, Docker, and package contracts stayed aligned.
+- Tests prove the behavior named by each test.
+- Shell and CI verification commands are not masked.
+- No secrets, local settings, coverage outputs, or unrelated files are staged.

.cursor/commands/plan-feature.mdc

@@ -0,0 +1,16 @@
+---
+description: Plan a DCGM Exporter feature before implementation
+globs: ["**/*"]
+---
+
+# Plan Feature
+
+Use the root `AGENTS.md` Feature Planning checklist and Repository Surface Map.
+
+Produce a concise plan that names:
+
+- the owning surface and scoped `AGENTS.md` files to read;
+- source-of-truth files, tests, and generated artifacts to verify;
+- user-visible contracts affected;
+- implementation approach and smallest useful abstraction;
+- validation commands and prerequisite-dependent gates.