fix(nemo-deployments): ship openshell SDK in nmp-api and clarify dropped-executor error

[maxdubrinsky]

Summary

Fixes AIRCORE-978. Two defects in the published nmp-api image: OpenShell-sandboxed deployments don't work, and a dropped executor is indistinguishable from a typo.

Defect 1 - openshell SDK missing from the image

The image installs the functional-services dep group, which carried nemo-deployments-plugin[docker,k8s] but not the openshell extra. So OpenShellDeploymentBackend couldn't initialise, the plugin skipped the openshell-local executor at startup, and any deployment naming it failed with Executor 'openshell-local' is not registered.

Fix: add nemo-deployments-plugin[openshell] to functional-services.

It goes in functional-services (image-only) rather than the bootstrap groups (enabled-plugins, dev) on purpose: the openshell wheel needs glibc >= 2.39 and has no sdist, and uv markers can't gate on glibc, so a hard dependency would break bare uv sync on older dev hosts. The image builds on a compatible base, so that's the right home for this required-in-image capability.

Defect 2 - dropped executor looks like a typo

A configured executor whose backend was skipped raised the same is not registered error as a misspelled name.

Fix: ExecutorRegistry now remembers skipped executors and resolve() raises a distinct message: Executor 'X' is configured but its backend 'Y' is unavailable: <reason>.

Testing

• pytest plugins/nemo-deployments/tests/unit/test_registry.py -> 12 passed. Added test_unavailable_executor_reports_distinct_error; tightened test_missing_executor_raises to assert the is not registered wording.
• uv lock --check in sync; openshell/grpcio/protobuf resolve for all three locked platforms (darwin-arm64, linux-x86_64, linux-aarch64).
• ruff clean; ty shows only pre-existing diagnostics.

Fixes AIRCORE-978
https://linear.app/nvidia/issue/AIRCORE-978

Summary by CodeRabbit

• Bug Fixes

  • Improved executor errors when a deployment backend is unavailable, clearly identifying the backend and missing dependency.
  • Clarified errors for executor names that are not registered.

• New Features

  • Added support for the OpenShell deployment SDK in the functional services environment, enabling deployments that require this integration.

[github-actions]

Suite	Lines Covered	Line Rate	Branch Rate
Unit Tests	30782/39277	78.4%	62.8%
Integration Tests	18075/37229	48.5%	21.1%

[coderabbitai]

📝 Walkthrough

Walkthrough

The executor registry now records executors skipped because of missing backend dependencies. Resolution errors identify the configured backend and failure reason. The functional-services dependency group includes the OpenShell extra, and unit tests cover both error paths.

Changes

Executor availability handling

Layer / File(s)	Summary
Track unavailable backends plugins/nemo-deployments/src/nemo_deployments_plugin/backends/registry.py, pyproject.toml	The registry records unavailable backends and their failure reasons. The functional-services group includes the OpenShell extra.
Report resolution failures plugins/nemo-deployments/tests/unit/test_registry.py, plugins/nemo-deployments/src/nemo_deployments_plugin/backends/registry.py	Resolution distinguishes unregistered executors from unavailable backends. Tests verify both messages and their diagnostic details.

Possibly related PRs

• NVIDIA-NeMo/nemo-platform#663: Introduced the executor registry used by this change.
• NVIDIA-NeMo/nemo-platform#1082: Also updates registry errors for unavailable and unregistered executors.

Suggested reviewers: benmccown, mckornfield

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes both main changes: shipping the OpenShell SDK and clarifying the dropped-executor error.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch mdubrinsky/aircore-978-published-nmp-api-image-ships-without-the-openshell-sdk

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@plugins/nemo-deployments/tests/unit/test_registry.py`:
- Around line 122-145: Update test_unavailable_executor_reports_distinct_error
to assert that the error message includes the configured executor name
"sandbox-local" together with the backend name "sandbox", preserving the
existing assertions for the distinct unavailable-backend reason.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 1670f98c-1526-4586-95f5-fb0598774bb3

📥 Commits

Reviewing files that changed from the base of the PR and between 795bba2 and 477604f.

⛔ Files ignored due to path filters (1)

• uv.lock is excluded by !**/*.lock

📒 Files selected for processing (3)

• plugins/nemo-deployments/src/nemo_deployments_plugin/backends/registry.py
• plugins/nemo-deployments/tests/unit/test_registry.py
• pyproject.toml