NVIDIA / NVIDIA/OpenShell

feat(tmachine): support runtime input overrides

Open
#3,464 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Rust
Stars
8.7k
Forks
1.3k
Avg merge
2d 11h
Merged PRs (30d)
253

Description

User Story

As an integration-test author, I want to override named tmachine inputs at runtime, so that CI can test candidate artifacts using values that are known only when the workflow runs.

Problem Statement

Tmachine inputs are currently fixed in its generated YAML configuration. The install and test commands have no runtime input override, even though release-specific values such as the candidate image SHA are available to the workflow only at runtime.

The Debian conformance lane introduced in #3461 therefore loads candidate supervisor and sandbox images under the fixed :tmachine tag and writes those references into the test user's gateway configuration. This safely prevents the packaged gateway from resolving its embedded dev or release-version tag, but it hides the candidate image identity and couples the playbook to a synthetic tag.

Using the package's embedded image tag is not a safe replacement. Release Dev could refresh :dev from the registry and test an older published image, while Release Tag runs before the versioned images are published.

Impact / Why This Matters

The current fixed-tag override keeps conformance deterministic, but CI cannot express the actual candidate image tag through tmachine. That makes artifact provenance less obvious and creates a risk that a future cleanup removes the override and accidentally pulls a previously published image during release validation.

Regenerating the Nix-backed tmachine configuration for every workflow value would move runtime CI data into build-time configuration and make local reproduction unnecessarily difficult.

Proposed Design

Add a repeatable runtime --input NAME=VALUE option to the tmachine install and test commands. Runtime values should override matching configured inputs before tmachine hashes the install layer and invokes Ansible.

The resulting workflow should allow release CI to retain the candidate images' immutable source-SHA tags and invoke tmachine with an override such as:

tmachine test ubuntu-docker-rootful deb conformance \
  --input openshell_runtime_image_tag="$SOURCE_SHA"

The Debian installer can then write the source-SHA image references into its minimal per-user gateway configuration. The packaged systemd unit and all other built-in gateway defaults remain unchanged.

Acceptance Criteria

  • tmachine install accepts repeatable --input NAME=VALUE overrides.
  • tmachine test accepts the same overrides for the installer phase.
  • Runtime values take precedence over matching installer inputs from config.yaml.
  • Malformed overrides and unknown input names fail with actionable errors.
  • Effective override values participate in the install-layer cache key.
  • Existing file-path inputs continue to be canonicalized when the referenced file exists, while scalar values do not need to name an existing file.
  • Unit tests cover parsing, precedence, validation, and cache invalidation.
  • Release Dev and Release Tag can pass the candidate source SHA to the Debian installer and remove the fixed :tmachine image references.
  • The Debian conformance lane continues to use a local immutable candidate image and does not pull dev or release-version runtime images from the registry.

Alternatives Considered

  • Keep the fixed :tmachine alias. This is deterministic but obscures which candidate image is under test and leaves the workflow coupled to a synthetic name.
  • Use the image tag embedded in the packaged gateway. This can pull an older dev image, and versioned release images are not published until after conformance succeeds.
  • Generate a new tmachine configuration in CI. This mixes runtime workflow values into Nix evaluation and makes equivalent local invocations harder to reproduce.

Agent Investigation

Tmachine currently stores installer and testsuite inputs as BTreeMap<String, PathBuf>. Existing paths are canonicalized before becoming Ansible extra variables; nonexistent paths are already passed through as literal values. Installer inputs are also part of the cached install-layer hash, so runtime overrides must be applied before both hashing and playbook execution.

Related to #3454 and follow-up to #3461.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the tmachine install and test command entry points and the config.yaml input handling described in the issue. Trace how installer inputs are canonicalized, included in the install-layer hash, and passed to Ansible. Add unit tests for parsing, precedence, validation, and cache invalidation; done means both commands accept repeatable overrides and the Debian conformance lane uses immutable candidate image references without fixed :tmachine tags.

Written by the indexing model from the issue text.

Assessment

Tech stack
rust
Domain
cli, release, testing-qa
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
68/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.