Repository navigation
Conversation
Prepare an Ubuntu-based image for headless Newton training without the Isaac Sim runtime. Document local build and validation while keeping the image independent of existing workflows.
Build docker/Dockerfile.newton from the public ubuntu:24.04 base and run the documented cartpole training probe plus a non-root runtime check, mirroring the folded verification steps used for the base and cuRobo images. Also pin the builder venv to the system Python 3.12 so it cannot drift from the runtime stage's libpython3.12, and fix the copyright header year range.
Force run_docker_tests=false in build.yaml's changes job and run_install_tests=false in install-ci.yml so all gated jobs skip via their existing if-gates, and run build-newton unconditionally on PRs. Must be reverted before final review.
…tless-container-codex
The bare train/play console scripts were removed from the isaaclab package by the central pyproject consolidation (isaac-sim#6009); the only remaining entry point is isaaclab. CI builds the PR merge commit, so the kitless image no longer contains .venv/bin/train and 'docker run ... train' fails with 'executable file not found'. Switch the docs verification command and the CI smoke test to 'isaaclab train', matching the uv workflow docs.
This reverts commit 2a664ff.
The kitless image copied environment.yml and tools/ into the dependency-install layer, although the install path reads neither: version pins come from [tool.isaaclab.versions] in the root pyproject.toml, and environment.yml is only read by the conda envs command. Both invalidated the cached install layer on unrelated edits. Also drop the uv cache clean step, a no-op under UV_NO_CACHE=1, and document why apps/ must stay: isaaclab parses apps/isaaclab.python.kit at import time to resolve the Nucleus asset root.
Rebuilding the Newton image from scratch on every triggered run re-downloads the full wheel set and occupies a GPU runner for the whole build. Reuse the ecr-build-push-pull action like the base and curobo images: an existing per-commit or deps-hash tag in ECR skips the build entirely, and full builds reuse the registry layer cache. The training smoke now mounts source/, scripts/ and apps/ from the checkout so a deps-cached image still runs the PR's code; this is sound because the isaaclab packages are installed editable from source/.
The Newton kitless job validates the containerized install/build path; library code is covered by the test matrix and the native install CI. Give it a dedicated run_newton_tests gate that fires only on install inputs (Dockerfile.newton, isaaclab.sh, root and per-package pyproject.toml, the install CLI implementation, and the job's own wiring), aligned with the deps-hash inputs of the ECR cache action, so unrelated source PRs no longer occupy a GPU runner. Also add pyproject.toml and isaaclab.sh to the shared docker-test patterns so dependency pin bumps trigger the docker jobs at all. Dependency drift with no repo change is accepted to surface on the next install-input PR rather than on a schedule.
The ECR login lives in a temporary docker config that the action removes in its cleanup step, so a job step running after the action cannot pull. On a deps-cache hit the action only aliased the commit tag registry-side, leaving jobs that docker-run the image in later steps of the same job without a local copy and without credentials to fetch one. Add an opt-in pull-on-deps-hit input (default false, no behavior change for existing callers) that pulls and tags the aliased image while the login is still valid, and reduce the Newton job's local image check to a fail-fast assertion.
The Newton kitless job is an install/build test of the shipped container artifact, matching the Installation Tests workflow in shape rather than the build.yaml docker jobs that feed the test matrix. Give it a dedicated workflow with its own single-output changes gate and concurrency group, and revert build.yaml's changes job to its original single-gate form (keeping the pyproject.toml and isaaclab.sh trigger patterns, which the other images consume too). The job's steps, the ecr-build-push-pull usage with pull-on-deps-hit, the per-commit tag scheme, and the ECR cache tags are unchanged, so cache history carries over.
The Newton docker workflow carries its own pyproject.toml and isaaclab.sh trigger patterns, so the copies added to build.yaml's shared docker-test gate no longer serve this PR. Removing them leaves build.yaml untouched by the PR; widening the shared gate for the other images can be considered separately.
Three workflows (build.yaml, install-ci.yml, newton-docker.yml) carried near-identical ~80-line copies of the changed-paths gating engine, differing only in pattern lists, output names, and summary wording. Move the engine into .github/actions/detect-changes with patterns/summary-title/summary-note inputs and a single run output; each caller shrinks to a sparse checkout plus one action call, with its existing output name mapped so downstream if: conditions are untouched. Gating semantics are preserved per workflow: same patterns, fail-open on non-PR events and on changed-file listing failures, and the skip-reports-green contract for branch protection.
github-actions
Bot
added
documentation
The composite-action form required each caller to add a checkout step, since local ./ actions resolve against the runner workspace. A workflow_call reusable workflow is fetched server-side from the caller's own commit, so the gate job keeps its original shape: a single API-only step with no checkout. Callers shrink to a uses: job with the patterns input; consumers reference the generic run output instead of the per-workflow output names (five mechanical renames). The gate check is now reported as 'Detect Changes / Detect Changes'; it is not a required check, so branch protection is unaffected.
…tless-container-codex
…detect-changes-action
Collaborator
|
Resolved by #6699 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
.github/workflows/detect-changes.yml(net -98 lines, and the next gated workflow gets it for free).workflow_callworkflows are fetched server-side from the caller's own commit, unlike local composite actions which need the repo on disk.1. Design
1.1 Reusable workflow over composite action
A composite action would force each caller to add a checkout step (local
./actions resolve against the runner workspace).workflow_callkeeps the gate byte-for-byte equivalent to the previous inline jobs. This is the repo's first reusable workflow; it is GitHub's intended mechanism for a callable job.1.2 Inputs / output
patterns: oneregex<whitespace>descriptionper line (regexes are whitespace-free; the first whitespace run is the column separator — avoids the tab-quoting fragility of the previous bash arrays).summary-title/summary-note: absorb the per-workflow summary wording, including build.yaml's push-vs-PR "Triggered jobs" line via a caller-side expression.runoutput; consumers referenceneeds.changes.outputs.run(five mechanical renames from the old per-workflow output names).1.3 Self-triggering
Each workflow's pattern list includes
^\.github/workflows/detect-changes\.yml$, so changes to the gate re-run the gated jobs.2. Validation
gh api) — decisions match the previous per-workflow implementations exactly.3. Notes