imbajin opened a new pull request, #400:
URL: https://github.com/apache/hugegraph-ai/pull/400
## Problem
The current selector already isolates client, SDK, LLM and MCP changes well.
The remaining mismatch is deployment-only work: every non-documentation path
under `docker/` selects the complete LLM unit/integration job, even for a
Dockerfile, Compose file or Helm template change.
That costs an extra runner with dependency installation and a HugeGraph
service, while it still does not validate the changed deployment definition
itself.
## Changes
- Add a `deployment` selection with granular preflight checks.
- Keep all checks on the existing preflight runner; no deployment runner or
matrix is added.
- Map changes to the relevant validation:
- `Dockerfile.llm` / `Dockerfile.nk` / `.dockerignore` → BuildKit `build
--check` only, without executing image build steps;
- Compose files / environment template → `docker compose config --quiet`;
- `docker/charts/**` → `helm lint` and `helm template`;
- `scripts/build_llm_image.sh` → `bash -n` plus the referenced LLM
Dockerfile check.
- Keep Python helpers under `docker/` on LLM + quality + CodeQL because they
can affect runtime behavior.
- Add a manual `deployment` test-set and expose the exact selected checks in
the job summary.
- Extend selector tests for Dockerfile, Compose, Helm, script,
`.dockerignore`, and Python-helper boundaries.
## Runner behavior
| Change | Before | After |
| --- | --- | --- |
| Dockerfile / `.dockerignore` | preflight + full LLM + result | preflight
checks + result |
| Compose / Helm only | preflight + full LLM + result | preflight
config/lint + result |
| Python helper under `docker/` | preflight + LLM + quality/security as
applicable + result | unchanged runtime checks, plus the relevant Dockerfile
check when required |
The steady-state saving is one full LLM runner for each pure deployment PR.
The stable `CI Result` job remains unchanged for branch protection.
## Validation design
The PR itself changes `ci.yml` and `scripts/ci.py`, so it intentionally
selects every lightweight deployment check once. This verifies the hosted
runner's Buildx, Compose and Helm commands without introducing a permanent
extra runner.
Related work:
- `hugegraph/actions#41`: independent `rag` / `rag-bin` publication and
checkpoints.
- `hugegraph/actions#42`: reproducible image size/startup/container-memory
benchmark.
- `hugegraph/hugegraph-ai#97`: `rag-bin` dependency caching and
runtime-image slimming.
Source PR: https://github.com/hugegraph/hugegraph-ai/pull/98
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
