mirror of
https://github.com/meta-llama/llama-stack.git
synced 2025-12-21 23:02:25 +00:00
e710622d4c
# What does this PR do?
Various fixes to integration test recording + stainless calling of
integration tests:
1. only the library client was being run, they all should be
2. the git check grabs diffs like:
M tests/integration/client-typescript/package-lock.json
M tests/integration/client-typescript/package.json
it should not
additionally:
Fixes rebase conflicts when stainless workflow runs integration tests
with
record-if-missing mode on PRs. Previously, the workflow would:
1. Commit all files in tests/integration/ (including non-recordings)
2. Try to rebase and push to 'main' instead of the PR branch
3. Fail with merge conflicts on PR-specific changes
Changes:
- Add pr_head_ref and is_fork_pr parameters flowing through workflow
chain
- Use target-branch input instead of github.ref_name in recording
commits
- Detect and handle fork PRs by skipping push and uploading recordings
as artifacts
- Add 7-day artifact retention for fork PR recordings
- Support both workflow_call and direct pull_request trigger contexts
For same-repo PRs: recordings now commit/push to the PR branch correctly
For fork PRs: recordings upload as downloadable artifacts with
instructions
you can see a failing workflow:
5846590613
with the rebase issues.
---------
Signed-off-by: Charlie Doern <cdoern@redhat.com>
17 lines
548 B
JSON
17 lines
548 B
JSON
{
|
|
"default": [
|
|
{"suite": "base", "setup": "ollama"},
|
|
{"suite": "base", "setup": "ollama-postgres", "allowed_clients": ["server"], "stack_config": "server:ci-tests::run-with-postgres-store.yaml"},
|
|
{"suite": "vision", "setup": "ollama-vision"},
|
|
{"suite": "responses", "setup": "gpt"},
|
|
{"suite": "base-vllm-subset", "setup": "vllm"}
|
|
],
|
|
"stainless": [
|
|
{"suite": "base", "setup": "ollama", "inference_mode": "record-if-missing"}
|
|
],
|
|
"schedules": {
|
|
"1 0 * * 0": [
|
|
{"suite": "base", "setup": "vllm"}
|
|
]
|
|
}
|
|
}
|