ci: trigger dogfood eval pipeline on run-evals PR label#428
Closed
calvarjorge wants to merge 4 commits into
Closed
ci: trigger dogfood eval pipeline on run-evals PR label#428calvarjorge wants to merge 4 commits into
calvarjorge wants to merge 4 commits into
Conversation
Add a GitHub Actions workflow that launches the dogfood eval pipeline (job 398185277057549) when the `run-evals` label is added to a PR, and re-launches it on each new commit while the label stays on. The real PR head commit is passed as `appkit_ref` so the pipeline can pull the code; `prompt_preset=custom-pr` and `tags=appkit_pr:<number>` are also set. Authenticates as the apps-mcp-evals-runner service principal via OAuth M2M, and posts a sticky "Eval running" comment linking the evals-monitor PR page and the triggered job run. Comment logic lives in .github/scripts/upsert-eval-comment.cjs. Co-authored-by: Isaac Signed-off-by: Jorge Calvar <jorge.calvar@databricks.com>
Probe dogfood reachability + workspace OIDC discovery and run a forced oauth-m2m authenticated call with debug logging, to pin down the "cannot configure default credentials" failure. To be reverted once auth works. Co-authored-by: Isaac Signed-off-by: Jorge Calvar <jorge.calvar@databricks.com>
Allow manual runs to probe staging connectivity from an arbitrary runner group (runner_group/runner_labels inputs), since dogfood.staging blocks the default databricks-protected-runner-group at the network edge. A bare dispatch runs only the diagnostic; pass pr_number to also trigger the job and post the comment. Co-authored-by: Isaac Signed-off-by: Jorge Calvar <jorge.calvar@databricks.com>
The databricks-protected-runner-group's egress to internal Databricks hosts is gated by the GitHub OIDC identity. Without `id-token: write` the egress proxy returns 403 "RBAC: access denied" for every request (incl. anonymous curl to dogfood.staging), which is what broke OAuth M2M. All other Databricks workflows in this repo set this permission. Also revert the temporary manual-dispatch/configurable-runner testing scaffolding; back to label/synchronize on databricks-protected-runner-group. Co-authored-by: Isaac Signed-off-by: Jorge Calvar <jorge.calvar@databricks.com>
Contributor
Author
|
Closing for now — GitHub Actions runners can't reach dogfood.staging (network perimeter), so we can't trigger the eval job directly from CI. Keeping the branch |
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
1 participant
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.
What
Adds a GitHub Actions workflow that launches the dogfood eval pipeline (job 398185277057549) for a PR when the
run-evalslabel is present, and re-launches it on every new commit while the label stays on.How it works
pull_requestwithtypes: [labeled, synchronize].labeled+ label isrun-evals→ run.synchronize(new commit) + PR already hasrun-evals→ run.github.event.pull_request.head.sha(the real PR head commit, never the synthetic merge commit) as theappkit_refjob param, so the pipeline can pull the code. Also setsprompt_preset=custom-prandtags=appkit_pr:<number>.concurrencywithcancel-in-progress: true(grouped per PR) guarantees the sticky⏳ Eval runningcomment always reflects the most recently triggered commit, even if an earlier run's job finishes first..github/scripts/upsert-eval-comment.cjs) links the evals-monitor PR page (/prs/appkit/<number>) and the triggered job run (run_idfrom therun-nowresponse — no extra API call).apps-mcp-evals-runnerservice principal. Credentials are scoped to the trigger step only, so the PR-authored comment script never sees them.Security notes
pull_request(notpull_request_target): repo secrets are withheld from fork PRs, so an external contributor cannot exfil the credentials even by editing the workflow. Therun-evalslabel gate also requires write/triage access.Required setup before this works
run-evalslabel in the repo.apps-mcp-evals-runnerSP and add two repo secrets:EVALS_DATABRICKS_CLIENT_ID_DOGFOODEVALS_DATABRICKS_CLIENT_SECRET_DOGFOOD398185277057549.Testing
Because this is a
pull_request-triggered workflow, it runs the PR branch's version of the workflow, so it can be exercised on this PR once the label and secrets exist.This pull request and its description were written by Isaac.