Context

The upstream-sync.yml workflow in our django fork keeps each mongodb-X.Y.x fork branch current with upstream Django.

django-mongodb-backend's CI checks out one of these fork branches by ref e.g. mongodb-6.0.x and runs its test suites against it.

Nothing signals the backend that its dependency changed, so its CI does not re-run on its own. A rebase can pull in upstream changes that break the backend, and without a fresh run against the new commits, nobody would notice.

Definition of done

Add a reusable composite action in mongodb-labs/drivers-github-tools that takes two inputs: the downstream repo to re-trigger, and a ci_rerun spec describing how. The spec is a small JSON object, {"kind": ...}, with three kinds:

• evergreen:  post an evergreen retry comment on a backend PR, nudging Evergreen to re-run its patch.
• pr: re-queue the GitHub Actions runs on a backend PR's head commit.
• ref: dispatch the backend's test-python* workflows on a branch (a dispatchable ref, i.e. branch or tag — not a SHA).

This action should create a backend-scoped GitHub App token internally via actions/create-github-app-token, so callers pass only app_id / private_key and never handle a raw cross-repo token themselves.

The ci_rerun values are already defined per-branch in upstream-sync.yml's matrix:

• e.g. mongodb-6.1.x → {"kind":"evergreen","pr":422}). 

The action will accept that exact format as its input contract, so those values pass through unchanged.

Once the action is merged configure upstream-sync.yml to use it.

Pitfalls

Conceptual complexity.