[sitole]

Description

This PR moves the test step logic out of the workflows and into two composite actions:

• .github/actions/unit-tests runs the unit test shard for one package, including the host setup that the package needs.
• .github/actions/integration-tests runs the integration suite on one configuration, from database startup to the data race check.

The workflows (pr-tests.yml, integration_tests.yml) keep the matrices, the sharding strategy, and the Codecov and artifact uploads. With the logic in actions, other workflows can run the same test steps without a copy of them.

Changes in behavior

• The integration-tests action starts the databases itself. The action runs this step first, so the image pulls still overlap with the package builds.
• The integration-tests action takes test-shard, tests-only, and test-parallelism as inputs. Before, the workflow set these values through environment variables.
• The unit-tests action always writes coverage.txt and junit.xml. The Codecov upload steps in pr-tests.yml are unchanged.

No changes

• The test commands, timeouts, and host setup steps are identical to the previous inline versions.
• The matrices, the runners, and the aggregate jobs are unchanged.

Testing

• All four YAML files parse.
• The refactor moves the steps verbatim; the diff shows the moved blocks.

[cursor]

PR Summary

Low Risk
CI-only structural change; test commands and host setup are unchanged aside from new optional orchestrator-path wiring that defaults to the existing path.

Overview
This refactor centralizes PR unit-test and integration-test execution in reusable composite actions so other workflows can run the same steps without duplicating them. The workflows keep matrices, sharding, and Codecov or artifact uploads; shard, allow-list, and parallelism settings for integration tests move from inline env vars to action inputs. The integration action also adds optional alternate orchestrator build path support with a pgrep check so a mismatched start-services version does not silently run the default binary.

^{Reviewed by Cursor Bugbot for commit a92fbdf. Bugbot is set up for automated code reviews on this repo. Configure here.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!

[claude]

LGTM — clean, verbatim extraction of existing CI steps into reusable composite actions.

Extended reasoning...

Overview

This PR refactors .github/workflows/pr-tests.yml and .github/workflows/integration_tests.yml by extracting their inline step sequences into two new composite actions, .github/actions/unit-tests and .github/actions/integration-tests. The workflows retain the matrices, sharding, and Codecov/artifact upload steps; the actions own the mechanical work (host setup, test execution, data-race checks). I diffed the moved blocks against the new action files and confirmed the step bodies, env vars, and shell commands are unchanged — this is a verbatim move, not a rewrite. The new orchestrator-make-path input on the integration-tests action is unused by both call sites in this PR (default value only), so it introduces no behavior change today; it's scaffolding for a future caller.

Security risks

None. This is CI-only, no production code paths, and secrets handling (CODECOV_TOKEN, test-only fake credentials in start-services) is passed through unchanged via existing secret/input mechanisms.

Level of scrutiny

Low-to-moderate is appropriate: this is a CI workflow refactor, not application logic, and the description explicitly claims a verbatim move which I verified against the diff. A regression here would surface immediately and loudly (broken CI on this or the next PR) rather than silently, which lowers the cost of being wrong.

Other factors

The bug-hunting system found no issues, and the two candidate issues it investigated (orchestrator-make-path being ignored by start-services, and the pgrep verification pattern) were correctly ruled out — that input/step pair is dead code under the current call sites (always default value), so it can't affect this PR's actual behavior. No outstanding review comments exist on the PR.