Skip to content

Recover search-engine builds and tests

runbookobservedEvidence reviewed 2026-10-06

An isolated appliance search-engine worktree on Shipyard. Real devices, cameras and production services are not targets.

Run the worktree script’s status from the selected checkout. Read its stateDirectory and private build/service logs. A build-lock wait can be another legitimate compile; check the owning task before interrupting anything. Confirm the configured base includes both the lifecycle script and test-port support.

For a compile failure, fix the code/toolchain cause and rerun setup; the last working service should remain until a successful replacement. Rust does not hot reload. Tests must use worktree.py test, which supplies separate dependencies and endpoint overrides. Do not invoke raw Cargo against shared fixed ports as a workaround.

Ordinary test failures clean test containers. After SIGKILL or machine failure, use the selected task’s archive path to clean abandoned owned resources after inspecting ownership, retaining its data/config. Do not remove the shared Cargo cache to fix every build failure.

Check the runtime URL, run the relevant isolated test filter and confirm no neighboring task was affected. Existing validation passed the same 194 tests in library and binary targets; doctests are not covered by the custom runner. Fixed-port test failures on a different branch may be the unresolved application compatibility dependency, not a local service outage.