Repository navigation
test(config): pin the direct-exchange Finalize status CAS and the checked_at future bound against a real database #1869
Description
Non-blocking test gaps raised by the cold reviews of #1858 (#1836).
- Finalize's
AND status = 'submitted'(internal/config/store_postgres_ri_exchange_direct.go) is pinned only by the pgxmock SQL-string test; no real-database test fails if it is loosened. Add an integration test that finalizes a row twice and a resolved/unknown row and expectsErrRIExchangeDirectExecutionNotSubmitted. - The future bound on
checked_atis pinned only at now+1h (store_postgres_ri_exchange_direct_resolve_boundary_integration_test.go); a +1s or +1m loosening would likely pass. Add cases just past the database clock. - Optional: a metric or alert on the error-level log for a late provider outcome that contradicts an operator resolution.
Activity
cristim commentedon Nov 10, 2026
Claimed by claude-ui-fix-1 (coordinator: next unclaimed platform item). Items 1 and 2 (real-database tests for the Finalize status CAS and the checked_at bound just past the database clock); item 3 (alert on a contradicting late outcome) is left for a decision.
cristim commentedon Nov 10, 2026
Landed on main as 8ea1a88 (claude-ui-fix-1), closing this. Item 1: a real-database test finalizes rows already completed, failed or unknown and expects ErrRIExchangeDirectExecutionNotSubmitted with the stored row untouched (a double finalize was already pinned by TestPostgresStoreDB_RIExchangeDirectExecution). Item 2: integration cases a few seconds and a minute past the database clock, plus a fixed-clock unit test of checkRIExchangeDirectResolve pinning the exact edge (dbNow accepted, one microsecond past it refused). Item 3 (alert on a late outcome that contradicts an operator resolution) is not done: it needs a decision on where such an alert would go; the error-level log stays.