Repository navigation
Validate existing Business ID table shape before allocation #171
Description
Problem
JdbcBusinessIdAllocator.ensureSchema() currently accepts an existing table when scope_key, current_value, version, and updated_at names are present. It does not verify that scope_key is unique/primary or that the counter and version columns are integral, non-null, and large enough. The example JDBC adapter currently returns only column_name, so allocator-only checks cannot prove this contract.
A pre-existing wrong-shape table can pass startup and fail later or compromise allocation correctness. This is separate from #169, which makes table creation portable and checks required column presence.
Acceptance
- Define a provider-neutral metadata contract for column type, nullability, and the unique/primary key needed by Business ID allocation.
- Fail closed at explicit
context.ensureSchema()with a table/column-specific diagnostic when an existing table cannot satisfy the allocator contract; do not silently rewrite production tables. - Add negative tests for missing uniqueness, nullable counters, and incompatible/narrow counter types, plus repeat-ensure positive tests.
- Exercise SQLite and the available live PostgreSQL/MySQL/SQL Server gates; document any database-specific metadata limits rather than treating skipped or absent gates as passing.
- Keep this issue separate from the 1.549 release and stacked Fail fast on non-contention JDBC Business ID errors #166/Make JDBC Business ID schema setup portable on SQL Server #170 PRs until those prerequisites land.
Activity
philipgreat commentedon Sep 23, 2026
Draft implementation is PR #172, exact head bb3d5b4. Local SQLite example tests pass 12/12 and the concurrent wrong-shape race test passes 1/1, all zero skips. Exact-head hosted Build and Test, examples, PostgreSQL/MySQL Business ID (2/2), and SQL Server Business ID (2/2) checks all pass with zero failures/errors/skips. This remains a stacked, unpublished change: signed 1.549 release and prerequisite PRs #166/#170 must land before merge. DM8/Oracle shape validation and released-package replay are not claimed.
philipgreat commentedon Sep 23, 2026
Additional local live evidence on exact PR #172 head bb3d5b4: isolated DM8 and Oracle tests ran together with TEAQL_REQUIRE_LIVE_DB=true and an isolated DM8 guard; Surefire reported 2 tests, 0 failures, 0 errors, 0 skips. Each covers repeat ensure, 80 allocations across independent connections, and restart continuation. These are local-only checks, not hosted CI or published-package evidence. The two dedicated test containers and their nonpersistent databases were removed afterward.
philipgreat commentedon Sep 23, 2026
Exact-head d3958bf strengthens PR #172 with live malformed-table negatives. Hosted PostgreSQL/MySQL Business ID suite: 4/4; hosted SQL Server Business ID suite: 3/3; each zero failures/errors/skips. Full Build and Test plus examples also pass on the same head. This adds real-provider fail-closed evidence but remains draft/unmerged pending signed 1.549 release and prerequisite PRs #166/#170.