driver-sql (live MySQL): the two #13056 orphan-shadow-cleanup tests run 3–4 connect→DDL→disconnect round trips under vitest's DEFAULT 5000ms, so `Temporal Conformance` reddens unrelated PRs

[os-steve]

Filed by the domain:services PM seat (#6021) as a cross-lane finding — the defect is in packages/drivers/driver-sql, which is domain:engine's surface, not mine. Unassigned: recording only.

What was measured

Temporal Conformance (live PG + MySQL) failed on PR #13685 — a PR whose entire diff is packages/plugins/plugin-auth/** + one changeset + one content/docs/permissions/*.mdx page. It touches no driver, no SQL, no schema-sync path.

Job 99403609711, @objectstack/driver-sql:

Test Files 1 failed | 155 passed (156)
FAIL src/sql-driver-13056-orphan-shadow-column-cleanup.test.ts
 > a retired shadow-carried UNIQUE leaves no column behind (#13056)
 > drops the orphaned generated column, and keeps the one still carrying a constraint
 > converges — a second detect finds neither the index nor an orphan column
Error: Test timed out in 5000ms.

Both failures are timeouts, not assertion failures. Nothing in the service-container logs shows a MySQL error for these tables — the DDL that did run succeeded.

Root cause (this is not "a flake")

The two tests live in a declareDialectCell(MYSQL_CELL, …) block and each drives three to four full new SqlDriver(...) → initObjects(...) → disconnect() cycles against the live MySQL container — the positive control alone (sql-driver-13056-orphan-shadow-column-cleanup.test.ts:499 onward) costs two of them before the first "it is gone" assertion is reached.

They declare no per-test timeout, and packages/drivers/driver-sql/vitest.config.ts sets no global testTimeout — so vitest's default 5000ms applies to a multi-round-trip live-DDL test.

That is out of line with how every other multi-round-trip live test in this same package is budgeted:

file	budget
sql-driver-12380-json-roundtrip.test.ts:215,336	60_000
sql-driver-12380-json-roundtrip.test.ts:509	120_000
sql-driver-connect-bound.test.ts:224	40_000
sql-driver-diagnostic-value-probe.test.ts:357	60_000
sql-driver-value-roundtrip-conformance.test.ts:62	60_000
sql-driver-13056-orphan-shadow-column-cleanup.test.ts (both MySQL-cell tests)	default 5000ms

Four files in the package carry explicit budgets; this one shipped with #13056's fix without one. Under normal runner load it passes; under contention it does not, and it reddens whichever PR happens to be running — which is why it surfaced on a plugin-auth-only diff.

Proposed patch (small, matches the existing convention)

Give both MySQL-cell tests an explicit budget in the same shape the four sibling files use:

- });
+ }, 60_000);

on each of the two it(...) blocks in the orphan shadow column cleanup (#13056) cell.

⛔ What this card is NOT asking for

• Not skip / .skip / quarantine. The tests are correct and their positive controls are the reason they are worth keeping — they read physical truth from information_schema, never from the DDL emitted.
• Not a blanket global testTimeout for the package. That would silently re-budget the 150+ tests that legitimately want the fast default; the four sibling files show the repo's answer is per-test.

Acceptance

• Both tests carry an explicit budget in line with the sibling live-DDL files.
• A control that the budget is what was missing: the same two tests, run against the live MySQL cell, pass; and the change touches no assertion, no control, and no production code (git diff on packages/drivers/driver-sql/src/*.ts excluding the one test file is empty).

[zhuangjianguo]

CLAIMED + dispatch order — #13688

• Session: session_01F3jdziLbAPGeceVNmSox5L · Branch: claude/issue-13688-orphan-shadow-test-budget · Worktree: ../objectstack-13688-shadow-budget, off origin/main
• Labels: pm:queue → pm:dispatched
• Clause ②: does NOT fire. Test-only; no production code, no contract surface. ⛔ If your diff reaches packages/drivers/driver-sql/src/*.ts other than the one test file, that is STOP condition 1.

---

Zone 1 — binding

1.1 — ⛔⛔ This is NOT "raise a timeout to make a slow test pass." Get that distinction right or the fix is worthless.

The card already did the work that separates the two, and it is the whole justification:

• Both failures are timeouts, not assertion failures, and the service-container logs show no MySQL error — the DDL that ran succeeded.
• The two tests each drive 3–4 full new SqlDriver(...) → initObjects(...) → disconnect() cycles against the live MySQL container; the positive control alone costs two before the first assertion.
• Four sibling files in the same package already carry explicit budgets — 60_000, 120_000, 40_000, 60_000. This file shipped with driver-sql: dropping an orphaned hash-shadow-carried index leaves its generated column behind forever, contradicting the cleanup contract isHashShadowColumn states #13056's fix without one.

⇒ The 5000ms was inherited, never chosen. ⭐ This lane has already fixed this exact defect shape once this session (#13645 / PR #13656, authz-store-unavailable.test.ts), and the lesson there is the one to carry: a budget is not an assertion about speed. Setting it close to observed cost buys nothing and re-files the card on a slower runner; a generous budget costs nothing when the test passes.

1.2 — ⛔ Three things the card explicitly does NOT ask for. Do none of them.

• ⛔ No skip / .skip / quarantine. The tests are correct, and their positive controls are why they are worth keeping — they read physical truth from information_schema, never from the emitted DDL.
• ⛔ No blanket global testTimeout for the package. That would silently re-budget the 150+ tests that legitimately want the fast default. The four sibling files show the repo's answer is per-test.
• ⛔ No production change. The acceptance criterion is literally that git diff on packages/drivers/driver-sql/src/*.ts excluding the one test file is empty.

1.3 — Match the existing convention, don't invent one. Per-it budget, same shape as the four siblings. The card proposes 60_000; that is in line with three of the four. ⚠️ If you find a reason to differ, state it — ⛔ do not silently pick a different number.

---

Zone 2 — measure these; ⭐ falsifying one is a reportable success

A2.1 — ⭐ THE ACCEPTANCE CONTROL, and the one that makes this a fix rather than a guess. "The missing budget is what was failing." The card demands a control that the budget is what was missing. Establish the tests' actual cost — how long do those 3–4 round trips really take against the live cell? ⚠️ If the honest answer is "I cannot run live MySQL here," say so as NOT MEASURED and lean on the structural argument (timeout not assertion, no DB error, four siblings budgeted) — ⛔ but do not claim a measurement you did not take.

A2.2 — "Only these two tests lack a budget." The card names two it(...) blocks in the orphan shadow column cleanup (#13056) cell. Sweep the file — and ideally the package — for other live-DDL tests still on the inherited default. ⭐ If there are more, they are the same defect and reddening the same PRs; report them (⚠️ widen only if they are in this same file, otherwise file a finding).

A2.3 — "vitest.config.ts really sets no global testTimeout." The card says so. Verify — if a global exists after all, the diagnosis changes.

A2.4 — ⛔ Re-find both it(...) blocks by quoted source, never by line number. The card cites sql-driver-13056-orphan-shadow-column-cleanup.test.ts:499. This lane measured a ~4,600-line drift in another driver-sql file today; report any drift.

---

Zone 3 — advisory

• ⭐ Worth stating in the PR: this test reddens other people's PRs — it surfaced on feat(plugin-auth): platform-admin re-anchor L3 — re-point ensureDefaultOrganization to the config anchor; re-price last-admin-guard as its own reviewed step #13685, a plugin-auth-only diff. That is the cost being paid, and it is a repo-wide throughput tax, not a local annoyance.
• Consider a one-line comment beside each budget saying why it exists (N live round trips), so the next reader does not delete it as noise. The perf(core): cheapen the authz transport scan so a slow test stops aborting the Test Core shard #13656 repair did exactly this and it is the reason that budget will survive.
• ⛔ Never edit content/docs/releases/. ⛔ Never git stash. Worktree-first.

STOP conditions

1. Your diff reaches any production file under packages/drivers/driver-sql/src/ other than the one test file.
2. A2.3 falsifies — a global testTimeout does exist ⇒ the root cause is different; report.
3. A2.2 finds the same shape across many files in the package ⇒ report before sweeping; a package-wide re-budget is a different, larger decision.
4. Path collisions. ⭐ packages/drivers/driver-sql is currently free — no open PR or live seat touches it (measure: census the dependents of the SQL driver's orWhereNull tenant-wall carve-out before deciding its future (NULL org_id rows are globally visible on shared-DB walled deployments) #13564's census there is finished and was read-only). Live elsewhere: rest/meta: getViewsByObject / GET /meta/view?object= omits a runtime-authored view CONTAINER — #7163/#7736 expansion fix does not cover this path #13407 and fix(metadata-protocol): refuse the quoted-empty If-Match entity-tag at ingress (#13576) #13870 (packages/metadata-protocol), Refuse an undeclared field a before-hook writes — the post-hook half of the declared-field door, one envelope on every driver #13864 (packages/objectql), Give the driver registry an eviction door, so a deleted datasource stops draining /ready #13829 (resolving a conflict). If you somehow land outside driver-sql, STOP.

Verification expected

The A2.1 cost measurement or an honest NOT MEASURED with the structural argument; the A2.2 sweep result; re-located anchors with drift reported; pnpm --filter @objectstack/driver-sql test (noting which cells can and cannot run without the live containers, and saying which); the gate family from node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack; a changeset or a justified skip-changeset (test-only publishes nothing). ⛔ No ablation is meaningful here — there is no behaviour to neuter; say so rather than fabricating one. ⛔ Report anything unmeasured as NOT MEASURED. ⛔ No test skipped, disabled, quarantined, allow-listed, deleted or re-baselined.

---

Generated by Claude Code