CI flake: `@objectstack/example-showcase#test` fails on mass sys_* `Find operation failed` + `Acquire connection error: operation timed out` — hit a merge-queue build and a PR run within 2h on unrelated prose-only diffs

[qq9340100]

Filed by the spec-lane PM seat (session session_016gcKVsiywU9CcS96S5t9qD) after the same signature killed two unrelated CI runs in one afternoon. Filed unlabeled for triage grading; the fix does not land in packages/spec — routing is triage's.

The signature

Test Core (3/3) shard → turbo task @objectstack/example-showcase#test exits 1 (bare ELIFECYCLE aggregate). The test log carries:

• dozens of [sql-driver] DATABASE_ERROR — the backend refused a read on 'sys_user' / 'sys_member' / 'sys_position' / 'sys_user_position' / 'sys_user_permission_set' / 'sys_setting' / 'sys_metadata' with the driver's no-verdict wrapper ("could not attribute the failure to any part of the request");
• the anchor line: Acquire connection error: Error: operation timed out for an unknown reason;
• zero AssertionError / Test timed out lines — suites that reach their summary in the same shard pass (e.g. rate-limit integration 8/8).

Two occurrences, both on diffs that cannot touch a query path

1. Merge-queue build 32374407797 (2026-08-20 ~13:37Z) — kicked PR docs(spec): widen the #8778 organizationField scope pin to name the cloud#1395 ruling and its three sanctioned consumers #10168, whose entire diff is JSDoc comments in packages/spec/src/data/object.zod.ts. Dequeued, re-queued once, then merged clean.
2. PR run 32383717677 (2026-08-20 ~15:20Z) — red on PR docs(spec): record the live read points of element:button.icon and object-metric.icon (#10053) #10270, whose diff is describe prose + accept-pins in packages/spec/src/ui/component.zod.ts.

Context: the merge-queue triage bot's 24h ledger that morning counted 38 failed queue builds (excluding the 13:37 one). This signature could not be aggregated by the bot either time — the vitest FAIL line with a test filename never reaches the readable log (the failing shard dies at the turbo/ELIFECYCLE layer), which is also why cross-PR aggregation misses it.

Initial judgment (PM, not a root-cause)

Connection-pool acquisition timing out under CI load in the showcase example's DB-backed suites — a load/infra class, not a per-PR regression: the same shard runs the heaviest example app, and the failures blanket every sys_* table at once rather than naming one query. Whether the fix is pool sizing in the showcase test harness, shard rebalancing, or a retry at the acquire layer is the implementing lane's call.

Re-check

Any red Test Core (3/3) job: fetch its log and grep for Acquire connection error + example-showcase#test. A hit is this card; an AssertionError is not.

Refs: PR #10168 (dequeue + re-queue audit comment with the first extraction) · PR #10270 (second occurrence) · merge-queue-triage workflow #4859 (whose FAIL-line parser this signature evades).

[qq9340100]

Triage: pm:queue · domain:devx · type Bug (triage seat, session session_01C5fGRTUHXFevsEdMEFDadM). Rationale: connection-pool acquisition timing out under CI load in the showcase example's DB-backed suites — the fix lands in the showcase test harness / Test Core shard configuration (pool sizing, shard rebalance, or an acquire-layer retry), which is the CI/gate surface domain:devx owns; it does not land in packages/spec despite both observed hits being spec-prose PRs. Two independent occurrences within 2h on prose-only diffs make it a load/infra class, not a per-PR regression. Flake-ledger note: the signature ("Acquire connection error" + example-showcase#test, no AssertionError, dies at the turbo/ELIFECYCLE layer so the FAIL-line parser cannot aggregate it) is a candidate for the merge-queue signature ledger — left as a request here, ⛔ not self-added, per the human-only escalation rule. Adjacent context for the implementing seat: #10227 (Test Core runner-contention decision), #10228 (turbo cache invalidation), #10260 (contended wall-clock finding).

---

Generated by Claude Code

[qq9340100]

Third occurrence today — now dequeuing PRs. Escalated to priority:p1 (triage seat, session session_01C5fGRTUHXFevsEdMEFDadM, during a live maintainer merge-queue review).

Merge-queue run 32391924279 (16:25–16:39Z, gh-readonly-queue/main/pr-10253-…): Test Core (3/3) → @objectstack/example-showcase#test exited 1 at the ELIFECYCLE layer after 12m43s, zero vitest FAIL lines in the readable log (the completeness guard's own note lists example-showcase among unattributed summaries — the same "signature evades the FAIL-line parser" property this card records). PR #10253 was dequeued and every speculative build behind it rebuilt (16:29 / 16:32 / 16:34 waves, another full-queue rebuild at 16:40) — each cascade adds ~20 minutes to every PR behind the failure, which is the measured queue-latency tax that motivated this escalation. Occurrences: merge-queue 13:37Z (#10168's build) · PR run 15:20Z (PR #10270) · merge-queue 16:39Z (this one).

For the implementing seat: the fix (pool sizing in the showcase test harness / acquire-layer retry / shard rebalance) is now the highest-leverage single change for merge-queue throughput. Priority p1 applies the queue's Bug-first tie-break; the signature-ledger addition request in the triage comment above stands (human-only escalation).

---

Generated by Claude Code

[qq9340100]

Claim: maintainer direct-dispatch (standing channel, 2026-08-10; this order verbatim: 「10293 10227-C 10228 这几个你立刻派发处理」, live session 2026-08-20 ~17:0xZ)
Session: session_01C5fGRTUHXFevsEdMEFDadM (triage seat acting under the direct-dispatch channel for these three named cards only)
Branch: claude/issue-10293-showcase-pool-flake
Worktree: objectstack-issue-10293
Domain: domain:devx
File surface: examples/app-showcase/** (test harness / vitest config / pool wiring) + at most a shard-placement line in scripts/partition-test-shards.mjs if rebalance is the chosen fix (stop on breach; explain in the report)
Container & model: M diagnosis-heavy Bug, mode:subagent, model: opus — dispatch-gates --tier this run: "no path-derived mandate … the tier stays the PM's per-card judgment call"
Clause-②: no — CI/test-infra reliability fix, no contract accept/reject change, no public surface
Serial constraints cleared: no open PR touches examples/app-showcase test harness (checked lane inventory this round); disjoint from the #10227/#10228 pair (.github/workflows/ci.yml) dispatched in the same order. Read-coupling note: if the fix is shard rebalance, coordinate with #10227's shard-count change via the issue — prefer the in-package fix precisely to stay off that file.

---

Generated by Claude Code