test(driver-turso): the cross-process autonumber pin starts its writers once and stops waiting on fsync

[objectstack-fleet]

Part of #23182 (row 2)

Clause-②: no

Row 2 of #23182 only: @objectstack/driver-turso, src/turso-remote-autonumber-concurrency.test.ts, read at 29.3 s in run 38228297391. Refresh PR #23141 lists it over 10 s in 6 of the 7 hourly full-battery runs that executed it. #23182 remains open for rows 1 and 3 to 6.

What changed (one file, test-only)

The file pins cross-process autonumber atomicity on the remote face: two writer processes and one libSQL database file must give N distinct numbers, monotonic per writer, with evidence that the writers overlapped. Most of its time went to work it does not assert on. Measured at d4aaeaf0e on this 4-vCPU container, one round took about 4.4 s:

phase	per round, before	what it was
writer start-up	~2.2 s	node --import tsx, then the driver's dependencies (@objectstack/driver-sql and what it imports, @objectstack/spec, @libsql/client) loaded through tsx's hook: ~2 s, against ~1.2 s under plain Node. The package's own source compiled in ~40 ms.
writes (2 x 200 creates)	~2.2 s	fsync-bound. A create is 3 statements (reserve, insert, read back). A single writer took 5.9 ms per create with synchronous = FULL and 1.2 ms with synchronous = OFF.

A round whose writers did not overlap also re-spawned both writers.

Three changes, none to what is asserted:

1. Writers start once. They start once per test and serve every round over stdin/stdout, one JSON line per command and per answer. A retry round costs only its writes.
2. Dependencies load before tsx. A writer requires the driver's dependencies with plain Node first, then calls require('tsx/cjs/api').register() and requires src/index.ts. The source is still compiled by tsx, and the dependencies are the same modules, already in the require cache. tsx is already a declared devDependency, and tsx/cjs/api is one of its public exports.
3. No fsync. Each writer runs PRAGMA synchronous = OFF on its own connection. That drops only the disk flush at commit, on a throwaway temp file. The journal mode, the locks and the busy timeout are unchanged, so the cross-process serialisation the pin depends on is unchanged.

The round's ready and write times now join the existing measurement line, so a slow CI run can be attributed from its log.

Kept as they were: N = 200 per writer, at most 3 rounds, every pin, and the NOT MEASURED skip. N was not reduced: writes are no longer the cost, and a smaller N would only lower how often a broken counter is caught (see the reverse verification below).

Readings

The CI per-file number is vitest's module line (✓ path (1 test) Nms). All readings below are from this container under scripts/pm/os-verify-lock.sh, on a shared box (other seats' unlocked work was running), with vitest's default reporter (agent-detection variables unset, as in CI).

density	BASE d4aaeaf0e	this PR
the file alone	4575, 5256, 4406 ms	1836, 1840, 1853 ms (83ece05da)
whole driver-turso package (88 files)	5735, 5914 ms	1912 ms (83ece05da)
package plus driver-sql's suite at once	10658, 8741, 8973 ms	4126 ms (794b74409, same run as BASE's 8973)

Writer changes per round: 2 to 40 before, 141 to 221 after (unmutated runs).

CI reading for the after-state: NOT MEASURED here. Job logs and artefacts redirect to an Azure blob host that this container's egress proxy refuses (403). This PR's own Test Core shard carries the file's module line and the measurement line in its log.

Reverse verification (ablation)

Mutation: the warm reservation (UPDATE … SET last_value = last_value + 1 … RETURNING) split into a SELECT last_value and then an UPDATE … SET last_value = ? (read-then-write), once with no gap and once with a 1 ms gap.

• Applied through scripts/ablation-replace.mjs: the anchor hit went 1 → 0 and the blob changed. The restore was proven after every leg: blob equal to HEAD and git diff HEAD empty.
• Reach: a tsx child spawned from the package root sees the mutated method, because nextRemoteSequenceValue.toString() carries the marker. No dist is involved, since both the parent and the writers run src.
• Predicted: red on the "no write failed" pin, with UNIQUE constraint failed in a writer's error list, as the file's header recorded. Observed: that direction in every red run.

shape	no gap	1 ms gap
this PR (794b74409 and 83ece05da)	red 8 of 11	red 7 of 7
this PR's first cut 7e0c6cc50 (before the dependency preload)	red 7 of 10	red 5 of 5
BASE, same box and hour, interleaved with the first cut	red 1 of 5	not run

A non-atomic counter is caught more often than before, not less: without the fsync the writers interleave far more finely. A zero-gap read-then-write remains a probabilistic catch, as it was at BASE.

Gates and tests (at 83ece05da unless named)

• Build: pnpm --filter '@objectstack/driver-turso...' build green, after merging origin/main.
• Typecheck and tests: pnpm --filter @objectstack/driver-turso typecheck (tsc --noEmit) is green, and its tsconfig includes src/**/*, tests included. The package suite ran 88/88 files: 2373 passed, 33 skipped, the same counts as BASE.
• Driver conformance: pnpm check:driver-conformance before the first edit and after the last commit read 50 covered cell(s), 0 in the DEBT ledger, 0 exempt both times.
• Derived gates: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack gave 51 commands, identical to the dispatch-time set, and all ran. The --ran verdict reads 51 derived, 51 run, 0 NOT-MEASURED, 0 UNRUN. Three first answered exit 3 (unmet prerequisites) and were re-run green after their remedy: check-plugin-teardown-shape --self-test needed git fetch --unshallow, and check:lean-entry-closure and check:dual-build-cjs-loads needed package dists, restored from the turbo cache.
• Lint, narrowed and proven: eslint with --no-inline-config on the one changed file linted 1 file with 0 errors and 0 warnings. The file is in eslint's population (--print-config returns a config for it). The config enables no type-aware linting (the resolved parserOptions are ecmaVersion and sourceType), so this diff cannot move any other file's verdict.
• Bytes: check:nul-bytes is green, and a control-byte self-scan of the file found 0 lines.

Changeset

None. The diff is test-only: the changed file is not in driver-turso's files[] (dist, README.md, CHANGELOG.md). Measured on a dist built after the merge, four symbols unique to the test (PRELOADED_DEPENDENCIES, WORKER_SOURCE, REPLY_DEADLINE_MS, turso-autonumber-concurrency-) have 0 hits, and the positive control nextRemoteSequenceValue has 6. Label: skip-changeset.

Acceptance notes

• Import time is shared, not this file's. The whole vitest process for this one file still takes about 16 to 19 s of wall here, and about 14 to 15 s of that is vitest's import of the driver's source graph. Every file in this package pays that import, and the per-file reading does not count it.
• @objectstack/core import cost. It alone takes about 0.9 s to import under plain Node and dominates a writer's remaining start-up of about 1.2 s. That is an observation about another package's import cost, not a defect; it is not filed.
• The old ablation reading was optimistic. The file's header recorded the zero-gap ablation as red in 2 of 3 runs at BASE. On this box the same BASE shape was red in 1 of 5. The new shape records its own reading beside the old one.
• The card's pin can only be read after landing: "the next refresh PR's list no longer names the file".

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

Nothing in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 0 changed package(s)), so this run has no opinion about the docs.

What this run could not see

• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 0 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 75cada0ea6859b735635688e07e5fcba324e90e8 → packageMentionDocs.