test(upgrade): pin typed-struct plateau on legacy-origin in-place upgrade (QA-606) by kriszyp · Pull Request #1870 · HarperFast/harper

kriszyp · 2026-07-20T14:15:54Z

What this pins

A qa-explorer regression anchor for the §2 G(cross-major-upgrade) × A(width-heterogeneous) interaction — the corner QA-478 explicitly left uncovered.

QA-478 proved a harperdb@4.7.34 → current-main in-place upgrade migrates width-drifting records decode-clean, but it never drove new writes against the upgraded table with randomAccessFields: true active — so it never exercised the unbounded typed-struct minting path (the #1453-class OOM). This spec closes that gap.

Controlled A/B on one instance, identical width-heterogeneous generator + cadence on both arms:

upgraded — WidthDrift, the v4.7.34-created, in-place-migrated dataset (a table with no persisted randomAccessFields directive, so it follows the global default on open).
fresh control — FreshControl, created by current main with an explicit randomAccessFields: true.

Verdict (green): both arms plateau at typedStructs = 38 (cap 256), upgraded/fresh ratio 1.00x — the OOM class does not reproduce on a legacy-origin table; the upgraded arm tracks the fresh-install control.

Provenance & gating

Promoted from qa-explorer QA-606 / candidate P-395.
Cold-gated GREEN on main 56a2bace9 — 3/3 runs (both arms tailFlat=true, legacy precondition fired, i.e. not a silent skip). oxlint + prettier clean.

⚠️ CI note — legacy provisioning required

This spec skips (like its sibling integrationTests/upgrade/4.x-upgrade.test.ts) unless HARPER_LEGACY_VERSION_PATH points at an installed v4.x harperdb package. It rides the same env var and skip-guard the existing 4.x upgrade suite already uses — no new mechanism — but CI must provision a v4.x harperdb install for it to actually run rather than skip.

🤖 Generated with Claude Code

…rade (QA-606) Regression anchor for the §2 G(cross-major-upgrade) × A(width-heterogeneous) interaction: a harperdb@4.7.34 -> current-main in-place-upgraded table, once `randomAccessFields: true` is active and NEW width-drifting writes are driven against it, must NOT route onto the unbounded typed-struct minting path (#1453-class OOM). It must plateau its typedStructs count identically to a fresh-install control under the same generator. Controlled A/B on one instance: the "upgraded" arm (WidthDrift, the QA-478 v4-origin migrated dataset) vs a "fresh control" arm (FreshControl, created by current main with an explicit randomAccessFields directive). Both driven by the identical width-heterogeneous generator at the same cadence. Verdict: both plateau at typedStructs=38 (cap 256), ratio 1.00x -> the OOM class does not reproduce on legacy-origin tables. Skips (like the sibling 4.x-upgrade.test.ts) unless HARPER_LEGACY_VERSION_PATH points at a v4.x `harperdb` install -- CI must provision one or this silently skips. Promoted from qa-explorer QA-606 / candidate P-395. Cold-gated GREEN on main 56a2bac (3/3 runs). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

gemini-code-assist

Code Review

This pull request introduces a new integration test suite (qa606-upgrade-structgrowth.test.ts) and its associated custom resource components to verify that an in-place upgraded database table plateaus its typed-struct count under a width-heterogeneous write workload when randomAccessFields is enabled, preventing OOM regressions. The reviewer feedback suggests replacing several instances of the global fetch API with the authenticated client.reqRest helper to ensure consistency, proper authorization, and cleaner response parsing.

gemini-code-assist · 2026-07-20T14:17:38Z

+				);
+
+				// ── PRECONDITION 2 (hard): the migrated dataset decodes clean BEFORE we touch it ─
+				const scan = await fetch(`${ctx.harper.httpURL}/ScanVerify606/?table=WidthDrift`).then((r) => r.json());


Use the authenticated client.reqRest helper instead of the global fetch API. This ensures consistency with the rest of the test suite, automatically includes the required authorization headers, and avoids potential 401 Unauthorized errors if authentication is enforced on custom resources.

Suggested change

const scan = await fetch(`${ctx.harper.httpURL}/ScanVerify606/?table=WidthDrift`).then((r) => r.json());

const scan = (await client.reqRest('/ScanVerify606/?table=WidthDrift')).body;

gemini-code-assist · 2026-07-20T14:17:38Z

+				const preFlip = (await (
+					await fetch(`${ctx.harper.httpURL}/StructReport606/?table=WidthDrift`)
+				).json()) as StructReportResult;


Use the authenticated client.reqRest helper instead of the global fetch API to maintain consistency, ensure proper authorization, and simplify the response parsing.

const preFlip = (await client.reqRest('/StructReport606/?table=WidthDrift')).body as StructReportResult;

gemini-code-assist · 2026-07-20T14:17:39Z

+				const postFlipWidthDrift = (await (
+					await fetch(`${ctx.harper.httpURL}/StructReport606/?table=WidthDrift`)
+				).json()) as StructReportResult;


Use the authenticated client.reqRest helper instead of the global fetch API to maintain consistency, ensure proper authorization, and simplify the response parsing.

const postFlipWidthDrift = (await client.reqRest('/StructReport606/?table=WidthDrift')).body as StructReportResult;

gemini-code-assist · 2026-07-20T14:17:39Z

+				const postFlipFreshControl = (await (
+					await fetch(`${ctx.harper.httpURL}/StructReport606/?table=FreshControl`)
+				).json()) as StructReportResult;


Use the authenticated client.reqRest helper instead of the global fetch API to maintain consistency, ensure proper authorization, and simplify the response parsing.

const postFlipFreshControl = (await client.reqRest('/StructReport606/?table=FreshControl')).body as StructReportResult;

claude · 2026-07-20T14:22:58Z

Reviewed; no blockers found.

cb1kenobi

I'm not sure I understand what this pull request is doing.

cb1kenobi · 2026-07-22T22:22:05Z

+			typedStructs: enc && Array.isArray(enc.typedStructs) ? enc.typedStructs.length : null,
+			classicStructures: enc && Array.isArray(enc.structures) ? enc.structures.length : null,
+			maxOwnStructures: enc ? (enc.maxOwnStructures ?? null) : null,
+			storeCtor: store && store.constructor && store.constructor.name,


nit: could be simplified:

Suggested change

storeCtor: store && store.constructor && store.constructor.name,

storeCtor: store?.constructor?.name,

Ethan-Arrowood

LGTM. Solid regression anchor — it pins the real invariant (a legacy-origin, in-place-upgraded table plateaus its typed-struct count instead of the #1453 OOM growth under randomAccessFields writes), and it's genuinely CI-exercised rather than dark: the integration workflow installs harperdb@4 and sets HARPER_LEGACY_VERSION_PATH, and the fail-loud preconditions (migration ran, decode-clean, structure flipped) stop it from silent-skipping to a false green. Green across Node 22/24/26.

Two non-blocking notes: worth folding in the fetch→client.reqRest consistency cleanup, and keep an eye on the added CI wall-clock (legacy boot + migrate + 2 restarts + 36k inserts, ~600s timeout) as the upgrade suite grows.

_{sent with Claude Opus 4.8}

gemini-code-assist Bot reviewed Jul 20, 2026

View reviewed changes

kriszyp marked this pull request as ready for review July 21, 2026 18:43

github-actions Bot requested review from Ethan-Arrowood, cb1kenobi and heskew July 21, 2026 18:43

cb1kenobi approved these changes Jul 22, 2026

View reviewed changes

Ethan-Arrowood approved these changes Jul 23, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

test(upgrade): pin typed-struct plateau on legacy-origin in-place upgrade (QA-606)#1870

test(upgrade): pin typed-struct plateau on legacy-origin in-place upgrade (QA-606)#1870
kriszyp wants to merge 1 commit into
mainfrom
kris/qa-promote-upgrade-structgrowth

kriszyp commented Jul 20, 2026

Uh oh!

gemini-code-assist Bot left a comment

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Uh oh!

claude Bot commented Jul 20, 2026

Uh oh!

cb1kenobi left a comment

Uh oh!

cb1kenobi Jul 22, 2026

Uh oh!

Ethan-Arrowood left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	const scan = await fetch(`${ctx.harper.httpURL}/ScanVerify606/?table=WidthDrift`).then((r) => r.json());
	const scan = (await client.reqRest('/ScanVerify606/?table=WidthDrift')).body;

	storeCtor: store && store.constructor && store.constructor.name,
	storeCtor: store?.constructor?.name,

Uh oh!

Conversation

kriszyp commented Jul 20, 2026

What this pins

Provenance & gating

⚠️ CI note — legacy provisioning required

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist Bot Jul 20, 2026

Choose a reason for hiding this comment

Uh oh!

claude Bot commented Jul 20, 2026

Uh oh!

cb1kenobi left a comment

Choose a reason for hiding this comment

Uh oh!

cb1kenobi Jul 22, 2026

Choose a reason for hiding this comment

Uh oh!

Ethan-Arrowood left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants