Repository navigation
metadata: the cached /meta item read sends Last-Modified as the time of the read, not of the item, so the header carries no information #31196
Description
domain:server#4 · marchtian · session_017ihL2dLQEXWGsE2SZ3Qn5F
Filing gate: ① product defect (an HTTP cache header that states something false), found while fixing #31171 (PR #31195, dev hand-off; the flake's cause). ⛔ Not a claim. Triage sets the grade and the lane.
Reader: the seat that owns packages/metadata-protocol (protocol-search-and-cache.ts) and packages/rest (meta-item-read-routes.ts).
Dedupe (open AND closed): Last-Modified meta → 4 hits, the only related one is #31171 (the flake this causes in a test, fixed in the test only).
reach: every cached GET /api/v1/meta/:type/:name; read from the code on main, observed in queue builds 42383103865 and 42385456389 (two reads of one unchanged item answered Last-Modified one second apart).
class: c · type: Bug
维护者速读
元数据接口的缓存响应里，Last-Modified 填的是「这次读取的时间」，不是条目真正最后修改的时间。所以这个头没有任何信息：同一条没改过的元数据，两次读到的值不一样；按 If-Modified-Since 做的缓存校验永远判成「已修改」。现在缓存靠 ETag 正常工作，所以没坏功能，但这个头在说假话，今天还让一个测试在合并队列里抖了两次（#31171）。
What is wrong
packages/metadata-protocol/src/protocol-search-and-cache.ts~147 returnslastModified: new Date().toISOString()beside the item's contentetag.packages/rest/src/meta-item-read-routes.ts~974 renders it asLast-Modified. Two reads of an unchanged item a second apart answer differentLast-Modifiedvalues with the sameETag.
Expected
Last-Modifiedis the item's own last change (its storedupdated_at/ publish time) when the protocol knows it, or the header is omitted when it does not — never the read time. Freshness stays gated by the ETag (no-cache), unchanged.
Acceptance: two reads of one unchanged item answer the same Last-Modified (or none); a publish moves it.
Activity
objectstack-fleet commentedon Nov 7, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: audit · p3 ·
domain:server· 2026-11-07T17:55Z⛔ Not a claim.
GET /api/v1/meta/:type/:namesendsLast-Modifiedas the time of the read, not the item's own modification time. The header carries no information, and anIf-Modified-Sincerevalidation always answers "modified". ETag revalidation works, so nothing is broken. It is also the cause of the Queue-flake anchor: src/domains/mcp-object-schema-mask.test.ts #31171 flake, which was fixed in the test only.domain:server(metadata-protocol+rest).findingremoved,pm:queue.Generated by Claude Code
objectstack-fleet commentedon Nov 7, 2026
domain:server#5 · os-project-manager · session_01FEkQ5o8PE7ZtVjcor51wBZ
Claim: PM loop round 1 · 2026-11-07T18:10Z
Session:
session_01FEkQ5p9eu5KqN5GP7JZAGe(card thread, dev) · seatsession_01FEkQ5o8PE7ZtVjcor51wBZAccount:
os-project-managerBranch:
claude/project-thread-qb6zilWorktree:
objectstack-issue-31196Domain:
domain:serverSeat:
domain:server#5File surface:
packages/metadata-protocol/src/protocol-search-and-cache.ts, possiblypackages/rest/src/meta-item-read-routes.ts, their tests, root.changeset/; ⛔packages/spec/**Container & model: S,
mode:thread,model: default tier(dispatch-gates --tier: no path-derived mandate),effort: thread default(order asked medium; the coordinator sets effort only on the maintainer's word)Clause-②: no —
lastModifiedis already optional in the publishedhttp-cache.zod.ts; omitting it or sourcing the item's own time stays inside itMetadata-protocol: none
Responsibility:
packages/metadata-protocolcached item read (objectstack) | ETag revalidation (no-cache) keeps caching correct | every cachedGET /api/v1/meta/:type/:name(If-Modified-Sincealways answers modified)Thread-read: 6710978221
Premise: holds:
protocol-search-and-cache.ts:147setslastModified: new Date().toISOString()andmeta-item-read-routes.ts:973-974renders it, onorigin/main66bf5b4aSerial constraints cleared: none — no open PR or
pm:dispatchedcard touches either fileScope: Fixes #31196. Order:
/mnt/project-files/pm-orders/server/order-31196.md. Summons: the maintainer's 「server 再开两个座位」「并发都是5」 (this project, 2026-11-07); grading per 「定级你可以自己决定」 is triage's here (graded on the card).objectstack-fleet commentedon Nov 7, 2026
domain:server#5 · os-project-manager · session_01FEkQ5p9eu5KqN5GP7JZAGe
os-dev-report
{
"issue": 31196,
"status": "done",
"branch": "claude/project-thread-qb6zil",
"pr": "#31248",
"session": "session_01FEkQ5p9eu5KqN5GP7JZAGe",
"premise_still_valid": true,
"summary": "getMetaItemCached stamped lastModified with the read time (protocol-search-and-cache.ts); the REST door only renders it. The by-name read (readMetaItem) now records the adopted sys_metadata row's updated_at (created_at fallback, canonicalIsoInstant) on its active envelope as a non-enumerable Symbol.for key (new served-row-changed-at.ts); the cached read answers it, and answers no lastModified key for an item no stored row serves, so the door sends no Last-Modified. ETag, Cache-Control and the published GetMetaItemResponse envelope are unchanged. REST door unchanged; If-Modified-Since was never read by the protocol and still never answers 304 (pinned).",
"tests": "metadata-protocol src/protocol.cached-item-last-modified.test.ts 4/4 (same lastModified on two reads; save and publish each move it, a pending draft does not; code-shipped item has no key; envelope serialisation unchanged; ETag/no-cache CONTROL) and rest src/meta-item-cached-last-modified.test.ts 3/3, both re-run green on 7ff2b9e. Package suites on f59f3fa+565bea38 (later commits test-only): metadata-protocol 323 files passed | 4 skipped, 22531 tests passed | 26 skipped; rest 394 files passed, 7489 passed | 332 skipped; typecheck both packages exit 0 on 7ff2b9e; scoped eslint packages/metadata-protocol packages/rest exit 0 on 7ff2b9e. Ablation via ablation-replace.mjs (src-resolved, no dist): spread replaced by
lastModified: new Date().toISOString(),(anchor 1->0, replacement 0->1, blob c29eebfc->c033ddc6): 3 failed | 1 passed,expected '2026-11-07T08:05:00.000Z' to be '2026-11-07T08:00:00.000Z'andexpected [ Array(5) ] to not include 'lastModified'; restored blob == HEAD, git diff HEAD empty.","gates": "dispatch-gates --commands: 69 families on 7ff2b9e, all run, exit 0 each; --ran: 69 derived accounted for, 69 run, 0 NOT-MEASURED (record /mnt/project-files/pm-orders/server/31196-gates-ran.txt). Earlier rounds red only on the new test doubles, fixed in-test: engine-double-contract (ledger --write), objectql-double-limit, where-matcher, test-source-alias. CI check-runs on PR #31248: in_progress at report time.",
"dogfood_battery": "n/a — no
!in the PR title and no (narrowing) arm","mcp_calls": "0",
"api_writes": "2 — fleet-write relay pr_create (POST /repos/objectstack-ai/objectstack/dispatches → PR #31248); this os-dev-report comment via post-stamped.mjs (POST issues/31196/comments). git push not counted.",
"open_questions": [],
"security_findings": "none",
"out_of_scope_findings": [
"carrier: seat · packages/runtime/src/domains/mcp-object-schema-mask.test.ts (PR #31195) comment says the protocol stamps lastModified with the read time; false after this PR for Last-Modified (the Date header half holds), masking harmless; not touched, outside the claimed surface",
"carrier: none · deviation: file surface grew past the claim by protocol-meta-item-reads.ts, new served-row-changed-at.ts, a one-word export in sys-metadata-repository.ts, and scripts/engine-double-contract.pinned.json (the change time is reachable only inside readMetaItem; getMetaItem envelope version is a hash digest)",
"carrier: none · deviation: origin/main merged once (no-op at 66bf5b4); main moved to 30f23fe since, touching none of these files; not re-merged"
]
}
Generated by Claude Code
objectstack-fleet commentedon Nov 7, 2026
domain:server#5 · os-project-manager · session_01FEkQ5o8PE7ZtVjcor51wBZ
ACCEPT → PR #31248 comment 6712419529 (Head-sha
7ff2b9e528). Ready, auto-merge armed; the card closes on merge.