setWindow publishes the new page before its include demand settles (on-demand materialize / toArray)
维护者通常 1 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 30/100
- Issue 类型
- 缺陷
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- typescript
- 领域
- databases
调研方向
Start by running the repro script from the issue (bun repro.ts) to confirm the early publish, then read the Atomic window publication and Pending demand sections of packages/db/src/query/live/ARCHITECTURE.md together with the decision in #2131. The question to settle is whether include demand created by a window move should hold publication, which needs a maintainer ruling first. Done means either a fix where the window publishes once with its includes, or a documented and exposed pending-move state, with tests covering the setWindow path.
由索引模型根据 Issue 内容生成。
描述
- I've validated the bug against the latest version of DB packages (
@tanstack/[email protected];mainhas only a docs commit since)
Describe the bug
When a windowed live query has an include over an on-demand source, setWindow() publishes the new page before its include demand settles. The new parent rows appear with empty includes (undefined for materialize(findOne()), [] for toArray()), and the children fill in on a later publication.
The initial load does not do this. preload() waits for the include demand, and the first publication already has every child. The setWindow() promise also waits correctly: it settles only after the children load. Only the publication during the move is early.
This seems to go against Atomic window publication ("the public Collection publishes the completed window once") and the decision in #2131 that an on-demand window "keeps publishing its last complete snapshot" while a window move is in flight. It may instead be intended under Pending demand does not hide the parent row. I couldn't find a rule for which of these applies to include demand that a window move creates.
To Reproduce
Standalone script, no framework: eager parents, so the window move needs no async load, and on-demand children whose loadSubset answers after 200 ms. Run with bun repro.ts.
repro.ts
// bun repro.ts
import {
BasicIndex,
createCollection,
createLiveQueryCollection,
eq,
extractSimpleComparisons,
localOnlyCollectionOptions,
materialize,
} from '@tanstack/db'
type Parent = { id: number }
type Child = { id: number; parentId: number; name: string }
const CHILD_LOAD_MS = 200
const indexed = { autoIndex: 'eager', defaultIndexType: BasicIndex } as const
const start = performance.now()
const log = (message: string) => console.log(`${(performance.now() - start).toFixed(0).padStart(4)} ms ${message}`)
// Eager parents, so the window move itself needs no async load
const parents = createCollection(
localOnlyCollectionOptions({
...indexed,
getKey: (p: Parent) => p.id,
initialData: Array.from({ length: 9 }, (_, i) => ({ id: i + 1 })),
}),
)
// On-demand children, served by a backend that answers after CHILD_LOAD_MS
const backend: Array<Child> = Array.from({ length: 9 }, (_, i) => ({
id: 101 + i,
parentId: i + 1,
name: `child-of-${i + 1}`,
}))
const children = createCollection<Child, number>({
...indexed,
getKey: (c) => c.id,
syncMode: 'on-demand',
sync: {
sync: ({ begin, write, commit, markReady }) => {
markReady()
return {
loadSubset: async (options) => {
const parentIds = new Set<number>()
for (const f of extractSimpleComparisons(options.where)) {
if (f.field.join('.') !== 'parentId') continue
for (const v of f.operator === 'in' ? (f.value as Array<number>) : [f.value as number]) parentIds.add(v)
}
log(` children.loadSubset parentIds=[${[...parentIds]}]`)
await new Promise((r) => setTimeout(r, CHILD_LOAD_MS))
begin()
for (const c of backend.filter((c) => parentIds.has(c.parentId))) write({ type: 'insert', value: c })
commit()
log(' children.loadSubset resolved')
},
}
},
},
})
const page = createLiveQueryCollection({
query: (q) =>
q
.from({ p: parents })
.orderBy(({ p }) => p.id)
.limit(3)
.select(({ p }) => ({
id: p.id,
child: materialize(
q
.from({ c: children })
.where(({ c }) => eq(c.parentId, p.id))
.findOne(),
),
})),
})
const show = () => `[${page.toArray.map((r) => `${r.id}:${r.child?.name ?? 'MISSING'}`).join(', ')}]`
page.subscribeChanges(() => log(`publish ${show()} isLoadingSubset=${page.isLoadingSubset}`))
await page.preload()
log(`preload settled ${show()}`)
log('setWindow({ offset: 3, limit: 3 })')
const settlement = page.utils.setWindow({ offset: 3, limit: 3 })
if (settlement !== true) await settlement
log(`setWindow settled ${show()}`)
Output:
26 ms children.loadSubset parentIds=[1,2,3]
230 ms publish [1:child-of-1, 2:child-of-2, 3:child-of-3] isLoadingSubset=false
231 ms children.loadSubset resolved
231 ms publish [1:child-of-1, 2:child-of-2, 3:child-of-3] isLoadingSubset=false
231 ms preload settled [1:child-of-1, 2:child-of-2, 3:child-of-3]
231 ms setWindow({ offset: 3, limit: 3 })
234 ms children.loadSubset parentIds=[4,5,6]
235 ms publish [4:MISSING, 5:MISSING, 6:MISSING] isLoadingSubset=true <-- page published before its includes
436 ms publish [4:child-of-4, 5:child-of-5, 6:child-of-6] isLoadingSubset=false
436 ms children.loadSubset resolved
438 ms setWindow settled [4:child-of-4, 5:child-of-5, 6:child-of-6]
Replacing materialize(...findOne()) with toArray(...) gives the same sequence, with [] instead of undefined.
Expected behavior
The same as the initial load: while the move is pending, keep publishing the last complete window ([1, 2, 3] with their children). Then publish [4, 5, 6] once, with their children, when the setWindow() promise settles.
If publishing the parents early is intended, it would help to document that for window moves, and for the hooks to expose that a move is pending. Today useLiveQuery's isLoading / isReady only cover the initial load, and isLoadingSubset isn't returned. An app can only hide the partial page by awaiting the setWindow() promise itself.
Real-world impact
A paginated table with numbered pages, over Electric on-demand collections. The query has an orderBy and materialize includes for the related rows, and the page is moved with setWindow (the workaround described in #2291). On every page change, the cells built from includes (reference, client, delivery date, a status computed from a related row) show their empty fallback for about 900 ms. Then they switch to the real values. A status derived from the related row can briefly be wrong, not just empty.
Environment
@tanstack/[email protected]- Bun 1.4.0, Windows 11 (the app: Chrome,
@tanstack/[email protected],@tanstack/[email protected])
Additional context
- #2291 / #2316: page moves through
useLiveQuery. #2316 lists "new include-publication semantics" as out of scope, so this behavior would carry over to adapter-owned page windows. - #2131: the policy decision to hold the last complete snapshot while a window move is in flight.
- #2156: includes are routed only from the rows the window shows. That is why a move creates new child demand at all.
- 主要语言
- TypeScript
- 星标
- 4k
- 派生
- 277
- 平均合并
- 1 天 7 小时
- 30 天内合并 PR
- 176
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue,再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库,在一个分支上完成修改。
- 提交 Pull Request,并在描述里引用这个 Issue 编号。
TanStack/db 的其他 Issue
-
难度 1/5 1-3 小时 新手友好度 88/100
维护者通常 1 天内回复
-
OfflineExecutor.dispose() never settles pendingTransactionPromises可能重新可做 @Claire-ovo 于 8 天前认领,目前没有进行中的 PR。 未关闭
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 78/100
维护者通常 1 天内回复
-
Two @tanstack/db tests time out under CPU load because a cold module import counts against the 5 s default timeout可能已有人在做 @jp-knj 于 6 天前认领。 未关闭
难度 2/5 1-3 小时 新手友好度 82/100
维护者通常 1 天内回复
-
难度 3/5 1-2 天 新手友好度 52/100
维护者通常 1 天内回复
相似的 Issue
-
needs investigation
难度 2/5 1-3 小时 新手友好度 60/100
-
难度 2/5 1-3 小时 新手友好度 72/100
维护者通常 1 天内回复
-
area:visualizer tech-debt
难度 2/5 1 小时以内 新手友好度 66/100
jmbarzee/temporal-architect#292 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 66/100
riverqueue/river#1576 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 70/100
colbymchenry/codegraph#3228 ·
维护者通常 1 天内回复