[2.0 rc.15] frames: late-mounted consumers report a false ASYNC_WATERFALL for an already-streaming response
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
- 42/100
- Issue 类型
- 缺陷
- 描述清晰度
- 基本清楚
- 活跃度
- 活跃
- 技术栈
- javascript, playwright, typescript, vite
调研方向
Read the frame client’s landing() implementation and frame handler, then trace how request origin reaches root, nested, staged, and single-flight deliveries. Start with the inline Vite reproduction and compare the late-mounted case with the immediate control; add regression coverage for the other delivery paths and retries. Done means an already-streaming frame no longer reports ASYNC_WATERFALL, while genuinely new sequential work and retries remain attributed correctly.
由索引模型根据 Issue 内容生成。
描述
Describe the bug
With server components enabled, a consumer of a server-function component reference that mounts late records ASYNC_WATERFALL for a frame response that was already streaming before the delay. There is only one request, and the content renders correctly. The finding is false: the frame delivery started before the preceding client wait resolved, not after it.
The frames client wraps the already-running delivery in a new landing promise, and that promise carries no flight origin, so attribution sees it as work that started when the consumer mounted. The repro below isolates this by acquiring the reference outside the reactive async graph, then mounting it through dynamic behind a 200 ms gate. The same page without the gate records no waterfall.
This only affects development diagnostics. The finding has info severity, so the default console doesn't print it; the repro subscribes to diagnostics to show it. It matters for tooling and tests that treat every diagnostic as a defect, where it shows up under slow mounting or CPU contention. Production rendering is unaffected.
Your Example Website or App
Minimal code is inline below (Vite + @solidjs/vite-plugin start mode, no other dependencies).
Steps to Reproduce the Bug or Issue
Create a Vite app with these three files and solid-js, @solidjs/web and @solidjs/vite-plugin installed (versions under Platform). Start mode needs no other entry files.
vite.config.ts:
import { defineConfig } from 'vite';
import solid from '@solidjs/vite-plugin';
export default defineConfig({
plugins: [solid({ start: true, ssr: true, serverFunctions: { components: true } })],
});
src/panel.tsx:
import { createMemo } from 'solid-js';
export async function serverPanel() {
'use server';
return function Panel() {
const text = createMemo(async () => {
await new Promise(resolve => setTimeout(resolve, 1000));
return 'Frame finished';
});
return <p>{text()}</p>;
};
}
src/App.tsx:
import { createSignal, createMemo, Show, Loading, OBSERVE } from 'solid-js';
import type { Component } from 'solid-js';
import { dynamic } from '@solidjs/web';
import { attribution } from 'solid-js/attribution';
import { serverPanel } from './panel';
if (typeof window !== 'undefined') {
attribution.enable({ log: false });
// Depth-two findings are informational and silent in the default console.
OBSERVE?.diagnostics.subscribe(event => {
if (event.code === 'ASYNC_WATERFALL') console.info(event.message);
});
}
export default function App() {
const [panel, setPanel] = createSignal<Component>();
async function start() {
// Acquire the reference outside a reactive async node, so request latency
// doesn't become an extra upstream flight in the immediate control.
const reference = await serverPanel();
setPanel(() => reference);
}
function Demo() {
// The server is already rendering its 1 s async body before this gate.
const ready = createMemo(() => new URLSearchParams(location.search).get('late') === '0'
? true : new Promise<true>(resolve => setTimeout(() => resolve(true), 200)));
const Panel = dynamic(() => panel());
return <Loading fallback={<p>Loading frame</p>}><Show when={ready()}><Mount /></Show></Loading>;
function Mount() {
return <Panel />;
}
}
return <main><button onClick={start}>Start</button><Show when={panel()}><Demo /></Show></main>;
}
- Run
vite, open/?late=1and open the browser console. - Click Start. The server begins its 1,000 ms async render immediately; the client mounts the consumer after its 200 ms gate, while the response is still streaming.
- Frame finished renders, and the console shows the
ASYNC_WATERFALLfinding below. The network panel shows a single frame request. - Control: open
/?late=0and click Start. The consumer mounts immediately and there is no finding.
Reproduced in 5 of 5 late runs, with 5 of 5 controls clean.
Expected behavior
A frame delivery that is already running should keep its request's flight origin when a new landing wrapper consumes it. Mounting the consumer later should not classify that delivery as new sequential async work, so no ASYNC_WATERFALL should be reported.
Actual (durations vary):
[ASYNC_WATERFALL] 2 sequential async flights — "ready" (201ms) → "computed" (789ms) — 989ms serialized: each began only after the previous resolved (as far as this graph can see). If a later request doesn't need the earlier response, derive both from the same inputs so they start together; if the dependency is intrinsic, preload the dependent data or join the requests server-side. If this work WAS already started elsewhere (a preloader or request cache), have that layer stamp its promises with attribution.markFlight() from "@solidjs/signals/attribution".
The frame is attributed 789 ms, although the server render takes 1,000 ms from the click: the request was in flight during the 200 ms gate.
Possible fix. Use the mechanism from #4108: capture the origin before the fetch, carry it through frame delivery, and call OBSERVE.attribution.markFlight(pending, startedAt) on the promise the landing memo returns. The origin should survive direct-root, nested, staged and single-flight deliveries, including commits delayed until later; retries should stay new work. The bookkeeping can be confined to dev/observe builds. This repro covers the ordinary root-delivery path; the other paths need their own regression tests. On next (d2ad8b2), landing() still returns an unstamped .then(() => value) wrapper and the frame handler doesn't propagate the request origin. I checked that by reading the source; I haven't run the repro against next.
Screenshots or Videos
(leave empty)
Platform
- OS: Linux x64
- Browser: Chromium 149 (headless, Playwright 1.61.1)
- Version:
solid-js,@solidjs/weband@solidjs/signals2.0.0-rc.15;@solidjs/vite-plugin3.0.0-next.50; Vite 8.3.3; Node 24.21.0
- 主要语言
- TypeScript
- 星标
- 36.1k
- 派生
- 1.1k
- 平均合并
- 11 小时 42 分钟
- 30 天内合并 PR
- 350
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。
solidjs/solid 的其他 Issue
-
`lazy()`: remounting an already-loaded component renders empty (or the Suspense fallback) since 1.9.14可能已有人在做 @itisrmk 于 3 天前认领。 未关闭1.x
难度 2/5 1-3 小时 新手友好度 62/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 48/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 54/100
维护者通常 1 天内回复
-
难度 5/5 一周以上 新手友好度 35/100
维护者通常 1 天内回复
-
难度 4/5 3-5 天 新手友好度 56/100
维护者通常 1 天内回复
相似的 Issue
-
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
-
bug
难度 2/5 1-3 小时 新手友好度 84/100
维护者通常 1 天内回复
-
bug
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 67/100
mobile-next/mobile-mcp#526 ·
维护者通常 1 天内回复
-
e2e-failure ready-to-code
难度 1/5 1 小时以内 新手友好度 73/100
redhat-developer/rhdh-plugin-export-overlays#4470 · 1 条评论 ·
维护者通常 1 天内回复