I ran into this on celld 0.4.1 while using Flue. When one request wakes async work started by another, that work can stall after the waking request finishes. This affects streams and background tasks.

To reproduce, deploy this Worker with a Durable Object binding named `PROBE` for `StreamProbe`:

```js import { DurableObject } from "cloudflare:workers";

export class StreamProbe extends DurableObject { phase = "idle";

fetch(request) { const path = new URL(request.url).pathname; if (path === "/start") { this.ctx.waitUntil((async () => { await new Promise(resolve => { this.notify = resolve; }); this.phase = "resumed"; await new Promise(resolve => setTimeout(resolve, 100)); this.phase = "completed"; })()); return new Response("admitted"); } if (path === "/notify") { this.notify(); return new Response("notified"); } return Response.json({ phase: this.phase }); } }

export default { fetch(request, env) { return env.PROBE.get(env.PROBE.idFromName("probe")).fetch(request); } }; ```

Set `CELLD_URL` to the Worker’s URL, then run:

```sh curl -fsS -X POST "$CELLD_URL/start" curl -fsS -X POST "$CELLD_URL/notify" sleep 2 curl -fsS "$CELLD_URL/stats" ```

Expected: `{"phase":"completed"}`. On 0.4.1 it stays at `{"phase":"resumed"}`—the task never gets past the 100 ms timeout.

The same pattern in a `ReadableStream` producer leaves readers stuck after the first chunk.

Here is an attempt at a patch that fixes it. Works for me, but there might be some other approach that is best

[0001-Fix-cross-request-async-operation-lifetimes.patch](https://github.com/user-attachments/files/32165011/0001-Fix-cross-request-async-operation-lifetimes.patch)