`sleepAfter` alarm fires during long-running `containerFetch()` operations

[ghostwriternr]

What happened

A container was killed by the sleepAfter alarm while containerFetch() was actively awaiting a response from the container. The operation (a matplotlib chart generation via Sandbox SDK's exec(), which uses containerFetch internally) had been running for approximately 6 seconds when the container received SIGTERM.

Client-side error:

Error: ReadableStream received over RPC disconnected prematurely.

Container-side error:

SessionDestroyedError: Session 'main' was destroyed during command execution

Note: the exact error messages come from Sandbox SDK, but sharing it here to be illustrative.

Reproduction

1. Create a DO subclass extending Container
2. Set sleepAfter to a short duration (e.g., "10s")
3. Call containerFetch() with a request that takes longer than sleepAfter to complete (e.g., a command execution that runs for 15+ seconds)
4. The alarm fires while containerFetch() is still awaiting the response, and onActivityExpired() kills the container
   This also affects back-to-back operations: if several quick operations consume most of the sleepAfter window, a subsequent longer operation can be killed even though it just started, because renewActivityTimeout() resets the deadline to now + sleepAfter, not now + sleepAfter from when the operation finishes.

Expected behaviour

The sleepAfter timer should not fire while containerFetch() is actively awaiting a response. The container is clearly in use when there's an in-flight request.

Current workaround

Before each long-running operation, extend the sleepAfter property and refresh the timeout:

this.sleepAfter = estimatedDurationSeconds + 30;
this.renewActivityTimeout();

This works but requires subclasses to predict operation duration and manage the timeout around every containerFetch() call.

Potentially related issues

• Container terminated mid-processing after ~8-10 minutes of active work #138: Container killed mid-processing (likely same root cause)
• WebSocket connections don't renew activity timeout - container sleeps despite active connections #147: WebSocket connections don't trigger activity renewal (related: no refresh at all for non-containerFetch paths)
• "SleepAfter" is not effective #127: sleepAfter appears ineffective

Environment

Observed in production using v0.1.1 of @cloudflare/containers

[whoiskatrin]

This should be fixed now, thank you