[Bug]: commands.run hangs indefinitely when sandbox becomes unreachable — request_timeout does not set read timeout on streaming calls

[simsinght]

Sandbox ID or Build ID

No response

Environment

e2b==2.8.1
python-sdk

Timestamp of the issue

2026-02-11 15:00 PST

Frequency

One-time occurrence

Expected behavior

if sandbox is unreachable, commands.run should error out

Actual behavior

if sandbox is unreachable, commands.run hangs forever

Issue reproduction

1. Create a sandbox and run some commands successfully
2. Wait for the sandbox to become unreachable (e.g., sandbox times out or dies)
3. Call commands.run() — it hangs indefinitely because there is no read timeout on the HTTP stream

Additional context

There was a blip in e2b of some sort and we saw couple of our sandboxes die or timeout at the same time that a long running command had been issued; the command was retried automatically with the commands.run, but then all of those tasks in our system hung indefinitely getting stuck at commands.run.

My (claude's) best guess:
When a sandbox becomes unreachable (e.g., sandbox dies or network connectivity is lost), commands.run() hangs indefinitely because the underlying call_server_stream does not set a read timeout on the HTTP stream. The request_timeout parameter only sets connect and pool timeouts for streaming calls, unlike unary calls which correctly set all four (connect, pool, read, write).

In e2b_connect/client.py, _prepare_unary_request (used by files.make_dir, files.write, etc.) correctly sets all timeouts:

  extensions = {
      "timeout": {
          "connect": request_timeout,
          "pool": request_timeout,
          "read": request_timeout,
          "write": request_timeout,
      }
  }

But _prepare_server_stream_request (used by commands.run) omits read and write:

  extensions = {
      "timeout": {"connect": request_timeout, "pool": request_timeout}
  }

The timeout parameter is sent as a connect-timeout-ms header for server-side enforcement, but if the server is unreachable, the client never gets a response and httpcore's ConnectionPool.stream() blocks on read forever.

[linear]

ENG-3490

[VintLin]

Can I submit a PR to fix this issue?

I checked the Python SDK and noticed _prepare_unary_request sets all timeout phases (connect/pool/read/write), but _prepare_server_stream_request currently sets only connect/pool.
This can cause commands.run() to hang when the sandbox is unreachable or the stream stops responding.

If this approach sounds good, I can open a minimal PR to:

1. add read/write timeout for server-stream requests, and
2. add focused tests to ensure request_timeout is fully applied on the stream path.

[mishushakov]

@VintLin feel free to attempt, I will review and we can see if it fixes the issue. Also would love if you could do a minimal reproduction so we could validate the solution

[jakubno]

Fixed in #1197