fix(python-sdk): return from _iterate_events after end event to unblock wait()

[FisherXZ]

Issue

Fixes #1034.

AsyncCommandHandle.wait() blocks indefinitely after kill().

Investigation

Traced the hang through the Python async SDK:

handle.kill()
  → Commands.kill(handle.pid)            [commands/__init__.py]
      → send_signal(SIGKILL)             [envd RPC — process is killed]

handle.wait()
  → await self._wait                     [command_handle.py]
      → _handle_events()
          → async for ... in _iterate_events()
              → async for event in self._events   ← BLOCKS HERE

_iterate_events receives the "end" event from envd, sets self._result — then continues the async for loop waiting for the next event. If envd is slow to close the HTTP stream after sending "end", the loop waits indefinitely. wait() blocks with it.

A secondary issue was also found during investigation: even after fixing the blocking, self._events (the acall_server_stream async generator) held its async with http_resp: block open until the handle was garbage-collected. Because HTTP/1.1 is used (not HTTP/2), each unclosed handle tied up one TCP connection. In workloads that accumulate handles, this can exhaust the connection pool (max_connections=2000).

Fix

Commit 1 — stop blocking: Add return after setting self._result in _iterate_events. The generator exits immediately on "end", so _handle_events terminates and wait() unblocks regardless of stream-close timing.

if event.event.HasField("end"):
    self._result = CommandResult(...)
    return  # ← exit generator; stream-close timing no longer matters

Commit 2 — release connection: Add await self._events.aclose() in a finally block in _handle_events. This releases the HTTP connection deterministically in all three paths:

Path	Behaviour
Normal "end" received	async for ends; finally calls aclose()
Exception during iteration	exception caught; finally calls aclose()
disconnect() → task cancelled	CancelledError breaks loop; finally calls aclose()

Calling aclose() in the finally of the coroutine that owns the iteration avoids the concurrent-access RuntimeError that's why the # await self._events.aclose() line in disconnect() was commented out.

Two lines changed total. No new state, no new exception types, no behavioural change to kill() or disconnect().

This also applies to PTY handles — AsyncCommandHandle is shared by pty.py, so PTY sessions now terminate promptly and release their connection after the process exits.

Tests

• test_kill_via_handle — verifies handle.kill() (the instance method) returns True and the process is dead. The existing test_kill_process uses sandbox.commands.kill(pid) — a different code path.
• test_kill_handle_wait_raises — verifies handle.wait() raises CommandExitException within 5 seconds after handle.kill(). asyncio.wait_for(..., timeout=5.0) is a safety net: if the fix regresses, the test fails with TimeoutError instead of hanging silently.

Out of scope

• __aiter__ on AsyncCommandHandle — PR fix: add __aiter__ to AsyncCommandHandle for async iteration support #1192 creates a second consumer of self._events → RuntimeError. Separate issue.
• AsyncWatchHandle has the identical # await self._events.aclose() pattern. Same fix applies but left for a separate PR to keep scope small.
• JS SDK — AbortController already handles clean stream cancellation. Unaffected.

🤖 Generated with Claude Code

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[changeset-bot]

🦋 Changeset detected

Latest commit: a6bb6ad

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@e2b/python-sdk	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: dbbad45b2c

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Close event stream before returning on end event

Returning immediately here stops consuming self._events, but the stream generator is kept on the handle and is not closed in wait() (and disconnect() does not aclose() it). The async stream implementation (e2b_connect.Client.acall_server_stream) holds the HTTP stream inside an async with, so exiting early without closing/exhausting the generator can keep connections open until the handle is garbage-collected; in workloads that keep many handles, this can accumulate and exhaust the async connection pool.

Useful? React with 👍 / 👎.