Skip to content

[Bug] Sandbox SDK's _send loses data with custom timeout settings #256

Description

@Bug-Keeper

sdk/python/forkd/sandbox.py:207-220:

def _send(self, msg: dict) -> dict:
    host, _, port_s = self.target.rpartition(":")
    port = int(port_s)
    with socket.create_connection((host, port), timeout=5) as s:
        s.settimeout(self.timeout + 5)
        s.sendall((json.dumps(msg) + "\n").encode())
        s.shutdown(socket.SHUT_WR)
        buf = bytearray()
        while True:
            chunk = s.recv(65536)
            if not chunk:
                break
            buf.extend(chunk)
    return json.loads(buf.decode())

Two issues:

  1. rpartition(":") mishandles IPv6 targets. If FORKD_TARGET=[::1]:8888, rpartition(":") splits at the last colon, giving host="[::1]" and port_s="8888". The [ and ] are then passed to create_connection, which (depending on Python version) either dies or silently fails. Not common today since the default is 10.42.0.2:8888, but the target argument is user-configurable and this'll bite someone.

  2. json.loads(buf.decode()) on an empty buf raises JSONDecodeError. If the guest agent dies mid-response or the timeout fires before any data is received, the user sees a baffling parse error instead of "connection closed" / "timeout". The TS SDK handles the empty-body case (if (!text) return null as T; at controller.ts:251); the Python sandbox doesn't.

For #1: use urllib.parse.urlsplit("//" + target).hostname and .port. Handles IPv6 brackets correctly.

For #2: if not buf: raise RuntimeError(f"forkd: empty response from {self.target}") before json.loads.

Both small, both worth a regression test in sdk/python/tests/ (which doesn't currently exist for sandbox — only example.py shows usage).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions