nemo_gym.agent_utils.process_supervisor

Standalone Linux process supervisor to upload beside a sandboxed harness.

Uses only the standard library; Gym need not be installed in the task sandbox. The receipt confirms descendant cleanup independently of the harness result.

Module Contents

Classes

Name	Description
`CleanupReceipt`	Outcome written only after supervision and bounded cleanup finish.

Functions

Name	Description
`_drain_children`	Reap tool descendants, including processes that detached with setsid().
`_positive_seconds`	-
`_supervise`	Enforce a harness deadline, then acknowledge cleanup after all descendants exit.
`exec_timeout`	Reserve TERM grace, harness process reaping, and descendant draining.
`main`	Run COMMAND and atomically write its cleanup receipt outside the task workdir.

Data

CLEANUP_PHASE_COUNT

DEFAULT_CLEANUP_TIMEOUT

API

class nemo_gym.agent_utils.process_supervisor.CleanupReceipt

Bases: typing.TypedDict

Outcome written only after supervision and bounded cleanup finish.

cleanup_confirmed

bool

error

str | None

return_code

int | None

timed_out

bool

nemo_gym.agent_utils.process_supervisor._drain_children(
    timeout: float
) -> None

Reap tool descendants, including processes that detached with setsid().

nemo_gym.agent_utils.process_supervisor._positive_seconds(
    value: str
) -> float

nemo_gym.agent_utils.process_supervisor._supervise(
    command: list[str],
    timeout: float,
    cleanup_timeout: float = DEFAULT_CLEANUP_TIMEOUT,
    stop_path: pathlib.Path | None = None
) -> nemo_gym.agent_utils.process_supervisor.CleanupReceipt

Enforce a harness deadline, then acknowledge cleanup after all descendants exit.

nemo_gym.agent_utils.process_supervisor.exec_timeout(
    timeout: float,
    cleanup_timeout: float = DEFAULT_CLEANUP_TIMEOUT
) -> float

Reserve TERM grace, harness process reaping, and descendant draining.

The extra 30 seconds cover startup, polling, and receipt I/O beyond those bounded cleanup phases. Keep their count shared with the controller budget.

nemo_gym.agent_utils.process_supervisor.main() -> int

Run COMMAND and atomically write its cleanup receipt outside the task workdir.

nemo_gym.agent_utils.process_supervisor.CLEANUP_PHASE_COUNT = 3

nemo_gym.agent_utils.process_supervisor.DEFAULT_CLEANUP_TIMEOUT = 10.0