home-assistant-core/homeassistant/runner.py

"""Run Home Assistant."""
from __future__ import annotations

import asyncio
from asyncio import events
import dataclasses
import logging
import os
import subprocess
import threading
from time import monotonic
import traceback
from typing import Any

import packaging.tags

from . import bootstrap
from .core import callback
from .helpers.frame import warn_use
from .util.executor import InterruptibleThreadPoolExecutor
from .util.thread import deadlock_safe_shutdown

#
# Some Python versions may have different number of workers by default
# than others.  In order to be consistent between
# supported versions, we need to set max_workers.
#
# In most cases the workers are not I/O bound, as they
# are sleeping/blocking waiting for data from integrations
# updating so this number should be higher than the default
# use case.
#
MAX_EXECUTOR_WORKERS = 64
TASK_CANCELATION_TIMEOUT = 5

_LOGGER = logging.getLogger(__name__)


@dataclasses.dataclass(slots=True)
class RuntimeConfig:
    """Class to hold the information for running Home Assistant."""

    config_dir: str
    skip_pip: bool = False
    skip_pip_packages: list[str] = dataclasses.field(default_factory=list)
    recovery_mode: bool = False

    verbose: bool = False

    log_rotate_days: int | None = None
    log_file: str | None = None
    log_no_color: bool = False

    debug: bool = False
    open_ui: bool = False

    safe_mode: bool = False


def can_use_pidfd() -> bool:
    """Check if pidfd_open is available.

    Back ported from cpython 3.12
    """
    if not hasattr(os, "pidfd_open"):
        return False
    try:
        pid = os.getpid()
        os.close(os.pidfd_open(pid, 0))
    except OSError:
        # blocked by security policy like SECCOMP
        return False
    return True


class HassEventLoopPolicy(asyncio.DefaultEventLoopPolicy):
    """Event loop policy for Home Assistant."""

    def __init__(self, debug: bool) -> None:
        """Init the event loop policy."""
        super().__init__()
        self.debug = debug
        self._watcher: asyncio.AbstractChildWatcher | None = None

    def _init_watcher(self) -> None:
        """Initialize the watcher for child processes.

        Back ported from cpython 3.12
        """
        with events._lock:  # type: ignore[attr-defined] # pylint: disable=protected-access
            if self._watcher is None:  # pragma: no branch
                if can_use_pidfd():
                    self._watcher = asyncio.PidfdChildWatcher()
                else:
                    self._watcher = asyncio.ThreadedChildWatcher()
                if threading.current_thread() is threading.main_thread():
                    self._watcher.attach_loop(
                        self._local._loop  # type: ignore[attr-defined] # pylint: disable=protected-access
                    )

    @property
    def loop_name(self) -> str:
        """Return name of the loop."""
        return self._loop_factory.__name__  # type: ignore[no-any-return,attr-defined]

    def new_event_loop(self) -> asyncio.AbstractEventLoop:
        """Get the event loop."""
        loop: asyncio.AbstractEventLoop = super().new_event_loop()
        loop.set_exception_handler(_async_loop_exception_handler)
        if self.debug:
            loop.set_debug(True)

        executor = InterruptibleThreadPoolExecutor(
            thread_name_prefix="SyncWorker", max_workers=MAX_EXECUTOR_WORKERS
        )
        loop.set_default_executor(executor)
        loop.set_default_executor = warn_use(  # type: ignore[method-assign]
            loop.set_default_executor, "sets default executor on the event loop"
        )
        # bind the built-in time.monotonic directly as loop.time to avoid the
        # overhead of the additional method call since its the most called loop
        # method and its roughly 10%+ of all the call time in base_events.py
        loop.time = monotonic  # type: ignore[method-assign]
        return loop


@callback
def _async_loop_exception_handler(_: Any, context: dict[str, Any]) -> None:
    """Handle all exception inside the core loop."""
    kwargs = {}
    if exception := context.get("exception"):
        kwargs["exc_info"] = (type(exception), exception, exception.__traceback__)

    logger = logging.getLogger(__package__)
    if source_traceback := context.get("source_traceback"):
        stack_summary = "".join(traceback.format_list(source_traceback))
        logger.error(
            "Error doing job: %s: %s",
            context["message"],
            stack_summary,
            **kwargs,  # type: ignore[arg-type]
        )
        return

    logger.error(
        "Error doing job: %s",
        context["message"],
        **kwargs,  # type: ignore[arg-type]
    )


async def setup_and_run_hass(runtime_config: RuntimeConfig) -> int:
    """Set up Home Assistant and run."""
    hass = await bootstrap.async_setup_hass(runtime_config)

    if hass is None:
        return 1

    # threading._shutdown can deadlock forever
    # pylint: disable-next=protected-access
    threading._shutdown = deadlock_safe_shutdown  # type: ignore[attr-defined]

    return await hass.async_run()


def _enable_posix_spawn() -> None:
    """Enable posix_spawn on Alpine Linux."""
    if subprocess._USE_POSIX_SPAWN:  # pylint: disable=protected-access
        return

    # The subprocess module does not know about Alpine Linux/musl
    # and will use fork() instead of posix_spawn() which significantly
    # less efficient. This is a workaround to force posix_spawn()
    # when using musl since cpython is not aware its supported.
    tag = next(packaging.tags.sys_tags())
    # pylint: disable-next=protected-access
    subprocess._USE_POSIX_SPAWN = "musllinux" in tag.platform


def run(runtime_config: RuntimeConfig) -> int:
    """Run Home Assistant."""
    _enable_posix_spawn()
    asyncio.set_event_loop_policy(HassEventLoopPolicy(runtime_config.debug))
    # Backport of cpython 3.9 asyncio.run with a _cancel_all_tasks that times out
    loop = asyncio.new_event_loop()
    try:
        asyncio.set_event_loop(loop)
        return loop.run_until_complete(setup_and_run_hass(runtime_config))
    finally:
        try:
            _cancel_all_tasks_with_timeout(loop, TASK_CANCELATION_TIMEOUT)
            loop.run_until_complete(loop.shutdown_asyncgens())
            loop.run_until_complete(loop.shutdown_default_executor())
        finally:
            asyncio.set_event_loop(None)
            loop.close()


def _cancel_all_tasks_with_timeout(
    loop: asyncio.AbstractEventLoop, timeout: int
) -> None:
    """Adapted _cancel_all_tasks from python 3.9 with a timeout."""
    to_cancel = asyncio.all_tasks(loop)
    if not to_cancel:
        return

    for task in to_cancel:
        task.cancel("Final process shutdown")

    loop.run_until_complete(asyncio.wait(to_cancel, timeout=timeout))

    for task in to_cancel:
        if task.cancelled():
            continue
        if not task.done():
            _LOGGER.warning(
                "Task could not be canceled and was still running after shutdown: %s",
                task,
            )
            continue
        if task.exception() is not None:
            loop.call_exception_handler(
                {
                    "message": "unhandled exception during shutdown",
                    "exception": task.exception(),
                    "task": task,
                }
            )