"""Provide methods to bootstrap a Home Assistant instance.""" from __future__ import annotations import asyncio from collections import defaultdict import contextlib from functools import partial from itertools import chain import logging import logging.handlers from operator import contains, itemgetter import os import platform import sys import threading from time import monotonic from typing import TYPE_CHECKING, Any # Import cryptography early since import openssl is not thread-safe # _frozen_importlib._DeadlockError: deadlock detected by _ModuleLock('cryptography.hazmat.backends.openssl.backend') import cryptography.hazmat.backends.openssl.backend # noqa: F401 import voluptuous as vol import yarl from . import ( block_async_io, config as conf_util, config_entries, core, loader, requirements, ) # Pre-import frontend deps which have no requirements here to avoid # loading them at run time and blocking the event loop. We do this ahead # of time so that we do not have to flag frontend deps with `import_executor` # as it would create a thundering heard of executor jobs trying to import # frontend deps at the same time. from .components import ( api as api_pre_import, # noqa: F401 auth as auth_pre_import, # noqa: F401 config as config_pre_import, # noqa: F401 default_config as default_config_pre_import, # noqa: F401 device_automation as device_automation_pre_import, # noqa: F401 diagnostics as diagnostics_pre_import, # noqa: F401 file_upload as file_upload_pre_import, # noqa: F401 group as group_pre_import, # noqa: F401 history as history_pre_import, # noqa: F401 http, # not named pre_import since it has requirements image_upload as image_upload_import, # noqa: F401 - not named pre_import since it has requirements logbook as logbook_pre_import, # noqa: F401 lovelace as lovelace_pre_import, # noqa: F401 onboarding as onboarding_pre_import, # noqa: F401 recorder as recorder_import, # noqa: F401 - not named pre_import since it has requirements repairs as repairs_pre_import, # noqa: F401 search as search_pre_import, # noqa: F401 sensor as sensor_pre_import, # noqa: F401 system_log as system_log_pre_import, # noqa: F401 webhook as webhook_pre_import, # noqa: F401 websocket_api as websocket_api_pre_import, # noqa: F401 ) from .components.sensor import recorder as sensor_recorder # noqa: F401 from .const import ( FORMAT_DATETIME, KEY_DATA_LOGGING as DATA_LOGGING, REQUIRED_NEXT_PYTHON_HA_RELEASE, REQUIRED_NEXT_PYTHON_VER, SIGNAL_BOOTSTRAP_INTEGRATIONS, ) from .exceptions import HomeAssistantError from .helpers import ( area_registry, category_registry, config_validation as cv, device_registry, entity, entity_registry, floor_registry, issue_registry, label_registry, recorder, restore_state, template, translation, ) from .helpers.dispatcher import async_dispatcher_send from .helpers.storage import get_internal_store_manager from .helpers.system_info import async_get_system_info from .helpers.typing import ConfigType from .setup import ( BASE_PLATFORMS, DATA_SETUP_STARTED, async_get_setup_timings, async_notify_setup_error, async_set_domains_to_be_loaded, async_setup_component, ) from .util.async_ import create_eager_task from .util.logging import async_activate_log_queue_handler from .util.package import async_get_user_site, is_virtual_env with contextlib.suppress(ImportError): # Ensure anyio backend is imported to avoid it being imported in the event loop from anyio._backends import _asyncio # noqa: F401 if TYPE_CHECKING: from .runner import RuntimeConfig _LOGGER = logging.getLogger(__name__) SETUP_ORDER_SORT_KEY = partial(contains, BASE_PLATFORMS) ERROR_LOG_FILENAME = "home-assistant.log" # hass.data key for logging information. DATA_REGISTRIES_LOADED = "bootstrap_registries_loaded" LOG_SLOW_STARTUP_INTERVAL = 60 SLOW_STARTUP_CHECK_INTERVAL = 1 STAGE_1_TIMEOUT = 120 STAGE_2_TIMEOUT = 300 WRAP_UP_TIMEOUT = 300 COOLDOWN_TIME = 60 DEBUGGER_INTEGRATIONS = {"debugpy"} CORE_INTEGRATIONS = {"homeassistant", "persistent_notification"} LOGGING_INTEGRATIONS = { # Set log levels "logger", # Error logging "system_log", "sentry", } FRONTEND_INTEGRATIONS = { # Get the frontend up and running as soon as possible so problem # integrations can be removed and database migration status is # visible in frontend "frontend", } RECORDER_INTEGRATIONS = { # Setup after frontend # To record data "recorder", } DISCOVERY_INTEGRATIONS = ("bluetooth", "dhcp", "ssdp", "usb", "zeroconf") STAGE_1_INTEGRATIONS = { # We need to make sure discovery integrations # update their deps before stage 2 integrations # load them inadvertently before their deps have # been updated which leads to using an old version # of the dep, or worse (import errors). *DISCOVERY_INTEGRATIONS, # To make sure we forward data to other instances "mqtt_eventstream", # To provide account link implementations "cloud", # Ensure supervisor is available "hassio", } DEFAULT_INTEGRATIONS = { # These integrations are set up unless recovery mode is activated. # # Integrations providing core functionality: "analytics", # Needed for onboarding "application_credentials", "backup", "frontend", "hardware", "logger", "network", "system_health", # # Key-feature: "automation", "person", "scene", "script", "tag", "zone", # # Built-in helpers: "counter", "input_boolean", "input_button", "input_datetime", "input_number", "input_select", "input_text", "schedule", "timer", } DEFAULT_INTEGRATIONS_RECOVERY_MODE = { # These integrations are set up if recovery mode is activated. "frontend", } DEFAULT_INTEGRATIONS_SUPERVISOR = { # These integrations are set up if using the Supervisor "hassio", } CRITICAL_INTEGRATIONS = { # Recovery mode is activated if these integrations fail to set up "frontend", } SETUP_ORDER = ( # Load logging as soon as possible ("logging", LOGGING_INTEGRATIONS), # Setup frontend and recorder ("frontend, recorder", {*FRONTEND_INTEGRATIONS, *RECORDER_INTEGRATIONS}), # Start up debuggers. Start these first in case they want to wait. ("debugger", DEBUGGER_INTEGRATIONS), ) # # Storage keys we are likely to load during startup # in order of when we expect to load them. # # If they do not exist they will not be loaded # PRELOAD_STORAGE = [ "core.logger", "core.network", "http.auth", "image", "lovelace_dashboards", "lovelace_resources", "core.uuid", "lovelace.map", "bluetooth.passive_update_processor", "bluetooth.remote_scanners", "assist_pipeline.pipelines", "core.analytics", "auth_module.totp", ] async def async_setup_hass( runtime_config: RuntimeConfig, ) -> core.HomeAssistant | None: """Set up Home Assistant.""" hass = core.HomeAssistant(runtime_config.config_dir) async_enable_logging( hass, runtime_config.verbose, runtime_config.log_rotate_days, runtime_config.log_file, runtime_config.log_no_color, ) hass.config.safe_mode = runtime_config.safe_mode hass.config.skip_pip = runtime_config.skip_pip hass.config.skip_pip_packages = runtime_config.skip_pip_packages if runtime_config.skip_pip or runtime_config.skip_pip_packages: _LOGGER.warning( "Skipping pip installation of required modules. This may cause issues" ) if not await conf_util.async_ensure_config_exists(hass): _LOGGER.error("Error getting configuration path") return None _LOGGER.info("Config directory: %s", runtime_config.config_dir) loader.async_setup(hass) block_async_io.enable() config_dict = None basic_setup_success = False if not (recovery_mode := runtime_config.recovery_mode): await hass.async_add_executor_job(conf_util.process_ha_config_upgrade, hass) try: config_dict = await conf_util.async_hass_config_yaml(hass) except HomeAssistantError as err: _LOGGER.error( "Failed to parse configuration.yaml: %s. Activating recovery mode", err, ) else: if not is_virtual_env(): await async_mount_local_lib_path(runtime_config.config_dir) basic_setup_success = ( await async_from_config_dict(config_dict, hass) is not None ) if config_dict is None: recovery_mode = True elif not basic_setup_success: _LOGGER.warning("Unable to set up core integrations. Activating recovery mode") recovery_mode = True elif any(domain not in hass.config.components for domain in CRITICAL_INTEGRATIONS): _LOGGER.warning( "Detected that %s did not load. Activating recovery mode", ",".join(CRITICAL_INTEGRATIONS), ) # Ask integrations to shut down. It's messy but we can't # do a clean stop without knowing what is broken with contextlib.suppress(TimeoutError): async with hass.timeout.async_timeout(10): await hass.async_stop() recovery_mode = True old_config = hass.config old_logging = hass.data.get(DATA_LOGGING) hass = core.HomeAssistant(old_config.config_dir) if old_logging: hass.data[DATA_LOGGING] = old_logging hass.config.skip_pip = old_config.skip_pip hass.config.skip_pip_packages = old_config.skip_pip_packages hass.config.internal_url = old_config.internal_url hass.config.external_url = old_config.external_url # Setup loader cache after the config dir has been set loader.async_setup(hass) if recovery_mode: _LOGGER.info("Starting in recovery mode") hass.config.recovery_mode = True http_conf = (await http.async_get_last_config(hass)) or {} await async_from_config_dict( {"recovery_mode": {}, "http": http_conf}, hass, ) elif hass.config.safe_mode: _LOGGER.info("Starting in safe mode") if runtime_config.open_ui: hass.add_job(open_hass_ui, hass) return hass def open_hass_ui(hass: core.HomeAssistant) -> None: """Open the UI.""" import webbrowser # pylint: disable=import-outside-toplevel if hass.config.api is None or "frontend" not in hass.config.components: _LOGGER.warning("Cannot launch the UI because frontend not loaded") return scheme = "https" if hass.config.api.use_ssl else "http" url = str( yarl.URL.build(scheme=scheme, host="127.0.0.1", port=hass.config.api.port) ) if not webbrowser.open(url): _LOGGER.warning( "Unable to open the Home Assistant UI in a browser. Open it yourself at %s", url, ) async def async_load_base_functionality(hass: core.HomeAssistant) -> None: """Load the registries and cache the result of platform.uname().processor.""" if DATA_REGISTRIES_LOADED in hass.data: return hass.data[DATA_REGISTRIES_LOADED] = None def _cache_uname_processor() -> None: """Cache the result of platform.uname().processor in the executor. Multiple modules call this function at startup which executes a blocking subprocess call. This is a problem for the asyncio event loop. By primeing the cache of uname we can avoid the blocking call in the event loop. """ _ = platform.uname().processor # Load the registries and cache the result of platform.uname().processor translation.async_setup(hass) entity.async_setup(hass) template.async_setup(hass) await asyncio.gather( create_eager_task(get_internal_store_manager(hass).async_initialize()), create_eager_task(area_registry.async_load(hass)), create_eager_task(category_registry.async_load(hass)), create_eager_task(device_registry.async_load(hass)), create_eager_task(entity_registry.async_load(hass)), create_eager_task(floor_registry.async_load(hass)), create_eager_task(issue_registry.async_load(hass)), create_eager_task(label_registry.async_load(hass)), hass.async_add_executor_job(_cache_uname_processor), create_eager_task(template.async_load_custom_templates(hass)), create_eager_task(restore_state.async_load(hass)), create_eager_task(hass.config_entries.async_initialize()), create_eager_task(async_get_system_info(hass)), ) async def async_from_config_dict( config: ConfigType, hass: core.HomeAssistant ) -> core.HomeAssistant | None: """Try to configure Home Assistant from a configuration dictionary. Dynamically loads required components and its dependencies. This method is a coroutine. """ start = monotonic() hass.config_entries = config_entries.ConfigEntries(hass, config) await async_load_base_functionality(hass) # Set up core. _LOGGER.debug("Setting up %s", CORE_INTEGRATIONS) if not all( await asyncio.gather( *( create_eager_task(async_setup_component(hass, domain, config)) for domain in CORE_INTEGRATIONS ) ) ): _LOGGER.error("Home Assistant core failed to initialize. ") return None _LOGGER.debug("Home Assistant core initialized") core_config = config.get(core.DOMAIN, {}) try: await conf_util.async_process_ha_core_config(hass, core_config) except vol.Invalid as config_err: conf_util.async_log_schema_error(config_err, core.DOMAIN, core_config, hass) async_notify_setup_error(hass, core.DOMAIN) return None except HomeAssistantError: _LOGGER.error( "Home Assistant core failed to initialize. Further initialization aborted" ) return None await _async_set_up_integrations(hass, config) stop = monotonic() _LOGGER.info("Home Assistant initialized in %.2fs", stop - start) if ( REQUIRED_NEXT_PYTHON_HA_RELEASE and sys.version_info[:3] < REQUIRED_NEXT_PYTHON_VER ): current_python_version = ".".join(str(x) for x in sys.version_info[:3]) required_python_version = ".".join(str(x) for x in REQUIRED_NEXT_PYTHON_VER[:2]) _LOGGER.warning( ( "Support for the running Python version %s is deprecated and " "will be removed in Home Assistant %s; " "Please upgrade Python to %s" ), current_python_version, REQUIRED_NEXT_PYTHON_HA_RELEASE, required_python_version, ) issue_registry.async_create_issue( hass, core.DOMAIN, "python_version", is_fixable=False, severity=issue_registry.IssueSeverity.WARNING, breaks_in_ha_version=REQUIRED_NEXT_PYTHON_HA_RELEASE, translation_key="python_version", translation_placeholders={ "current_python_version": current_python_version, "required_python_version": required_python_version, "breaks_in_ha_version": REQUIRED_NEXT_PYTHON_HA_RELEASE, }, ) return hass @core.callback def async_enable_logging( hass: core.HomeAssistant, verbose: bool = False, log_rotate_days: int | None = None, log_file: str | None = None, log_no_color: bool = False, ) -> None: """Set up the logging. This method must be run in the event loop. """ fmt = ( "%(asctime)s.%(msecs)03d %(levelname)s (%(threadName)s) [%(name)s] %(message)s" ) if not log_no_color: try: # pylint: disable-next=import-outside-toplevel from colorlog import ColoredFormatter # basicConfig must be called after importing colorlog in order to # ensure that the handlers it sets up wraps the correct streams. logging.basicConfig(level=logging.INFO) colorfmt = f"%(log_color)s{fmt}%(reset)s" logging.getLogger().handlers[0].setFormatter( ColoredFormatter( colorfmt, datefmt=FORMAT_DATETIME, reset=True, log_colors={ "DEBUG": "cyan", "INFO": "green", "WARNING": "yellow", "ERROR": "red", "CRITICAL": "red", }, ) ) except ImportError: pass # If the above initialization failed for any reason, setup the default # formatting. If the above succeeds, this will result in a no-op. logging.basicConfig(format=fmt, datefmt=FORMAT_DATETIME, level=logging.INFO) # Capture warnings.warn(...) and friends messages in logs. # The standard destination for them is stderr, which may end up unnoticed. # This way they're where other messages are, and can be filtered as usual. logging.captureWarnings(True) # Suppress overly verbose logs from libraries that aren't helpful logging.getLogger("requests").setLevel(logging.WARNING) logging.getLogger("urllib3").setLevel(logging.WARNING) logging.getLogger("aiohttp.access").setLevel(logging.WARNING) logging.getLogger("httpx").setLevel(logging.WARNING) sys.excepthook = lambda *args: logging.getLogger(None).exception( "Uncaught exception", exc_info=args ) threading.excepthook = lambda args: logging.getLogger(None).exception( "Uncaught thread exception", exc_info=( # type: ignore[arg-type] args.exc_type, args.exc_value, args.exc_traceback, ), ) # Log errors to a file if we have write access to file or config dir if log_file is None: err_log_path = hass.config.path(ERROR_LOG_FILENAME) else: err_log_path = os.path.abspath(log_file) err_path_exists = os.path.isfile(err_log_path) err_dir = os.path.dirname(err_log_path) # Check if we can write to the error log if it exists or that # we can create files in the containing directory if not. if (err_path_exists and os.access(err_log_path, os.W_OK)) or ( not err_path_exists and os.access(err_dir, os.W_OK) ): err_handler: ( logging.handlers.RotatingFileHandler | logging.handlers.TimedRotatingFileHandler ) if log_rotate_days: err_handler = logging.handlers.TimedRotatingFileHandler( err_log_path, when="midnight", backupCount=log_rotate_days ) else: err_handler = _RotatingFileHandlerWithoutShouldRollOver( err_log_path, backupCount=1 ) try: err_handler.doRollover() except OSError as err: _LOGGER.error("Error rolling over log file: %s", err) err_handler.setLevel(logging.INFO if verbose else logging.WARNING) err_handler.setFormatter(logging.Formatter(fmt, datefmt=FORMAT_DATETIME)) logger = logging.getLogger("") logger.addHandler(err_handler) logger.setLevel(logging.INFO if verbose else logging.WARNING) # Save the log file location for access by other components. hass.data[DATA_LOGGING] = err_log_path else: _LOGGER.error("Unable to set up error log %s (access denied)", err_log_path) async_activate_log_queue_handler(hass) class _RotatingFileHandlerWithoutShouldRollOver(logging.handlers.RotatingFileHandler): """RotatingFileHandler that does not check if it should roll over on every log.""" def shouldRollover(self, record: logging.LogRecord) -> bool: """Never roll over. The shouldRollover check is expensive because it has to stat the log file for every log record. Since we do not set maxBytes the result of this check is always False. """ return False async def async_mount_local_lib_path(config_dir: str) -> str: """Add local library to Python Path. This function is a coroutine. """ deps_dir = os.path.join(config_dir, "deps") if (lib_dir := await async_get_user_site(deps_dir)) not in sys.path: sys.path.insert(0, lib_dir) return deps_dir @core.callback def _get_domains(hass: core.HomeAssistant, config: dict[str, Any]) -> set[str]: """Get domains of components to set up.""" # Filter out the repeating and common config section [homeassistant] domains = { domain for key in config if (domain := cv.domain_key(key)) != core.DOMAIN } # Add config entry and default domains if not hass.config.recovery_mode: domains.update(DEFAULT_INTEGRATIONS) domains.update(hass.config_entries.async_domains()) else: domains.update(DEFAULT_INTEGRATIONS_RECOVERY_MODE) # Add domains depending on if the Supervisor is used or not if "SUPERVISOR" in os.environ: domains.update(DEFAULT_INTEGRATIONS_SUPERVISOR) return domains class _WatchPendingSetups: """Periodic log and dispatch of setups that are pending.""" def __init__( self, hass: core.HomeAssistant, setup_started: dict[tuple[str, str | None], float], ) -> None: """Initialize the WatchPendingSetups class.""" self._hass = hass self._setup_started = setup_started self._duration_count = 0 self._handle: asyncio.TimerHandle | None = None self._previous_was_empty = True self._loop = hass.loop def _async_watch(self) -> None: """Periodic log of setups that are pending.""" now = monotonic() self._duration_count += SLOW_STARTUP_CHECK_INTERVAL remaining_with_setup_started: defaultdict[str, float] = defaultdict(float) for integration_group, start_time in self._setup_started.items(): domain, _ = integration_group remaining_with_setup_started[domain] += now - start_time if remaining_with_setup_started: _LOGGER.debug("Integration remaining: %s", remaining_with_setup_started) elif waiting_tasks := self._hass._active_tasks: # pylint: disable=protected-access _LOGGER.debug("Waiting on tasks: %s", waiting_tasks) self._async_dispatch(remaining_with_setup_started) if ( self._setup_started and self._duration_count % LOG_SLOW_STARTUP_INTERVAL == 0 ): # We log every LOG_SLOW_STARTUP_INTERVAL until all integrations are done # once we take over LOG_SLOW_STARTUP_INTERVAL (60s) to start up _LOGGER.warning( "Waiting on integrations to complete setup: %s", self._setup_started, ) _LOGGER.debug("Running timeout Zones: %s", self._hass.timeout.zones) self._async_schedule_next() def _async_dispatch(self, remaining_with_setup_started: dict[str, float]) -> None: """Dispatch the signal.""" if remaining_with_setup_started or not self._previous_was_empty: async_dispatcher_send( self._hass, SIGNAL_BOOTSTRAP_INTEGRATIONS, remaining_with_setup_started ) self._previous_was_empty = not remaining_with_setup_started def _async_schedule_next(self) -> None: """Schedule the next call.""" self._handle = self._loop.call_later( SLOW_STARTUP_CHECK_INTERVAL, self._async_watch ) def async_start(self) -> None: """Start watching.""" self._async_schedule_next() def async_stop(self) -> None: """Stop watching.""" self._async_dispatch({}) if self._handle: self._handle.cancel() self._handle = None async def async_setup_multi_components( hass: core.HomeAssistant, domains: set[str], config: dict[str, Any], ) -> None: """Set up multiple domains. Log on failure.""" # Avoid creating tasks for domains that were setup in a previous stage domains_not_yet_setup = domains - hass.config.components # Create setup tasks for base platforms first since everything will have # to wait to be imported, and the sooner we can get the base platforms # loaded the sooner we can start loading the rest of the integrations. futures = { domain: hass.async_create_task( async_setup_component(hass, domain, config), f"setup component {domain}", eager_start=True, ) for domain in sorted( domains_not_yet_setup, key=SETUP_ORDER_SORT_KEY, reverse=True ) } results = await asyncio.gather(*futures.values(), return_exceptions=True) for idx, domain in enumerate(futures): result = results[idx] if isinstance(result, BaseException): _LOGGER.error( "Error setting up integration %s - received exception", domain, exc_info=(type(result), result, result.__traceback__), ) async def _async_resolve_domains_to_setup( hass: core.HomeAssistant, config: dict[str, Any] ) -> tuple[set[str], dict[str, loader.Integration]]: """Resolve all dependencies and return list of domains to set up.""" domains_to_setup = _get_domains(hass, config) needed_requirements: set[str] = set() platform_integrations = conf_util.extract_platform_integrations( config, BASE_PLATFORMS ) # Ensure base platforms that have platform integrations are added to # to `domains_to_setup so they can be setup first instead of # discovering them when later when a config entry setup task # notices its needed and there is already a long line to use # the import executor. # # For example if we have # sensor: # - platform: template # # `template` has to be loaded to validate the config for sensor # so we want to start loading `sensor` as soon as we know # it will be needed. The more platforms under `sensor:`, the longer # it will take to finish setup for `sensor` because each of these # platforms has to be imported before we can validate the config. # # Thankfully we are migrating away from the platform pattern # so this will be less of a problem in the future. domains_to_setup.update(platform_integrations) # Load manifests for base platforms and platform based integrations # that are defined under base platforms right away since we do not require # the manifest to list them as dependencies and we want to avoid the lock # contention when multiple integrations try to load them at once additional_manifests_to_load = { *BASE_PLATFORMS, *chain.from_iterable(platform_integrations.values()), } translations_to_load = additional_manifests_to_load.copy() # Resolve all dependencies so we know all integrations # that will have to be loaded and start right-away integration_cache: dict[str, loader.Integration] = {} to_resolve: set[str] = domains_to_setup while to_resolve or additional_manifests_to_load: old_to_resolve: set[str] = to_resolve to_resolve = set() if additional_manifests_to_load: to_get = {*old_to_resolve, *additional_manifests_to_load} additional_manifests_to_load.clear() else: to_get = old_to_resolve manifest_deps: set[str] = set() resolve_dependencies_tasks: list[asyncio.Task[bool]] = [] integrations_to_process: list[loader.Integration] = [] for domain, itg in (await loader.async_get_integrations(hass, to_get)).items(): if not isinstance(itg, loader.Integration): continue integration_cache[domain] = itg needed_requirements.update(itg.requirements) # Make sure manifests for dependencies are loaded in the next # loop to try to group as many as manifest loads in a single # call to avoid the creating one-off executor jobs later in # the setup process additional_manifests_to_load.update( dep for dep in chain(itg.dependencies, itg.after_dependencies) if dep not in integration_cache ) if domain not in old_to_resolve: continue integrations_to_process.append(itg) manifest_deps.update(itg.dependencies) manifest_deps.update(itg.after_dependencies) if not itg.all_dependencies_resolved: resolve_dependencies_tasks.append( create_eager_task( itg.resolve_dependencies(), name=f"resolve dependencies {domain}", loop=hass.loop, ) ) if unseen_deps := manifest_deps - integration_cache.keys(): # If there are dependencies, try to preload all # the integrations manifest at once and add them # to the list of requirements we need to install # so we can try to check if they are already installed # in a single call below which avoids each integration # having to wait for the lock to do it individually deps = await loader.async_get_integrations(hass, unseen_deps) for dependant_domain, dependant_itg in deps.items(): if isinstance(dependant_itg, loader.Integration): integration_cache[dependant_domain] = dependant_itg needed_requirements.update(dependant_itg.requirements) if resolve_dependencies_tasks: await asyncio.gather(*resolve_dependencies_tasks) for itg in integrations_to_process: for dep in itg.all_dependencies: if dep in domains_to_setup: continue domains_to_setup.add(dep) to_resolve.add(dep) _LOGGER.info("Domains to be set up: %s", domains_to_setup) # Optimistically check if requirements are already installed # ahead of setting up the integrations so we can prime the cache # We do not wait for this since its an optimization only hass.async_create_background_task( requirements.async_load_installed_versions(hass, needed_requirements), "check installed requirements", eager_start=True, ) # # Only add the domains_to_setup after we finish resolving # as new domains are likely to added in the process # translations_to_load.update(domains_to_setup) # Start loading translations for all integrations we are going to set up # in the background so they are ready when we need them. This avoids a # lot of waiting for the translation load lock and a thundering herd of # tasks trying to load the same translations at the same time as each # integration is loaded. # # We do not wait for this since as soon as the task runs it will # hold the translation load lock and if anything is fast enough to # wait for the translation load lock, loading will be done by the # time it gets to it. hass.async_create_background_task( translation.async_load_integrations(hass, translations_to_load), "load translations", eager_start=True, ) # Preload storage for all integrations we are going to set up # so we do not have to wait for it to be loaded when we need it # in the setup process. hass.async_create_background_task( get_internal_store_manager(hass).async_preload( [*PRELOAD_STORAGE, *domains_to_setup] ), "preload storage", eager_start=True, ) return domains_to_setup, integration_cache async def _async_set_up_integrations( hass: core.HomeAssistant, config: dict[str, Any] ) -> None: """Set up all the integrations.""" setup_started: dict[tuple[str, str | None], float] = {} hass.data[DATA_SETUP_STARTED] = setup_started watcher = _WatchPendingSetups(hass, setup_started) watcher.async_start() domains_to_setup, integration_cache = await _async_resolve_domains_to_setup( hass, config ) # Initialize recorder if "recorder" in domains_to_setup: recorder.async_initialize_recorder(hass) pre_stage_domains = [ (name, domains_to_setup & domain_group) for name, domain_group in SETUP_ORDER ] # calculate what components to setup in what stage stage_1_domains: set[str] = set() # Find all dependencies of any dependency of any stage 1 integration that # we plan on loading and promote them to stage 1. This is done only to not # get misleading log messages deps_promotion: set[str] = STAGE_1_INTEGRATIONS while deps_promotion: old_deps_promotion = deps_promotion deps_promotion = set() for domain in old_deps_promotion: if domain not in domains_to_setup or domain in stage_1_domains: continue stage_1_domains.add(domain) if (dep_itg := integration_cache.get(domain)) is None: continue deps_promotion.update(dep_itg.all_dependencies) stage_2_domains = domains_to_setup - stage_1_domains for name, domain_group in pre_stage_domains: if domain_group: stage_2_domains -= domain_group _LOGGER.info("Setting up %s: %s", name, domain_group) to_be_loaded = domain_group.copy() to_be_loaded.update( dep for domain in domain_group if (integration := integration_cache.get(domain)) is not None for dep in integration.all_dependencies ) async_set_domains_to_be_loaded(hass, to_be_loaded) await async_setup_multi_components(hass, domain_group, config) # Enables after dependencies when setting up stage 1 domains async_set_domains_to_be_loaded(hass, stage_1_domains) # Start setup if stage_1_domains: _LOGGER.info("Setting up stage 1: %s", stage_1_domains) try: async with hass.timeout.async_timeout( STAGE_1_TIMEOUT, cool_down=COOLDOWN_TIME ): await async_setup_multi_components(hass, stage_1_domains, config) except TimeoutError: _LOGGER.warning( "Setup timed out for stage 1 waiting on %s - moving forward", hass._active_tasks, # pylint: disable=protected-access ) # Add after dependencies when setting up stage 2 domains async_set_domains_to_be_loaded(hass, stage_2_domains) if stage_2_domains: _LOGGER.info("Setting up stage 2: %s", stage_2_domains) try: async with hass.timeout.async_timeout( STAGE_2_TIMEOUT, cool_down=COOLDOWN_TIME ): await async_setup_multi_components(hass, stage_2_domains, config) except TimeoutError: _LOGGER.warning( "Setup timed out for stage 2 waiting on %s - moving forward", hass._active_tasks, # pylint: disable=protected-access ) # Wrap up startup _LOGGER.debug("Waiting for startup to wrap up") try: async with hass.timeout.async_timeout(WRAP_UP_TIMEOUT, cool_down=COOLDOWN_TIME): await hass.async_block_till_done() except TimeoutError: _LOGGER.warning( "Setup timed out for bootstrap waiting on %s - moving forward", hass._active_tasks, # pylint: disable=protected-access ) watcher.async_stop() if _LOGGER.isEnabledFor(logging.DEBUG): setup_time = async_get_setup_timings(hass) _LOGGER.debug( "Integration setup times: %s", dict(sorted(setup_time.items(), key=itemgetter(1), reverse=True)), )