Files
Iliyan Angelov 6b247e5b9f Updates
2025-09-19 11:58:53 +03:00

541 lines
19 KiB
Python

"""Beat Scheduler Implementation."""
import datetime
import logging
import math
from multiprocessing.util import Finalize
try:
from zoneinfo import ZoneInfo # Python 3.9+
except ImportError:
from backports.zoneinfo import ZoneInfo # Python 3.8
from celery import current_app, schedules
from celery.beat import ScheduleEntry, Scheduler
from celery.utils.log import get_logger
from celery.utils.time import maybe_make_aware
from django.conf import settings
from django.core.exceptions import ObjectDoesNotExist
from django.db import close_old_connections, transaction
from django.db.models import Case, F, IntegerField, Q, When
from django.db.models.functions import Cast
from django.db.utils import DatabaseError, InterfaceError
from kombu.utils.encoding import safe_repr, safe_str
from kombu.utils.json import dumps, loads
from .clockedschedule import clocked
from .models import (ClockedSchedule, CrontabSchedule, IntervalSchedule,
PeriodicTask, PeriodicTasks, SolarSchedule)
from .utils import NEVER_CHECK_TIMEOUT, aware_now, now
# This scheduler must wake up more frequently than the
# regular of 5 minutes because it needs to take external
# changes to the schedule into account.
DEFAULT_MAX_INTERVAL = 5 # seconds
SCHEDULE_SYNC_MAX_INTERVAL = 300 # 5 minutes
ADD_ENTRY_ERROR = """\
Cannot add entry %r to database schedule: %r. Contents: %r
"""
logger = get_logger(__name__)
debug, info, warning = logger.debug, logger.info, logger.warning
class ModelEntry(ScheduleEntry):
"""Scheduler entry taken from database row."""
model_schedules = (
(schedules.crontab, CrontabSchedule, 'crontab'),
(schedules.schedule, IntervalSchedule, 'interval'),
(schedules.solar, SolarSchedule, 'solar'),
(clocked, ClockedSchedule, 'clocked')
)
save_fields = ['last_run_at', 'total_run_count', 'no_changes']
def __init__(self, model, app=None):
"""Initialize the model entry."""
self.app = app or current_app._get_current_object()
self.name = model.name
self.task = model.task
try:
self.schedule = model.schedule
except model.DoesNotExist:
logger.error(
'Disabling schedule %s that was removed from database',
self.name,
)
self._disable(model)
try:
self.args = loads(model.args or '[]')
self.kwargs = loads(model.kwargs or '{}')
except ValueError as exc:
logger.exception(
'Removing schedule %s for argument deseralization error: %r',
self.name, exc,
)
self._disable(model)
self.options = {}
for option in ['queue', 'exchange', 'routing_key', 'priority']:
value = getattr(model, option)
if value is None:
continue
self.options[option] = value
if getattr(model, 'expires_', None):
self.options['expires'] = getattr(model, 'expires_')
headers = loads(model.headers or '{}')
headers['periodic_task_name'] = model.name
self.options['headers'] = headers
self.total_run_count = model.total_run_count
self.model = model
if not model.last_run_at:
model.last_run_at = model.date_changed or self._default_now()
# if last_run_at is not set and
# model.start_time last_run_at should be in way past.
# This will trigger the job to run at start_time
# and avoid the heap block.
if self.model.start_time:
model.last_run_at = model.last_run_at \
- datetime.timedelta(days=365 * 30)
self.last_run_at = model.last_run_at
def _disable(self, model):
model.no_changes = True
model.enabled = False
model.save()
def is_due(self):
if not self.model.enabled:
# 5 second delay for re-enable.
return schedules.schedstate(False, 5.0)
# START DATE: only run after the `start_time`, if one exists.
if self.model.start_time is not None:
now = self._default_now()
if getattr(settings, 'DJANGO_CELERY_BEAT_TZ_AWARE', True):
now = maybe_make_aware(self._default_now())
if now < self.model.start_time:
# The datetime is before the start date - don't run.
# send a delay to retry on start_time
current_tz = now.tzinfo
start_time = self.model.due_start_time(current_tz)
time_remaining = start_time - now
delay = math.ceil(time_remaining.total_seconds())
return schedules.schedstate(False, delay)
# EXPIRED TASK: Disable task when expired
if self.model.expires is not None:
now = self._default_now()
if now >= self.model.expires:
self._disable(self.model)
# Don't recheck
return schedules.schedstate(False, NEVER_CHECK_TIMEOUT)
# ONE OFF TASK: Disable one off tasks after they've ran once
if self.model.one_off and self.model.enabled \
and self.model.total_run_count > 0:
self.model.enabled = False
self.model.total_run_count = 0 # Reset
self.model.no_changes = False # Mark the model entry as changed
self.model.save()
# Don't recheck
return schedules.schedstate(False, NEVER_CHECK_TIMEOUT)
# CAUTION: make_aware assumes settings.TIME_ZONE for naive datetimes,
# while maybe_make_aware assumes utc for naive datetimes
tz = self.app.timezone
last_run_at_in_tz = maybe_make_aware(self.last_run_at).astimezone(tz)
return self.schedule.is_due(last_run_at_in_tz)
def _default_now(self):
if getattr(settings, 'DJANGO_CELERY_BEAT_TZ_AWARE', True):
now = datetime.datetime.now(self.app.timezone)
else:
# this ends up getting passed to maybe_make_aware, which expects
# all naive datetime objects to be in utc time.
now = datetime.datetime.utcnow()
return now
def __next__(self):
self.model.last_run_at = self._default_now()
self.model.total_run_count += 1
self.model.no_changes = True
return self.__class__(self.model)
next = __next__ # for 2to3
def save(self):
# Object may not be synchronized, so only
# change the fields we care about.
obj = type(self.model)._default_manager.get(pk=self.model.pk)
for field in self.save_fields:
setattr(obj, field, getattr(self.model, field))
obj.save()
@classmethod
def to_model_schedule(cls, schedule):
for schedule_type, model_type, model_field in cls.model_schedules:
schedule = schedules.maybe_schedule(schedule)
if isinstance(schedule, schedule_type):
model_schedule = model_type.from_schedule(schedule)
model_schedule.save()
return model_schedule, model_field
raise ValueError(
f'Cannot convert schedule type {schedule!r} to model')
@classmethod
def from_entry(cls, name, app=None, **entry):
obj, created = PeriodicTask._default_manager.update_or_create(
name=name, defaults=cls._unpack_fields(**entry),
)
return cls(obj, app=app)
@classmethod
def _unpack_fields(cls, schedule,
args=None, kwargs=None, relative=None, options=None,
**entry):
entry_schedules = {
model_field: None for _, _, model_field in cls.model_schedules
}
model_schedule, model_field = cls.to_model_schedule(schedule)
entry_schedules[model_field] = model_schedule
entry.update(
entry_schedules,
args=dumps(args or []),
kwargs=dumps(kwargs or {}),
**cls._unpack_options(**options or {})
)
return entry
@classmethod
def _unpack_options(cls, queue=None, exchange=None, routing_key=None,
priority=None, headers=None, expire_seconds=None,
**kwargs):
return {
'queue': queue,
'exchange': exchange,
'routing_key': routing_key,
'priority': priority,
'headers': dumps(headers or {}),
'expire_seconds': expire_seconds,
}
def __repr__(self):
return '<ModelEntry: {} {}(*{}, **{}) {}>'.format(
safe_str(self.name), self.task, safe_repr(self.args),
safe_repr(self.kwargs), self.schedule,
)
class DatabaseScheduler(Scheduler):
"""Database-backed Beat Scheduler."""
Entry = ModelEntry
Model = PeriodicTask
Changes = PeriodicTasks
_schedule = None
_last_timestamp = None
_initial_read = True
_heap_invalidated = False
_last_full_sync = None
def __init__(self, *args, **kwargs):
"""Initialize the database scheduler."""
self._dirty = set()
Scheduler.__init__(self, *args, **kwargs)
self._finalize = Finalize(self, self.sync, exitpriority=5)
self.max_interval = (
kwargs.get('max_interval')
or self.app.conf.beat_max_loop_interval
or DEFAULT_MAX_INTERVAL)
def setup_schedule(self):
self.install_default_entries(self.schedule)
self.update_from_dict(self.app.conf.beat_schedule)
def all_as_schedule(self):
debug('DatabaseScheduler: Fetching database schedule')
s = {}
for model in self.enabled_models():
try:
s[model.name] = self.Entry(model, app=self.app)
except ValueError:
pass
return s
def enabled_models(self):
"""Return list of enabled periodic tasks.
Allows overriding how the list of periodic tasks is fetched without
duplicating the filtering/querying logic.
"""
return list(self.enabled_models_qs())
def enabled_models_qs(self):
next_schedule_sync = now() + datetime.timedelta(
seconds=SCHEDULE_SYNC_MAX_INTERVAL
)
exclude_clock_tasks_query = Q(
clocked__isnull=False,
clocked__clocked_time__gt=next_schedule_sync
)
exclude_cron_tasks_query = self._get_crontab_exclude_query()
# Combine the queries for optimal database filtering
exclude_query = exclude_clock_tasks_query | exclude_cron_tasks_query
# Fetch only the tasks we need to consider
return self.Model.objects.enabled().exclude(exclude_query)
def _get_crontab_exclude_query(self):
"""
Build a query to exclude crontab tasks based on their hour value,
adjusted for timezone differences relative to the server.
This creates an annotation for each crontab task that represents the
server-equivalent hour, then filters on that annotation.
"""
# Get server time based on Django settings
server_time = aware_now()
server_hour = server_time.hour
# Window of +/- 2 hours around the current hour in server tz.
hours_to_include = [
(server_hour + offset) % 24 for offset in range(-2, 3)
]
hours_to_include += [4] # celery's default cleanup task
# Regex pattern to match only numbers
# This ensures we only process numeric hour values
numeric_hour_pattern = r'^\d+$'
# Get all tasks with a simple numeric hour value
numeric_hour_tasks = CrontabSchedule.objects.filter(
hour__regex=numeric_hour_pattern
)
# Annotate these tasks with their server-hour equivalent
annotated_tasks = numeric_hour_tasks.annotate(
# Cast hour string to integer
hour_int=Cast('hour', IntegerField()),
# Calculate server-hour based on timezone offset
server_hour=Case(
# Handle each timezone specifically
*[
When(
timezone=timezone_name,
then=(
F('hour_int')
+ self._get_timezone_offset(timezone_name)
+ 24
) % 24
)
for timezone_name in self._get_unique_timezone_names()
],
# Default case - use hour as is
default=F('hour_int')
)
)
excluded_hour_task_ids = annotated_tasks.exclude(
server_hour__in=hours_to_include
).values_list('id', flat=True)
# Build the final exclude query:
# Exclude crontab tasks that are not in our include list
exclude_query = Q(crontab__isnull=False) & Q(
crontab__id__in=excluded_hour_task_ids
)
return exclude_query
def _get_unique_timezone_names(self):
"""Get a list of all unique timezone names used in CrontabSchedule"""
return CrontabSchedule.objects.values_list(
'timezone', flat=True
).distinct()
def _get_timezone_offset(self, timezone_name):
"""
Args:
timezone_name: The name of the timezone or a ZoneInfo object
Returns:
int: The hour offset
"""
# Get server timezone
server_time = aware_now()
# Use server_time.tzinfo directly if it is already a ZoneInfo instance
if isinstance(server_time.tzinfo, ZoneInfo):
server_tz = server_time.tzinfo
else:
server_tz = ZoneInfo(str(server_time.tzinfo))
if isinstance(timezone_name, ZoneInfo):
timezone_name = timezone_name.key
target_tz = ZoneInfo(timezone_name)
# Use a fixed point in time for the calculation to avoid DST issues
fixed_dt = datetime.datetime(2023, 1, 1, 12, 0, 0)
# Calculate the offset
dt1 = fixed_dt.replace(tzinfo=server_tz)
dt2 = fixed_dt.replace(tzinfo=target_tz)
# Calculate hour difference
offset_seconds = (
dt1.utcoffset().total_seconds() - dt2.utcoffset().total_seconds()
)
offset_hours = int(offset_seconds / 3600)
return offset_hours
def schedule_changed(self):
try:
close_old_connections()
# If MySQL is running with transaction isolation level
# REPEATABLE-READ (default), then we won't see changes done by
# other transactions until the current transaction is
# committed (Issue #41).
try:
transaction.commit()
except transaction.TransactionManagementError:
pass # not in transaction management.
last, ts = self._last_timestamp, self.Changes.last_change()
except DatabaseError as exc:
logger.exception('Database gave error: %r', exc)
return False
except InterfaceError:
warning(
'DatabaseScheduler: InterfaceError in schedule_changed(), '
'waiting to retry in next call...'
)
return False
try:
if ts and ts > (last if last else ts):
return True
finally:
self._last_timestamp = ts
return False
def reserve(self, entry):
new_entry = next(entry)
# Need to store entry by name, because the entry may change
# in the mean time.
self._dirty.add(new_entry.name)
return new_entry
def sync(self):
if logger.isEnabledFor(logging.DEBUG):
debug('Writing entries...')
_tried = set()
_failed = set()
try:
close_old_connections()
while self._dirty:
name = self._dirty.pop()
try:
self._schedule[name].save()
_tried.add(name)
except (KeyError, TypeError, ObjectDoesNotExist):
_failed.add(name)
except DatabaseError as exc:
logger.exception('Database error while sync: %r', exc)
except InterfaceError:
warning(
'DatabaseScheduler: InterfaceError in sync(), '
'waiting to retry in next call...'
)
finally:
# retry later, only for the failed ones
self._dirty |= _failed
def update_from_dict(self, mapping):
s = {}
for name, entry_fields in mapping.items():
try:
entry = self.Entry.from_entry(name,
app=self.app,
**entry_fields)
if entry.model.enabled:
s[name] = entry
except Exception as exc:
logger.exception(ADD_ENTRY_ERROR, name, exc, entry_fields)
self.schedule.update(s)
def install_default_entries(self, data):
entries = {}
if self.app.conf.result_expires:
entries.setdefault(
'celery.backend_cleanup', {
'task': 'celery.backend_cleanup',
'schedule': schedules.crontab('0', '4', '*'),
'options': {'expire_seconds': 12 * 3600},
},
)
self.update_from_dict(entries)
def schedules_equal(self, *args, **kwargs):
if self._heap_invalidated:
self._heap_invalidated = False
return False
return super().schedules_equal(*args, **kwargs)
@property
def schedule(self):
initial = update = False
current_time = datetime.datetime.now()
if self._initial_read:
debug('DatabaseScheduler: initial read')
initial = update = True
self._initial_read = False
self._last_full_sync = current_time
elif self.schedule_changed():
info('DatabaseScheduler: Schedule changed.')
update = True
self._last_full_sync = current_time
# Force update the schedule if it's been more than 5 minutes
if not update:
time_since_last_sync = (
current_time - self._last_full_sync
).total_seconds()
if (
time_since_last_sync >= SCHEDULE_SYNC_MAX_INTERVAL
):
debug(
'DatabaseScheduler: Forcing full sync after 5 minutes'
)
update = True
self._last_full_sync = current_time
if update:
self.sync()
self._schedule = self.all_as_schedule()
# the schedule changed, invalidate the heap in Scheduler.tick
if not initial:
self._heap = []
self._heap_invalidated = True
if logger.isEnabledFor(logging.DEBUG):
debug('Current schedule:\n%s', '\n'.join(
repr(entry) for entry in self._schedule.values()),
)
return self._schedule