Apscheduler

- **BREAKING** Added the ``cleanup()`` scheduler method and a configuration option
(``cleanup_interval``). A corresponding abstract method was added to the ``DataStore``
class. This method purges expired job results and schedules that have exhausted their
triggers and have no more associated jobs running. Previously, schedules were
automatically deleted instantly once their triggers could no longer produce any fire
times.
- **BREAKING** Made publishing ``JobReleased`` events the responsibility of the
``DataStore`` implementation, rather than the scheduler, for consistency with the
``acquire_jobs()`` method
- **BREAKING** The ``started_at`` field was moved from ``Job`` to ``JobResult``
- **BREAKING** Removed the ``from_url()`` class methods of ``SQLAlchemyDataStore``,
``MongoDBDataStore`` and ``RedisEventBroker`` in favor of the ability to pass a
connection url to the initializer
- Added the ability to pause and unpause schedules (PR by WillDaSilva)
- Added the ``scheduled_start`` field to the ``JobAcquired`` event
- Added the ``scheduled_start`` and ``started_at`` fields to the ``JobReleased`` event
- Fixed large parts of ``MongoDBDataStore`` still calling blocking functions in the
event loop thread
- Fixed JSON serialization of triggers that had been used at least once
- Fixed dialect name checks in the SQLAlchemy job store
- Fixed JSON and CBOR serializers unable to serialize enums
- Fixed infinite loop in CalendarIntervalTrigger with UTC timezone (PR by unights)
- Fixed scheduler not resuming job processing when ``max_concurrent_jobs`` had been
reached and then a job was completed, thus making job processing possible again
(PR by MohammadAmin Vahedinia)
- Fixed the shutdown procedure of the Redis event broker
- Fixed ``SQLAlchemyDataStore`` not respecting custom schema name when creating enums
- Fixed skipped intervals with overlapping schedules in ``AndTrigger``
(911 <https://github.com/agronholm/apscheduler/issues/911>_; PR by Bennett Meares)
- Fixed implicitly created client instances in data stores and event brokers not being
closed along with the store/broker

- **BREAKING** Renamed any leftover fields named ``executor`` to ``job_executor``
(this breaks data store compatibility)
- **BREAKING** Switched to using the timezone aware timestamp column type on Oracle
- **BREAKING** Fixed precision issue with interval columns on MySQL
- **BREAKING** Fixed datetime comparison issues on SQLite and MySQL
- **BREAKING** Worked around datetime microsecond precision issue on MongoDB
- **BREAKING** Renamed the ``worker_id`` field to ``scheduler_id`` in the
``JobAcquired`` and ``JobReleased`` events
- **BREAKING** Added the ``task_id`` attribute to the ``ScheduleAdded``,
``ScheduleUpdated`` and ``ScheduleRemoved`` events
- **BREAKING** Added the ``finished`` attribute to the ``ScheduleRemoved`` event
- **BREAKING** Added the ``logger`` parameter to ``Datastore.start()`` and
``EventBroker.start()`` to make both use the scheduler's assigned logger
- **BREAKING** Made the ``apscheduler.marshalling`` module private
- Added the ``configure_task()`` and ``get_tasks()`` scheduler methods
- Fixed out of order delivery of events delivered using worker threads
- Fixed schedule processing not setting job start deadlines correctly

- **BREAKING** The scheduler classes were moved to be importable (only) directly from
the ``apscheduler`` package (``apscheduler.Scheduler`` and
``apscheduler.AsyncScheduler``)
- **BREAKING** Removed the "tags" field in schedules and jobs (this will be added back
when the feature has been fully thought through)
- **BREAKING** Removed the ``JobInfo`` class in favor of just using the ``Job`` class
(which is now immutable)
- **BREAKING** Workers were merged into schedulers. As the ``Worker`` and
``AsyncWorker`` classes have been removed, you now need to pass
``role=SchedulerRole.scheduler`` to the scheduler to prevent it from processing due
jobs. The worker event classes (``WorkerEvent``, ``WorkerStarted``, ``WorkerStopped``)
have also been removed.
- **BREAKING** The synchronous interfaces for event brokers and data stores have been
removed. Synchronous libraries can still be used to implement these services through
the use of ``anyio.to_thread.run_sync()``.
- **BREAKING** The ``current_worker`` context variable has been removed
- **BREAKING** The ``current_scheduler`` context variable is now specified to only
contain the currently running instance of a **synchronous** scheduler
(``apscheduler.Scheduler``). The asynchronous scheduler instance can be fetched from
the new ``current_async_scheduler`` context variable, and will always be available
when a scheduler is running in the current context, while ``current_scheduler`` is
only available when the synchronous wrapper is being run.
- **BREAKING** Changed the initialization of data stores and event brokers to use a
single ``start()`` method that accepts an ``AsyncExitStack`` (and, depending on the
interface, other arguments too)
- **BREAKING** Added a concept of "job executors". This determines how the task function
is executed once picked up by a worker. Several data structures and scheduler methods
have a new field/parameter for this, ``job_executor``. This addition requires database
schema changes too.
- Dropped support for Python 3.7
- Added support for Python 3.12
- Added the ability to run jobs in worker processes, courtesy of the ``processpool``
executor
- Added the ability to run jobs in the Qt event loop via the ``qt`` executor
- Added the ``get_jobs()`` scheduler method
- The synchronous scheduler now runs an asyncio event loop in a thread, acting as a
façade for ``AsyncScheduler``
- Fixed the ``schema`` parameter in ``SQLAlchemyDataStore`` not being applied
- Fixed SQLalchemy 2.0 compatibility

- **BREAKING** Changed the scheduler API to always require a call to either
``run_until_stopped()`` or ``start_in_background()`` to start the scheduler (using it
as a context manager is no longer enough)
- **BREAKING** Replaced ``from_asyncpg_pool()`` with ``from_dsn()`` in the asyncpg event
broker
- Added an async Redis event broker
- Added automatic reconnection to the Redis event brokers (sync and async)
- Added automatic reconnection to the asyncpg event broker
- Changed ``from_async_sqla_engine()`` in asyncpg event broker to only copy the
connection options instead of directly using the engine
- Simplified the MQTT event broker by providing a default ``client`` instance if omitted
- Fixed ``CancelledError`` being reported as a crash on Python 3.7
- Fixed JSON/CBOR serialization of ``JobReleased`` events

This was a major rewrite/redesign of most parts of the project. See the
:doc:`migration section <migration>` section for details.

.. warning:: The v4.0 series is provided as a **pre-release** and may change in a
backwards incompatible fashion without any migration pathway, so do NOT use this
release in production!

- Made persistent data stores shareable between multiple processes and nodes
- Enhanced data stores to be more resilient against temporary connectivity failures
- Refactored executors (now called *workers*) to pull jobs from the data store so they
can be run independently from schedulers
- Added full async support (:mod:`asyncio` and Trio_) via AnyIO_
- Added type annotations to the code base
- Added the ability to queue jobs directly without scheduling them
- Added alternative serializers (CBOR, JSON)
- Added the ``CalendarInterval`` trigger
- Added the ability to access the current scheduler (under certain circumstances),
current worker and the currently running job via context-local variables
- Added schedule level support for jitter
- Made triggers stateful
- Added threshold support for ``AndTrigger``
- Migrated from ``pytz`` time zones to standard library ``zoneinfo`` zones
- Allowed a wider range of tzinfo implementations to be used (though ``zoneinfo`` is
preferred)
- Changed ``IntervalTrigger`` to start immediately instead of first waiting for one
interval
- Changed ``CronTrigger`` to use Sunday as weekday number 0, as per the crontab standard
- Dropped support for Python 2.X, 3.5 and 3.6
- Dropped support for the Qt, Twisted, Tornado and Gevent schedulers
- Dropped support for the Redis, RethinkDB and Zookeeper job stores

.. _Trio: https://pypi.org/project/trio/
.. _AnyIO: https://github.com/agronholm/anyio

* Removed a leftover check for pytz ``localize()`` and ``normalize()`` methods