Migrating from previous versions of APScheduler
From v3.x to v4.0
APScheduler 4.0 has undergone a partial rewrite since the 3.x series.
There is currently no way to automatically import schedules from a persistent 3.x job store, but this shortcoming will be rectified before the final v4.0 release.
Terminology and architectural design changes
Data stores, previously called job stores, have been redesigned to work with multiple running schedulers and workers, both for purposes of scalability and fault tolerance. Many data store implementations were dropped because they were either too burdensome to support, or the backing services were not sophisticated enough to handle the increased requirements.
Event brokers are a new component in v4.0. They relay events between schedulers and workers, enabling them to work together with a shared data store. External (as opposed to local) event broker services are required in multi-node or multi-process deployment scenarios.
Triggers are now stateful. This change was found to be necessary to properly support
combining triggers (
OrTrigger), as they needed to keep track of the next run
times of all the triggers contained within. This change also enables some more
sophisticated custom trigger implementations.
Entry points are no longer used or supported, as they were more trouble than they were worth, particularly with packagers like py2exe or PyInstaller which by default did not package distribution metadata. Thus, triggers and data stores have to be explicitly instantiated.
add_job() method is now
add_schedule(). The scheduler still has
a method named
add_job(), but this is meant for making one-off runs of
a task. Previously you would have had to call
add_job() with a
DateTrigger using the current time as the run time.
The two most commonly used schedulers,
BackgroundScheduler, have often caused confusion among users and have thus been
Scheduler. This new unified scheduler class has two methods that
start() method used previously:
start_in_background(). The former should be used if you previously used
BlockingScheduler, and the latter if you used
The asyncio scheduler has been replaced with a more generic
which is based on AnyIO and thus also supports Trio in addition to
The API of the async scheduler differs somewhat from its synchronous counterpart. In
particular, it requires itself to be used as an async context manager – whereas with
the synchronous scheduler, use as a context manager is recommended but not required.
All other scheduler implementations have been dropped because they were either too burdensome to support, or did not seem necessary anymore. Some of the dropped implementations (particularly Qt) are likely to be re-added before v4.0 final.
Schedulers no longer support multiple data stores. If you need this capability, you should run multiple schedulers instead.
Configuring and running the scheduler has been radically simplified. The
method is gone, and all configuration is now passed as keyword arguments to the
As the scheduler is no longer used to create triggers, any supplied datetimes will be
assumed to be in the local time zone. If you wish to change the local time zone, you
should set the
TZ environment variable to either the name of the desired timezone
Europe/Helsinki) or to a path of a time zone file. See the tzlocal
documentation for more information.
Jitter support has been moved from individual triggers to the schedule level. This not only simplified trigger design, but also enabled the scheduler to provide information about the randomized jitter and the original run time to the user.
CronTrigger was changed to respect the standard order of
weekdays, so that Sunday is now 0 and Saturday is 6. If you used numbered weekdays
before, you must change your trigger configuration to match. If in doubt, use
abbreviated weekday names (e.g.
IntervalTrigger was changed to start immediately, instead
of waiting for the first interval to pass. If you have workarounds in place to “fix”
the previous behavior, you should remove them.
From v3.0 to v3.2
Prior to v3.1, the scheduler inadvertently exposed the ability to fetch and manipulate
jobs before the scheduler had been started. The scheduler now requires you to call
scheduler.start() before attempting to access any of the jobs in the job stores. To
ensure that no old jobs are mistakenly executed, you can start the scheduler in paused
scheduler.start(paused=True)) (introduced in v3.2) to avoid any premature job
From v2.x to v3.0
The 3.0 series is API incompatible with previous releases due to a design overhaul.
The concept of “standalone mode” is gone. For
BlockingSchedulerinstead, and for
BackgroundScheduler. BackgroundScheduler matches the old default semantics.
Job defaults (like
coalesce) must now be passed in a dictionary as the
BaseScheduler.configure(). When supplying an ini-style configuration as the first argument, they will need a corresponding
The configuration key prefix for job stores was changed from
jobstores.to match the dict-style configuration better.
max_runsoption has been dropped since the run counter could not be reliably preserved when replacing a job with another one with the same ID. To make up for this, the
end_dateoption was added to cron and interval triggers.
The old thread pool is gone, replaced by
ThreadPoolExecutor. This means that the old
threadpooloptions are no longer valid.
The trigger-specific scheduling methods have been removed entirely from the scheduler. Use the generic
BaseScheduler.add_job()method or the
@BaseScheduler.scheduled_jobdecorator instead. The signatures of these methods were changed significantly.
close_jobstoresoptions have been removed from the
BaseScheduler.shutdown()method. Executors and job stores are now always shut down on scheduler shutdown.
Scheduler.unschedule_func()have been replaced by
BaseScheduler.remove_job(). You can also unschedule a job by using the job handle returned from
Job store changes
The job store system was completely overhauled for both efficiency and forwards compatibility. Unfortunately, this means that the old data is not compatible with the new job stores. If you need to migrate existing data from APScheduler 2.x to 3.x, contact the APScheduler author.
The Shelve job store had to be dropped because it could not support the new job store design. Use SQLAlchemyJobStore with SQLite instead.
From 3.0 onwards, triggers now require a pytz timezone. This is normally provided by the
scheduler, but if you were instantiating triggers manually before, then one must be
supplied as the
The only other backwards incompatible change was that
get_next_fire_time() takes two
arguments now: the previous fire time and the current datetime.
From v1.x to 2.0
There have been some API changes since the 1.x series. This document explains the changes made to v2.0 that are incompatible with the v1.x API.
The behavior of cron scheduling with regards to default values for omitted fields has been made more intuitive – omitted fields lower than the least significant explicitly defined field will default to their minimum values except for the week number and weekday fields
SchedulerShutdownError has been removed – jobs are now added tentatively and scheduled for real when/if the scheduler is restarted
Scheduler.is_job_active() has been removed – use
job in scheduler.get_jobs()instead
dump_jobs() is now print_jobs() and prints directly to the given file or sys.stdout if none is given
repeatparameter was removed from
@Scheduler.interval_schedulein favor of the universal
Scheduler.unschedule_func()now raises a
KeyErrorif the given function is not scheduled
The semantics of
Scheduler.shutdown()have changed – the method no longer accepts a numeric argument, but two booleans
The scheduler can no longer be reconfigured while it’s running