ischedule

Schedule periodic tasks in a Python program. Simple syntax, precise timing, no busy waiting. The project was originally made for an industrial automation and IoT application.

Basic example

from ischedule import schedule, run_loop

def task_1():
    print("task 1")

def task_2():
    print("task 2")

schedule(task_1, interval=1.0)
schedule(task_2, interval=0.2)

run_loop()

Output:

task 2
task 2
task 2
task 2
task 1
task 2
task 2
task 2
task 2
task 2
task 1

Implementational details

Quite importantly, and unlike some other packages, ischedule takes into account the time it takes for the task function to execute. For example, if a task that takes 0.9 seconds to complete is scheduled to run every second, the execution number 1000 will happen exactly 1000 seconds after the start of the program (± a few milliseconds).

There is no busy waiting. Inside the run_loop method, ischedule calculates the time until the next task becomes pending, and idles the CPU until this happens.

What happens during heavy loading

Heavy loading means that there is not enough computer resources to execute all tasks as scheduled. For example, a task that is scheduled to run every second, could take more than a second to complete. Graceful handling of this condition is essential in a well-implemented periodic scheduler.

• If more than one task become pending simultaneously, they will be executed in the order in which they were added to the schedule by schedule().
• Regardless of the load, no task will be completely starved. All pending tasks will be executed as soon as possible after they become pending.
• There is no build-up of delayed executions. If the execution of a task is delayed so much that the next execution of the same task become pending, an execution will be skipped.

Exceptions

Exceptions during the execution are propagated out of run_loop()/run_pending(), and can be dealt with by the caller.

Cancellable loops

If run_loop() is executed without parameters, it will continue running until the process is terminated.

If the program needs to be able to cancel it, it should supply a stop_event, which is expected to be a threading.Event. When this event is set, run_loop() will cleanly return to the caller after completing the currently pending tasks.

The call to run_loop() accepts an optinal parameter return_after, which allows the loop to return after a specified time, either as seconds or as a datetime.timedelta.

More advanced example

In this example, two tasks are scheduled for periodic execution. The first one is scheduled with an interval of 0.1 seconds, and the second one is scheduled with an interval of 0.5 seconds. The second task takes a lot of time to complete, stress-testing the scheduler.

import time

from ischedule import schedule, run_loop

start_time = time.time()

def task_1():
    dt = time.time() - start_time
    print(f"Started a _fast_ task at t={dt:.3f}")

def task_2():
    dt = time.time() - start_time
    print(f"Started a *slow* task at t={dt:.3f}")

    if dt < 2:
        time.sleep(0.91)
    else:
        time.sleep(0.09)

schedule(task_1, interval=0.1)
schedule(task_2, interval=0.5)

run_loop(return_after=3)
print("Finished")

Output:

Started a _fast_ task at t=0.100
Started a _fast_ task at t=0.200
Started a _fast_ task at t=0.300
Started a _fast_ task at t=0.400
Started a _fast_ task at t=0.500
Started a *slow* task at t=0.500
Started a _fast_ task at t=1.411
Started a *slow* task at t=1.411
Started a _fast_ task at t=2.323
Started a *slow* task at t=2.323
Started a _fast_ task at t=2.413
Started a _fast_ task at t=2.500
Started a *slow* task at t=2.500
Started a _fast_ task at t=2.600
Started a _fast_ task at t=2.700
Started a _fast_ task at t=2.800
Started a _fast_ task at t=2.900
Started a _fast_ task at t=3.000
Started a *slow* task at t=3.000
Finished

The fast task runs every 0.1 seconds, and completes quickly. The slow task is first scheduled for execution at t=0.5s. Initially it uses so much time that it blocks the other tasks from being executed. The scheduler becomes overloaded. It adapts by running the pending tasks as soon as it gets back the control at t=1.41s, and again at t=2.323.

After t=2.0s, the slow task changes to spend only 0.09 seconds. This is slow, but just fast enough not to create delays in the schedule. The scheduler is able to return to normal operation.

Limitations

If the scheduled tasks need to run concurrently on separate threads, then this package cannot be used. Multiprocesseing parallelism is however an excellent alternative in Python. An example implementation of multiprocessing with ischedule is tested as part of every release.

Decorator syntax

Decorator syntax is supported for scheduling tasks:

from ischedule import run_loop, schedule


@schedule(interval=0.1)
def task():
    print("Performing a task")


run_loop(return_after=1)

Timing Precision

Deviations from the scheduled time were thoroughly tested. In a typical 1-minute run, the median deviation is below 0.2 milliseconds, and maximum deviations is below 5 milliseconds. Larger deviations, on the order of tens of milliseconds, have been occasionally observed.

Feedback

The project has its main homepage on GitHub. Issue reports, improvement suggestions or general comments can be submitted at GitHub Issues.