Take into account execution times of jobstores, executors and events …

[kbasten]

…when determining next wakeup time

Whenever a jobstore, executor or event takes a while to run, the next wakeup time is calculated incorrectly. This can lead to missed jobs, as in the example below:

12-07-2023 00:30:19 DEBUG Looking for jobs to run
12-07-2023 00:30:19 INFO Running job "job1 (trigger: interval[0:15:00], next run at: 2023-07-12 00:45:19 CEST)" (scheduled at 2023-07-12 00:30:19.283364+02:00)
12-07-2023 00:31:14 DEBUG Next wakeup is due at 2023-07-12 00:31:12.973193+02:00 (in 53.519473 seconds)
12-07-2023 00:31:14 INFO Job "job1 (trigger: interval[0:15:00], next run at: 2023-07-12 00:45:19 CEST)" executed successfully
12-07-2023 00:32:07 DEBUG Looking for jobs to run
12-07-2023 00:32:07 WARNING Run time of job "job2 (trigger: interval[0:05:00], next run at: 2023-07-12 00:36:12 CEST)" was missed by 0:00:54.633316

This is fixed by calculating the next wakeup time using the current datetime, not the datetime at the start of the execution.

[agronholm]

I'm willing to accommodate this change, but I would need at least an additional test that verifies that this fixes the problem.

[kbasten]

Good, I will figure out how to write a test for this soon. Here is an example that misses jobs in the current version:

    def tick():
        print('Tick! The time is: %s' % datetime.now())


    def mk_slow_listener(execution_time_seconds):
        def slow_listener(event):
            print("Start slow event")
            time.sleep(execution_time_seconds)
            print("End slow event")

        return slow_listener


    job_interval = 10
    misfire_grace_time = 1

    logging.getLogger('apscheduler').setLevel(logging.DEBUG)
    scheduler = BackgroundScheduler()

    # add a listener that takes longer than the misfire grace time:
    scheduler.add_listener(mk_slow_listener(misfire_grace_time + 1), EVENT_JOB_SUBMITTED)

    scheduler.add_job(tick, 'interval', seconds=job_interval, misfire_grace_time=misfire_grace_time)
    scheduler.start()

    print('Press Ctrl+{0} to exit'.format('Break' if os.name == 'nt' else 'C'))

    try:
        # This is here to simulate application activity (which keeps the main thread alive).
        while True:
            input(">")
    except (KeyboardInterrupt, SystemExit):
        # Not strictly necessary if daemonic mode is enabled but should be done if possible
        scheduler.shutdown()

[agronholm]

I pushed a single commit with both this fix and a changelog entry where I give you credit. Thanks.