Task Scheduling¶

In the past you might have written a cron entry for each recurring task on your server. This quickly gets painful — the schedule lives outside source control and each change means SSHing in. arvel's command scheduler lets you define your recurring tasks in code, and needs only a single cron entry on the server.

Defining schedules¶

Define scheduled tasks in app/routes/console.py using the Schedule entry point:

# app/routes/console.py
from arvel import Schedule

Schedule.command("queue:work --stop-when-empty").hourly()
Schedule.call(prune_old_exports).daily_at("02:00")

You can schedule three kinds of work:

Schedule.call(callback)               # an async/sync callable
Schedule.command("cache:clear")       # an arvel console command by signature
Schedule.job(SendDigest())            # a queued job (see Queues)

Frequency options¶

Chain a frequency onto any scheduled task:

Method	Runs the task...
`.every_minute()`	every minute
`.hourly()`	every hour
`.daily()`	every day at midnight
`.daily_at("13:00")`	every day at a given time
`.twice_daily(1, 13)`	at two hours each day
`.weekly()` / `.monthly()` / `.quarterly()` / `.yearly()`	at the start of the period
`.weekdays()` / `.weekends()`	constrain to those days
`.cron("/5 * * *")`	a raw cron expression

Constraints compose — .daily().weekdays().between("9:00", "17:00") runs at midnight only on a weekday within business hours. Set the evaluation timezone with .timezone("America/New_York").

Truth-test & environment constraints¶

Schedule.command("report:send").daily().when(lambda: billing_enabled())
Schedule.command("report:send").daily().skip(lambda: holiday_today())
Schedule.command("report:send").daily().environments("production")

Preventing task overlaps¶

By default a task runs even if its previous run is still going. Guard long tasks so they never overlap:

Schedule.command("report:generate").hourly().without_overlapping()

On a multi-server deployment, ensure a task runs on one server only:

Schedule.command("report:generate").daily().on_one_server()

Both use a cache lock, so they need a shared cache store.

Task hooks¶

Schedule.command("backup:run").daily() \
    .before(notify_start) \
    .after(notify_done) \
    .on_success(record_ok) \
    .on_failure(alert)

Running the scheduler¶

The scheduler itself is driven by one cron entry that invokes schedule:run every minute; it works out which tasks are actually due:

* * * * * cd /path-to-app && arvel schedule:run >> /dev/null 2>&1