Schedule flow runs
Prefect can schedule when to automatically create new flow runs.
Prefect allows you to specify schedules on which your flows run.
You can add one or more schedules to any deployed flow.
Schedules tell Prefect when and how to create new flow runs.
You can add a schedule to a deployed flow in the Prefect UI, in the CLI through the prefect deployment schedule
command, or the prefect.yaml
configuration file.
Schedule types
Prefect supports three types of schedules:
Cron
is most appropriate for users who are already familiar withcron
from previous use.Interval
is best suited for deployments that run at some consistent cadence that isn’t related to absolute time.RRule
is best suited for deployments that rely on calendar logic for simple recurring schedules, irregular intervals, exclusions, or day-of-month adjustments.
Schedules can be inactive
When you create or edit a schedule, you can set the active
property to False
in Python (or false
in a YAML file) to
deactivate the schedule.
This is useful to keep the schedule configuration but temporarily stop the schedule from creating new flow runs.
Cron
A Cron
schedule uses a standard cron
string to define when flow runs should occur. This is ideal for schedules tied to specific clock times and dates (e.g., “every Monday at 9:00 AM”). You may also provide a timezone to enforce daylight saving time (DST) behaviors.
Prefect’s cron scheduling is based on the croniter
library, allowing it to interpret familiar cron expressions. For example, the expression 0 9 * * MON-FRI
means “at 9:00 AM on every Monday through Friday”.
Supported croniter
features
While Prefect supports most features of croniter
for creating cron
-like schedules, it does not support “R”
random or “H” hashed keyword expressions or the schedule jittering possible with those expressions.
Cron
properties include:
Property | Description |
---|---|
cron | A valid cron string. (Required) |
day_or | Boolean indicating how croniter handles day and day_of_week entries. Default is True . |
timezone | String name of a time zone. (See the IANA Time Zone Database for valid time zones.) |
How the day_or
property works
The day_or
property defaults to True
, matching the behavior of cron
.
In this mode, if you specify a day
(of the month)
entry and a day_of_week
entry, the schedule runs a flow on both the specified day of the month and on the specified day of the
week.
The “or” in day_or
refers to the two entries treated like an OR
statement. The schedule should include
both, as in the SQL statement:
For example, with day_or
set to True
, the cron schedule * * 3 1 2
runs a flow every minute on the third day of the month
and on Tuesday (the second day of the week) in January (the first month of the year).
With day_or
set to False
, the day
(of the month) and day_of_week
entries are joined with the more restrictive AND
operation, as in the SQL statement:
For example, the same schedule, when day_or
is False
, runs a flow on every minute on the third Tuesday in January.
This behavior matches fcron
instead of cron
.
Daylight saving time considerations
If the timezone
is a DST-observing one, then the schedule adjusts itself appropriately.
The cron
rules for DST are based on schedule times, not intervals. This means that an hourly cron
schedule fires on
every new schedule hour, not every elapsed hour.
For example, when clocks are set back, this results in a two-hour pause as the schedule will fire the first time 1 AM is reached and the first time 2 AM is reached, 120 minutes later.
Longer schedules, such as one that fires at 9 AM every morning, adjust for DST automatically.
Interval
As the name suggests, an Interval
schedule is best for deployments that need to run at a consistent cadence not strictly tied to absolute clock time (e.g., “every 10 minutes” or “every 3 hours”). It creates new flow runs on a regular interval, measured in seconds, which can be anchored to a specific anchor_date
.
Interval
properties include:
Property | Description |
---|---|
interval | datetime.timedelta indicating the time between flow runs. (Required) |
anchor_date | datetime.datetime indicating the starting or “anchor” date to begin the schedule. If no anchor_date is supplied, the current UTC time is used. |
timezone | String name of a time zone, used to enforce localization behaviors like DST boundaries. (See the IANA Time Zone Database for valid time zones.) |
The anchor_date
does not indicate a “start time” for the schedule, but a fixed point to
compute intervals.
If the anchor date is in the future, then schedule dates are computed by subtracting the interval
from it.
Daylight saving time considerations
If the schedule’s anchor_date
or timezone
are provided with a DST-observing timezone, then the schedule adjusts itself
appropriately.
Intervals greater than 24 hours will follow DST conventions, while intervals of less than 24 hours will follow UTC intervals.
For example, an hourly schedule will fire every UTC hour, even across DST boundaries. When clocks are set back, this results in two runs that appear to be scheduled for 1 AM local time, even though they are an hour apart in UTC time.
For longer intervals, like a daily schedule, the interval schedule adjusts for DST boundaries so that the clock-hour remains constant. This means that a daily schedule that always fires at 9 AM will observe DST and continue to fire at 9 AM in the local time zone.
RRule
For complex scheduling needs that go beyond simple cron expressions or fixed intervals, RRule
(Recurrence Rule) is the most powerful option. It’s ideal for deployments that rely on calendar-based logic, such as “the last weekday of each month,” “every other Tuesday,” or schedules with specific exclusions.
RRule
scheduling supports iCal recurrence rules, providing a rich syntax for defining a wide array of repetitive schedules, from yearly down to every minute.
RRules are appropriate for any kind of calendar-date manipulation, including simple repetition, irregular intervals, exclusions, week day or day-of-month adjustments, and more. RRules can represent complex logic like:
- The last weekday of each month
- The fourth Thursday of November
- Every other day of the week
RRule
properties include:
Property | Description |
---|---|
rrule | String representation of an RRule schedule. See the rrulestr examples for syntax. |
timezone | String name of a time zone. See the IANA Time Zone Database for valid time zones. |
You may find it useful to use an RRule string generator such as the iCalendar.org RRule Tool to help create valid RRules.
RRule restrictions
The max supported character length of an rrulestr
is 6,500 characters
COUNT
is not supported. Please use UNTIL
or the /deployments/{id}/runs
endpoint to schedule a fixed number of
flow runs.
Daylight saving time considerations
As a calendar-oriented standard, RRules
are sensitive to the initial timezone provided.
A 9 AM daily schedule with a DST-aware start date maintains a local 9 AM time through DST boundaries. A 9 AM daily schedule
with a UTC start date maintains a 9 AM UTC time.
How scheduling works
Prefect’s Scheduler
service evaluates each deployment’s schedules and creates new runs appropriately. It starts
automatically when prefect server start
is run and it is a built-in service of Prefect Cloud.
The Scheduler
creates the fewest runs that satisfy the following constraints, in order:
- No more than 100 runs will be scheduled.
- Runs will not be scheduled more than 100 days in the future.
- At least three runs will be scheduled.
- Runs will be scheduled until at least one hour in the future.
These behaviors are adjustable through the relevant settings with the prefect config view --show-defaults
command:
See the Settings docs for more information on altering your settings.
These settings mean that if a deployment has an hourly schedule, the default settings will create runs for the next four days (or 100 hours). If it has a weekly schedule, the default settings will maintain the next 14 runs (up to 100 days in the future).
Scheduler
does not affect execution
The Prefect Scheduler
service only creates new flow runs and places them in Scheduled
states.
It is not involved in flow or task execution.
If you change a schedule, previously scheduled flow runs that have not started are removed, and new scheduled flow runs are created to reflect the new schedule.
To remove all scheduled runs for a flow deployment, you can remove the schedule.