Oracle Cloud Infrastructure Documentation

Scheduler

The Data Science Scheduler can be used to periodically trigger other Data Science resources, for example, ML jobs, ML pipelines, and ML monitoring.

For interval-based schedules, you can configure a randomized start window to distribute schedule start times across a specified time range. Provide a start time and a random window duration. When the schedule becomes active, the scheduler chooses a one-time random offset within that window to set the first execution time. The scheduler then runs next executions at fixed intervals, based on the resolved first execution time and the interval you configure. You can preview the upcoming execution windows before you create or update the schedule.

You can:

  • Run a model training pipeline automatically once a week, every week.
  • Run an ML monitoring job every day.
  • Run a batch inference job every four hours.

For more information about Data Science resources that can be scheduled, see:

Terminology

start date and start time
The start date and time is when the randomization window begins. It is the earliest possible time for the randomized execution window.

Example:

If the start date and time is 09:10 UTC and the random window duration is 30 minutes, the first execution can occur any time between 09:10 UTC and 09:40 UTC.

Frequency
Used to describe the unit of time defined by an interval value, for example, days or hours.
Interval
A value.
Next five executions

A table that lists the next five upcoming schedule occurrences and their corresponding randomization windows.

Use this table to verify the schedule timing before you create or update the schedule.

Randomization window
A randomization window is a configured time range during which the scheduler can start an interval schedule execution at a random time. The randomization window begins at the start date and time and lasts for the configured random window duration.

Example:

If the start date and time is 09:10 UTC and the random window duration is 30 minutes, the randomization window is 09:10 UTC to 09:40 UTC. The scheduler selects the actual execution time from within that window.
Random window duration
Random window duration is the number of minutes after the start time during which the scheduler can randomly select the first execution time. The minimum value is 30 minutes. The value cannot exceed the configured schedule interval. If random start is enabled and you do not provide a random window duration, the service uses a default duration of half the configured interval.

Example:

If the random window duration is 30 minutes, the schedule can start at any random time during the 30-minute period that begins at the start date and time.

Summary
A summary is a generated sentence that describes the effective schedule configuration in readable form.

Example:

First run starts randomly within a 30-minute window; subsequent runs repeat daily from the resolved first run time.

Target Resource
The OCI resource the scheduler is configured to trigger.
Target Resource Type
The type of OCI resource the scheduler is configured to trigger, for example, ML Job or ML Pipeline Trigger.

Key Features

You can use the scheduler with the following Data Science resources:

  • Job Run
  • Pipeline Run

You can define a schedule by using:

  • A form interface for daily or hourly frequencies.
    • Randomized start windows for interval schedules
  • An iCal Expression.
  • A Cron Expression.