Scheduled mode runs monitors automatically at a fixed frequency. On-Demand mode runs monitors manually when you choose.
Every monitor runs in one of two modes: Scheduled or On-Demand. You select the mode when creating a monitor, and it cannot be changed after creation.
Scheduled monitors run automatically at the frequency you configure — hourly, daily, weekly, or custom intervals. Decube scans the asset at each interval, evaluates the monitored metric against the threshold or confidence interval, and opens an incident if the result falls outside the expected range.
Scheduled mode supports Smart Training. When Smart Training is enabled, the ML model trains on historical data to learn a dynamic confidence interval. See How Anomaly Detection Works for how training works.
All monitor types support Scheduled mode — Freshness, Volume, Field Health (including Cardinality), Custom SQL, Schema Drift, Job Failure.
On-Demand monitors do not run on a schedule. You trigger them manually from All Monitors, either immediately after creation (Save and Run) or at any later point (Run once from the ellipsis menu).
On-Demand mode is useful for ad-hoc investigation, validating a specific data load, or testing a monitor's configuration before committing to a schedule.
Constraints in On-Demand mode:
Smart Training is not available. Thresholds must be set manually.
Grouped By is not available.
Frequency is not configured — instead, you set a Lookback Period that defines how far back each run scans.
The Cardinality test type is not available.
Supported monitor types: Freshness, Volume, Field Health (excluding Cardinality), Custom SQL.
Runs automatically
Yes
No — manual trigger only
Smart Training
Yes (with Timestamp row creation)
No
Grouped By
Yes
No
Cardinality
Yes
No
Frequency setting
Required
Not applicable
Lookback Period
Set automatically from frequency
Required
Use Scheduled for production monitoring — tables that need continuous coverage, SLA tracking, or anomaly detection based on historical patterns.
Use On-Demand for development, investigation, or one-off validation checks where you don't need continuous alerting.
Last updated
