Skip to main content
Windows determine the scope of data that a validation runs against. Datachecks supports two window types: Global and Tumbling.

​
Global windows

A global window runs the validation against the entire dataset on every execution. There is no time-based partitioning — every row in the table is included each time. Best for: Small tables, or when you need a full-dataset view every run. Trade-offs: On large tables, global windows perform a full data load on every polling cycle, which can cause performance degradation and high resource consumption. No additional configuration is required for global windows.

​
Tumbling windows

A tumbling window divides data into fixed, non-overlapping time segments based on a timestamp column. Each segment is validated independently, making it easier to identify issues in a specific batch or time period. Best for: Large tables, continuous monitoring of recent data, identifying temporal anomalies, and batch-specific quality checks.

​
Configuration parameters

ParameterDescription
Date-Time FieldThe timestamp column used to partition data into segments (e.g. created_at, event_time).
Window SizeThe length of each time segment, combined with Window Unit (e.g. 2 days).
Window UnitUnit for Window Size — hours, days, or weeks.
Look Back PeriodHow far back in time to retrieve data, combined with Look Back Unit (e.g. 5 days).
Look Back UnitUnit for Look Back Period — hours, days, or weeks.

​
How it works

Given a Look Back Period of 10 days and a Window Size of 2 days, Datachecks creates 5 non-overlapping windows. Each window is validated independently. Only the data within the Look Back Period is considered — older records are excluded. Example: 
Look Back Period: 10 days
Window Size: 2 days

→ Window 1: days 1–2
→ Window 2: days 3–4
→ Window 3: days 5–6
→ Window 4: days 7–8
→ Window 5: days 9–10