This product is not supported for your selected
Datadog site. (
).
このページは日本語には対応しておりません。随時翻訳に取り組んでいます。
翻訳に関してご質問やご意見ございましたら、
お気軽にご連絡ください。
Overview
Quality Monitoring detects issues such as data freshness delays, unusual data patterns, and changes in column-level metrics before they affect dashboards, machine learning models, or other downstream systems. It alerts you to potential problems and provides context to trace them back to upstream jobs or sources.
Key capabilities
With Quality Monitoring, you can:
- Detect delayed updates and unexpected row count behavior in your tables
- Surface changes in column-level metrics such as null counts or uniqueness
- Set up monitors using static thresholds or historical baselines
- Trace quality issues using lineage views that show upstream jobs and downstream impact
- Explore end-to-end dependencies in Lineage to identify root causes and downstream impact
Supported data sources
Connect to these data warehouses:
Connect to these data lake sources:
Connect to these transactional databases:
Integrate with these transformation and orchestration tools:
Trace lineage from these ELT tools:
Track downstream impact in these BI tools:
Data sent to Datadog
Quality Monitoring runs aggregate queries in your data warehouse, such as COUNT(*), MAX(<timestamp_column>), null counts, and distinct counts. Only the resulting metric values are sent to Datadog, not row-level data from your tables.
Custom SQL monitors are the exception, because you control what the query returns:
- The query must return a value aliased as
dd_value, and Datadog stores that value. Queries that return full rows, such as SELECT *, fail validation and are not stored. - If the query includes a
GROUP BY clause, Datadog also stores the value of each grouped column. Each group is evaluated as a separate series.
For example, the following query stores each distinct region value alongside its count:
SELECT region, COUNT(*) AS dd_value
FROM <database>.<schema>.orders
WHERE status = 'FAILED'
GROUP BY region
If you group by a column that contains sensitive data, such as email or user_id, those values are sent to Datadog. Group only by non-sensitive columns. If your warehouse supports column masking, you can also mask sensitive columns so that Datadog reads masked values.
Column masking support varies by warehouse:
Further reading