SageMaker Catalog Table Data Profile Guide

Quick answer

Use the Data quality tab on a SageMaker Unified Studio catalog table to create DQDL rules, run them on demand, schedule recurring evaluation, inspect pass/fail metrics, and build profile history for anomaly detection.

This is an at-rest workflow. It monitors stored table data over time; it does not intercept data inside a Visual ETL job before a target write.

Prerequisites

Before creating a profile or ruleset, verify:

• the correct project, catalog, database, and table;
• project-role access to the underlying S3 or other data source;
• Lake Formation grants where required;
• an approved schedule and compute/cost owner;
• CloudWatch Logs and result-destination policy;
• KMS, retention, and sensitive-metric requirements;
• a reviewer for failed rules and anomaly observations.

A visible table is not proof that the evaluation role can read every partition or encrypted object.

Create a ruleset

From the table’s Data quality tab, create a named ruleset and write DQDL rules. AWS currently documents 31 rule types in the Unified Studio editor, including common checks for:

• row count;
• completeness and nulls;
• uniqueness;
• required columns;
• column length and values;
• custom SQL;
• referential integrity in supported catalog conditions.

Use names that encode purpose and owner, such as orders-ingest-contract, rather than generic ruleset-1. Store the reviewed DQDL in version control or another governed change system.

Run and interpret results

An evaluation produces:

• execution status;
• overall quality result and score;
• rules passed versus total;
• evaluated metric for each rule;
• failure message;
• run timestamps and CloudWatch log group.

The score is the percentage of configured rules that passed. It is not a universal measure of accuracy or business usefulness. One critical failed rule can matter more than many low-risk passes.

Schedule a useful cadence

Choose a schedule based on table updates and decision latency:

• after each material batch for contract enforcement;
• daily for operational tables;
• weekly for slow-moving reference data;
• separately after backfills or schema migrations.

Avoid a schedule that evaluates stale data repeatedly or misses the only period when bad data reaches consumers. Define late-run, missing-run, and failed-evaluation alerts.

Build profile history safely

Profiles and analyzers create statistics over time that can support learned anomaly observations. Keep cadence stable enough for the chosen mode and record deployments or business events that explain shifts.

AWS says anomaly detection needs at least three data points. A minimum is not a representative baseline. Include ordinary, seasonal, and recovery conditions before using anomalies for consequential automation.

Import existing Glue rulesets

AWS documents adding the tag:

Key: AmazonDataZoneProject
Value: <Unified Studio project ID>

The project role also needs access to the underlying data, including S3 and Lake Formation grants where applicable. Tagging makes the ruleset discoverable; it does not grant the missing data permissions or approve the rules.

Acceptance checklist

1. Run a known-good table and record expected metrics.
2. Inject a safe test failure or use a non-production fixture.
3. Verify rule details, logs, and alert routing.
4. Test a missing partition and late data.
5. Confirm owner response and rerun procedure.
6. Review profile history and anomaly feedback.
7. Document downstream consumers and severity per rule.

Compare profiles with explicit DQDL rules and use the readiness checklist before production.

Frequently asked questions

Where do catalog table data quality checks run?

They run at rest against a table registered in the catalog, either on demand or on a schedule, and results appear on the table’s Data quality tab.

Can an existing AWS Glue ruleset appear in Unified Studio?

Yes. AWS documents tagging an existing ruleset with the AmazonDataZoneProject key and project ID, plus granting the project role access to the table and underlying data.

Does a passing catalog score prove downstream fitness?

No. It covers the configured rules on the evaluated data. Consumers still need freshness, lineage, business semantics, access, and use-specific acceptance checks.

Official sources

• AWS: Data profiling in the catalog
• AWS: Create catalog data quality rulesets
• AWS: Run rulesets and view results

Source check: August 19, 2026. Recheck source support, permissions, Lake Formation, rules, schedules, result storage, encryption, quotas, and costs before use.