Data Quality

• Glue Data Quality is serverless and rules are written in DQDL
• The data quality score is the percentage of rules that pass
• Run the same ruleset at rest (catalog) or in transit (ETL job)
• Only the catalog entry point recommends rules for you
• Only the ETL path identifies the exact records that failed
• A failing ETL rule with the fail action stops the job
• Dynamic rules and ML anomaly detection are ETL-only
• Know the core DQDL rule types by what they assert
• Check freshness with a ColumnValues date expression or DataFreshness
• Use with threshold to allow a percentage of rows to match
• Glue Data Quality cannot evaluate nested or list columns
• DataBrew is the no-code path: profile to investigate first
• A DataBrew ruleset fails if any single rule fails
• DataBrew validation raises an EventBridge result event
• Random sampling is the default; stratify to keep rare classes
• Prefer incremental validation over rescanning the whole table
• Data skew shows up as one straggler task, not a quality failure
• Salt the hot key to spread skewed data across partitions
• Adaptive Query Execution splits skewed joins automatically in Glue 4.0+
• Glue Data Quality emits a 'Data Quality Evaluation Results Available' EventBridge event you can filter to FAILED
• Glue Schema Registry compatibility mode controls which schema evolutions are allowed
• CheckSchemaVersionValidity validates schema syntax without registering anything

Unlock with Premium — includes all practice exams and the complete study guide.