Planned order (low risk → high impact):
-
EventBridge + Step Functions
- Scheduled backfill/replay
- Orchestrated runs with retries/timeouts
- Manual approval + notifications
-
Glue (Catalog + Crawler + Job)
- Catalog tables for Silver Parquet
- Partition management
- Batch compaction / repartition
-
Data Quality (Great Expectations)
- Validate Silver outputs as a gate
- Store validation results in S3 + metrics
-
EMR (or EMR Serverless)
- Large-scale joins/aggregations/historical recompute
- Workloads too heavy for Lambda/Glue alone