Redshift in an analytics architecture

Redshift is rarely deployed alone. It sits in the middle of a pipeline: sources feed into S3 and into processing services, the warehouse holds the curated result, and reporting tools and applications read from it. The diagram below shows the arrangement that most AWS analytics stacks converge on.

flowchart TD subgraph "Data Sources" S3[S3 Data Lake] RDS[Operational DBs/RDS] Stream[Streaming Data] end subgraph "Processing Layer" Glue[AWS Glue ETL] EMR[EMR Processing] Kinesis[Kinesis] end subgraph "Redshift Layer" RS["Redshift (Serverless or provisioned)"] RSSpec[Redshift Spectrum] RSMl[Redshift ML] end subgraph "Consumption Layer" BI[BI Tools] Custom[Custom Apps] Athena[Athena Queries] end S3 --> Glue RDS --> Glue Stream --> Kinesis Kinesis --> S3 Glue --> RS EMR --> S3 S3 --> RSSpec RS --> BI RS --> Custom RSSpec --> Athena classDef source fill:#e1f3d8,stroke:#333,stroke-width:2px; classDef process fill:#dae8fc,stroke:#333,stroke-width:2px; classDef redshift fill:#f9d1ff,stroke:#333,stroke-width:2px; classDef consume fill:#fff2cc,stroke:#333,stroke-width:2px; class S3,RDS,Stream source; class Glue,EMR,Kinesis process; class RS,RSSpec,RSMl redshift; class BI,Custom,Athena consume;

Where zero-ETL integrations cover the source — Aurora, RDS, DynamoDB, streaming services — they replace the Glue step entirely, and data arrives in the warehouse continuously without a pipeline to maintain.

What the warehouse does

Enterprise data warehouse. The central store for curated structured data, holding history over years, optimised for the aggregations that reporting asks for and able to serve many concurrent users.

Query engine. Complex SQL, joins across many tables, window functions and aggregations, and materialized views that pre-compute the expensive parts of a frequently run query.

Integration point. Federated queries reach into Aurora and RDS to join live operational data with warehouse history. Redshift Spectrum reaches into S3 to join warehouse tables against the data lake. Data sharing exposes datasets to other Redshift warehouses without copying them.

Performance layer. Columnar storage, massively parallel execution, result caching and automatic workload management are what let a single query scan a very large table in a time a person will wait for.

Typical workloads

• Business intelligence reporting, through QuickSight or a third-party tool.
• Historical analysis over years of transactional history.
• Complex analytical queries that would lock up an OLTP database.
• Near-real-time dashboards, fed by streaming ingestion or a zero-ETL integration.
• In-warehouse machine learning through Redshift ML, which trains and invokes models from SQL.