Amazon Redshift Cheat Sheet

Updated 2026-04-12

🧠Study flashcards on this topic97 cards · spaced repetition→

Amazon Redshift is a fully managed, petabyte-scale cloud data warehouse service developed by AWS that enables fast, cost-effective analytics on massive datasets using columnar storage and massively parallel processing (MPP) architecture. Designed for OLAP workloads, it seamlessly integrates with the broader AWS ecosystem—querying data lakes via Spectrum, streaming from Kinesis, and federating to RDS—while offering both provisioned clusters for predictable workloads and serverless for on-demand elasticity. A critical mental model: Redshift distributes data across compute nodes using distribution keys and sorts it using sort keys; choosing these wisely is the single most impactful optimization you can make, as poor choices force costly data shuffles across the network during query execution.

What This Cheat Sheet Covers

This topic spans 15 focused tables and 104 indexed concepts, 97 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Cluster Architecture and Node TypesTable 2: Distribution StylesTable 3: Sort KeysTable 4: Data Loading and UnloadingTable 5: Redshift Spectrum and External TablesTable 6: Materialized ViewsTable 7: Performance Tuning TechniquesTable 8: Workload Management (WLM)Table 9: Concurrency and Query OptimizationTable 10: VACUUM and ANALYZE MaintenanceTable 11: Compression EncodingsTable 12: Security FeaturesTable 13: System Tables and MonitoringTable 14: Redshift Serverless vs ProvisionedTable 15: Advanced Features

Quick IndexSubscribe to unlock

A jump-to index of every table row in this cheat sheet.

Mind MapSubscribe to unlock

An interactive map of every table and concept in this topic.

Table 1: Cluster Architecture and Node Types

Every Redshift query flows through the same two-tier shape: a leader node that plans and coordinates, and compute nodes—each carved into slices—that do the parallel work. Knowing what each piece does, and where RA3's separation of compute from managed storage fits versus the older DC2 nodes, is the foundation for sizing a cluster and reasoning about why a query is fast or slow.

Component	Example	Description
Leader node	`Cluster with 1 leader + N compute nodes`	Receives queries from client applications, parses SQL, develops execution plans, and coordinates parallel query execution across compute nodes.
Compute node	`dc2.large`, `ra3.xlplus`	• Executes query portions assigned by the leader node • each has dedicated CPU, memory, and disk storage • stores data and performs computations.
Node slice	`16 slices on ra3.4xlarge`	• A partition of a compute node's memory and disk space • each slice processes a portion of the workload in parallel • the unit of data distribution.
RA3 nodes	`ra3.xlplus`, `ra3.4xlarge`	• Modern node type with managed storage that scales compute and storage independently • uses high-performance SSDs for hot data and S3 for cold data • recommended for most workloads.
DC2 nodes (legacy)	`dc2.large`, `dc2.8xlarge`	• Dense compute nodes with local SSD storage tightly coupled to compute • lower cost per GB but require cluster resize for storage expansion • reaching end of life.
Single-node cluster	`1 node acting as both leader and compute`	• One node serves as both leader and compute node • suitable for development, testing, or small datasets under 160 GB.
Multi-node cluster	`1 leader node + 2+ compute nodes`	• Separates leader and compute roles • leader node is free • scales horizontally by adding more compute nodes for larger datasets and higher concurrency.
AQUA (Advanced Query Accelerator)	`Enabled on RA3.xlplus nodes`	• Hardware-accelerated distributed cache that accelerates scans, aggregates, and filtering operations by up to 10x • automatically enabled on supported RA3 nodes.

Table 2: Distribution Styles

How rows are spread across slices decides whether a join happens locally or forces data to fly across the network. KEY collocates matching rows, EVEN spreads everything round-robin, ALL replicates small tables everywhere, and AUTO lets Redshift switch styles as a table grows—pick wrong and you pay for it on every join.

Style	Example	Description
KEY	`DISTKEY(customer_id)`	• Distributes rows based on values in a specified column • rows with the same key value land on the same node slice, enabling collocated joins without network transfers.

The rest of this section is part of the full cheat sheet

Table 3: Sort Keys

Sort keys physically order rows on disk so Redshift can skip whole blocks that can't match a filter—the secret behind fast range scans. The choice comes down to your query shape: compound keys reward a consistent leading-column filter, interleaved keys give every column equal weight, and the AUTO/MDDL options let Redshift learn the right layout from your actual workload.

Type	Example	Description
COMPOUND sort key	`SORTKEY(date, region, product)`	• Sorts rows sequentially by each column in order • highly effective when queries filter on a consistent prefix of columns (e.g., always filtering by date, then region).

The rest of this section is part of the full cheat sheet

Table 4: Data Loading and Unloading

Getting data in and out of Redshift efficiently means leaning on parallelism. COPY is the workhorse that loads from S3 across every slice at once, with manifests for precise file control and auto-copy jobs for continuous ingestion, while UNLOAD exports results back to S3 in parallel. The cardinal rule the rows make clear: avoid row-by-row INSERT for anything but trivial data.

Command	Example	Description
COPY	`COPY sales FROM 's3://bucket/data/'` `IAM_ROLE 'arn:...'` `FORMAT AS PARQUET;`	• Loads data in parallel from S3, DynamoDB, or remote hosts using MPP architecture • the most efficient loading method supporting Parquet, ORC, JSON, CSV, and Avro formats.

The rest of this section is part of the full cheat sheet

Table 1: Cluster Architecture and Node Types

Component	Example	Description
Leader node	`Cluster with 1 leader + N compute nodes`	Receives queries from client applications, parses SQL, develops execution plans, and coordinates parallel query execution across compute nodes.
Compute node	`dc2.large`, `ra3.xlplus`	• Executes query portions assigned by the leader node • each has dedicated CPU, memory, and disk storage • stores data and performs computations.
Node slice	`16 slices on ra3.4xlarge`	• A partition of a compute node's memory and disk space • each slice processes a portion of the workload in parallel • the unit of data distribution.
RA3 nodes	`ra3.xlplus`, `ra3.4xlarge`	• Modern node type with managed storage that scales compute and storage independently • uses high-performance SSDs for hot data and S3 for cold data • recommended for most workloads.
DC2 nodes (legacy)	`dc2.large`, `dc2.8xlarge`	• Dense compute nodes with local SSD storage tightly coupled to compute • lower cost per GB but require cluster resize for storage expansion • reaching end of life.
Single-node cluster	`1 node acting as both leader and compute`	• One node serves as both leader and compute node • suitable for development, testing, or small datasets under 160 GB.
Multi-node cluster	`1 leader node + 2+ compute nodes`	• Separates leader and compute roles • leader node is free • scales horizontally by adding more compute nodes for larger datasets and higher concurrency.
AQUA (Advanced Query Accelerator)	`Enabled on RA3.xlplus nodes`	• Hardware-accelerated distributed cache that accelerates scans, aggregates, and filtering operations by up to 10x • automatically enabled on supported RA3 nodes.

Table 2: Distribution Styles

Style	Example	Description
KEY	`DISTKEY(customer_id)`	• Distributes rows based on values in a specified column • rows with the same key value land on the same node slice, enabling collocated joins without network transfers.

Table 3: Sort Keys

Type	Example	Description
COMPOUND sort key	`SORTKEY(date, region, product)`	• Sorts rows sequentially by each column in order • highly effective when queries filter on a consistent prefix of columns (e.g., always filtering by date, then region).

Table 4: Data Loading and Unloading

Command	Example	Description
COPY	`COPY sales FROM 's3://bucket/data/'` `IAM_ROLE 'arn:...'` `FORMAT AS PARQUET;`	• Loads data in parallel from S3, DynamoDB, or remote hosts using MPP architecture • the most efficient loading method supporting Parquet, ORC, JSON, CSV, and Avro formats.