Redshift

A Redshift cluster has one leader node and one or more compute nodes. The leader node receives queries, builds query plans, and coordinates execution. Compute nodes execute the plan in parallel across slices (virtual CPU cores within each node). Data is distributed across slices according to the distribution style.

Distribution style determines how rows are spread across slices. Sort key determines the physical order of data on disk. These two design choices have more impact on Redshift query performance than almost anything else.

Redshift Serverless eliminates cluster management entirely. It automatically scales capacity and charges per RPU-second (Redshift Processing Unit). Provisioned clusters give you more control over node types and pricing.

Redshift can query data directly in S3 without loading it (Spectrum), and can receive near-real-time data from Aurora and DynamoDB via Zero-ETL.

-- COPY data from S3 (most efficient ingestion method)
COPY sales
FROM 's3://my-bucket/sales/2024/'
IAM_ROLE 'arn:aws:iam::123456789012:role/RedshiftS3Role'
FORMAT AS PARQUET;

-- Query S3 data directly with Spectrum
SELECT s.customer_id, SUM(s.amount)
FROM spectrum_schema.cold_sales s
JOIN public.customers c ON s.customer_id = c.id
WHERE s.sale_date >= '2023-01-01'
GROUP BY 1;

-- UNLOAD results to S3
UNLOAD ('SELECT * FROM sales WHERE sale_date = CURRENT_DATE')
TO 's3://my-bucket/daily-export/'
IAM_ROLE 'arn:aws:iam::123456789012:role/RedshiftS3Role'
FORMAT PARQUET PARALLEL ON;

Redshift performance degrades over time without maintenance. The three main operational tasks are WLM tuning (managing query queues), VACUUM (reclaiming space from deletes/updates), and ANALYZE (updating table statistics).