Redshift Serverless: Data Warehousing Without Capacity Planning

Provisioned Redshift clusters require upfront decisions: How many nodes? What instance type? Redshift Serverless eliminates this guesswork. You run queries; AWS handles the infrastructure. In 2025, it’s become the default choice for new Redshift workloads.

Provisioned vs Serverless

Aspect	Provisioned	Serverless
Capacity	Choose nodes upfront	Auto-scales 8-512 RPU
Idle costs	Pays 24/7	Scales to zero after idle
Scaling	Manual resize	Automatic
Best for	Predictable, steady load	Variable, bursty workloads

Creating a Serverless Namespace

import boto3

client = boto3.client('redshift-serverless')

# Step 1: Create namespace (logical database container)
namespace = client.create_namespace(
    namespaceName='analytics',
    adminUsername='admin',
    adminUserPassword='SecurePass123!',
    dbName='warehouse',
    defaultIamRoleArn='arn:aws:iam::123456789012:role/RedshiftRole'
)

# Step 2: Create workgroup (compute layer)
workgroup = client.create_workgroup(
    workgroupName='analytics-compute',
    namespaceName='analytics',
    baseCapacity=32,  # RPUs (8-512)
    maxCapacity=128,  # Auto-scale limit
    publiclyAccessible=False,
    subnetIds=['subnet-abc123', 'subnet-def456'],
    securityGroupIds=['sg-12345678']
)

print(f"Endpoint: {workgroup['workgroup']['endpoint']['address']}")

Query Examples

-- Load data from S3
COPY sales
FROM 's3://my-bucket/sales/'
IAM_ROLE 'arn:aws:iam::123456789012:role/RedshiftRole'
FORMAT PARQUET;

-- Query data lake directly (no COPY needed)
SELECT
    product_category,
    SUM(revenue) as total_revenue,
    COUNT(*) as transactions
FROM spectrum.external_sales  -- S3 data via Spectrum
WHERE sale_date >= CURRENT_DATE - 30
GROUP BY product_category
ORDER BY total_revenue DESC;

-- ML in SQL
SELECT
    customer_id,
    ml_predict_customer_churn(
        customer_tenure,
        monthly_spend,
        support_tickets
    ) as churn_probability
FROM customers;

Cost Optimization

RPU Pricing (December 2025)

Base: $0.36/RPU-hour (on-demand)
Compute: Charged only when queries run
Storage: $0.024/GB-month (managed storage)

Example: 32 RPU base, 4 hours active/day = ~$1,400/month

Equivalent provisioned: dc2.large 4-node = ~$2,200/month (running 24/7)

2025 Features

AI-driven scaling: Predicts query complexity, pre-scales
Cross-account data sharing: Share tables without copying
Streaming ingestion: Real-time from Kinesis/MSK
Zero-ETL: Query Aurora/RDS without data movement

Redshift Serverless is the future of AWS data warehousing. Provisioned clusters still have their place for predictable workloads, but serverless is where the innovation is happening.