DynamoDB CDC Benchmark Results

Benchmark suite: internal/impl/aws/dynamodb/bench/

2026-03 — Initial Benchmark

PR: https://github.com/redpanda-data/connect/pull/4102

Environment:

• DynamoDB Local running in Docker (amazon/dynamodb-local:latest)
• In-memory mode (-sharedDb -inMemory)
• Local development machine

Dataset: 450,000 items (3 tables x 150,000 items)

• bench-users: 150K items with user profile data (~1KB each)
• bench-products: 150K items with product catalog data (~1KB each)
• bench-orders: 150K items with order data (~1KB each)

Configuration:

• batch_size: 1000
• 3 tables monitored via DynamoDB Streams

Results:

INFO rolling stats: 99000 msg/sec, 204 MB/sec
INFO rolling stats: 95516 msg/sec, 198 MB/sec
INFO rolling stats: 102000 msg/sec, 216 MB/sec

Observations:

• DynamoDB Local uses a single shard per table — with 3 tables the connector fully saturates each shard
• After all records are consumed, throughput drops to 0 until new writes arrive
• Real AWS DynamoDB scales horizontally with multiple shards per table, so production throughput could be higher
• DynamoDB Streams retain records for only 24 hours — seed data and run the benchmark promptly