Problem

Fixes #59

The endpoint loads ALL table data into memory before sending the response. For large databases (1GB+), this causes:

1. Out-of-memory crashes — entire DB is materialized in a string
2. 30-second timeout — Cloudflare Workers kills long-running requests

Solution

Rewrote to use **** with LIMIT/OFFSET batching:

• Fetches rows in batches of 500 ( constant)
• Streams statements incrementally via
• Yields control between batches (10ms breathing interval) to avoid starving other concurrent requests
• Proper SQL escaping: identifiers with double-quotes, values with correct handling for strings, NULL, blobs, booleans
• Peak memory is now O(BATCH_SIZE) instead of O(database_size)

Changes

• — complete rewrite using streaming
• — updated tests + new tests for NULL handling and multi-batch streaming

Testing

All 25 export tests pass:

Design Decisions

1. Batch size 500 — small enough to fit comfortably in memory, large enough to minimize query round-trips
2. 10ms breathing interval — prevents the Durable Object from being locked during long exports
3. **** — wraps table names in double-quotes to handle special characters safely
4. Streaming via ReadableStream — standard Web API supported by Cloudflare Workers, no extra dependencies