2026-04-06 | 🖼️ Reducing Image Backfill Capacity 🔧
🎯 Summary
🔧 This was a quick surgical change to reduce the image backfill capacity from 4 images per hourly run down to 2.
🤔 Why Reduce the Limit
🖼️ The image backfill pipeline generates AI images for blog posts and library content that lack them. 🕐 Each image generation consumes API quota and takes time. 📉 Halving the per-run capacity from 4 to 2 spreads the work out more evenly over time, reducing burst load on the image generation API and keeping each scheduled run lighter.
🔨 What Changed
📝 Two files were updated in this change.
- 🏗️ In RunScheduled.hs, the bfcMaxImages configuration value was changed from 4 to 2, which directly controls how many images can be generated in a single scheduled run.
- 📋 In the image generation spec, the documented default was updated from 4 to 2 to match the new code behavior.
🧮 Impact
⏱️ Each hourly run will now generate at most 2 images instead of 4. 🐢 This means the full backfill queue will take roughly twice as long to complete, but each individual run finishes faster and uses fewer resources. 🎯 The tradeoff favors steady, predictable throughput over burst capacity.
📚 Book Recommendations
📖 Similar
- Thinking in Systems by Donella H. Meadows is relevant because adjusting rate limits in a pipeline is fundamentally a systems thinking exercise about balancing throughput, resource usage, and stability.
- The Goal by Eliyahu M. Goldratt is relevant because it explores constraint management and throughput optimization in production systems, directly analogous to tuning backfill batch sizes.
↔️ Contrasting
- Antifragile by Nassim Nicholas Taleb offers a contrasting perspective where systems benefit from stress and variability rather than smoothing out load, challenging the assumption that lower burst capacity is always better.
🔗 Related
- Designing Data-Intensive Applications by Martin Kleppmann is relevant because it covers rate limiting, backpressure, and batch processing patterns that inform decisions like tuning image backfill capacity.
- Release It! by Michael T. Nygard is relevant because it discusses capacity planning and stability patterns for production systems, including the kind of throttling decisions made here.