Case Studyscylladbfeature storedragonflydbcompaction tuning
Agoda Scales Feature Store To Maintain 10ms P99 Latency
9.2
Relevance Score
Agoda scaled its ScyllaDB-backed online feature store after server traffic spiked 50× between Jan 2023 and Feb 2025, aiming to maintain a 10ms P99 latency SLA. Engineers used DragonflyDB caching, client-side caching, compaction tuning, and a SATA→NVMe disk upgrade; combined changes increased cold-cache capacity from about 5K EPS to roughly 300K EPS, restoring stability.


