How Databricks Feature Store Serves Features With Sub-Second Freshness
Databricks, Monday, August 17th, 2026
Databricks Feature Store now reaches 200ms p99 latency for streaming features using Spark Real-Time Mode.
Databricks describes how its Feature Store now delivers real-time freshness for machine learning features. Streaming aggregations from Kafka can reach the online feature store with 200ms p99 latency, collapsing feature lag from minutes or hours to milliseconds.
Spark Real-Time Mode makes millisecond feature computation possible by processing rows continuously instead of waiting for microbatches. RTM updates rolling-window aggregates per event rather than per batch. It also amortizes checkpointing to keep stateful streaming performant at that cadence.