How Medium Powers Real-Time Recommendations at 1M OPS
Episode 12 min 8s

How Medium Powers Real-Time Recommendations at 1M OPS

This story was originally published on HackerNoon at: https://hackernoon.com/how-medium-powers-real-time-recommendations-at-1m-ops.
Discover how Medium redesigned its ScyllaDB feature store with list-based features to reduce queries, improve latency, & support up to 1M operations per second.
Check more stories related to undefined at: https://hackernoon.com/c/undefined. You can also check exclusive content about #scylladb-data-modeling, #scylladb-list-features, #scylladb-feature-store, #scylladb-ttl, #machine-learning-feature-store, #scylladb-vs-dynamodb-benchmark, #medium-scylladb-feature-store, #good-company, and more.

This story was written by: @scylladb. Learn more about this writer by checking @scylladb's about page, and for more stories, please visit hackernoon.com.

Medium redesigned its ScyllaDB-based feature store to support the high-throughput workloads behind its recommendation system. Its original relational feature model required separate lookups and relied heavily on ALLOW FILTERING, causing unnecessary reads and query overhead. The new list-based model stores related items in a single partition, enabling one-query retrieval, timestamp-based ordering, and automatic TTL expiration. Benchmarks showed ScyllaDB delivering lower and more predictable tail latencies than DynamoDB for Medium’s tested workloads, although the comparison did not include an additional DynamoDB caching layer.

Community discussion