--- title: When your data model is the bottleneck: lessons from... description: Medium's principal software engineer explains how the team rebuilt their feature store data model on ScyllaDB to handle 1M operations per second for real-time... canonical: https://daily.dev/posts/when-your-data-model-is-the-bottleneck-lessons-from-medium-s-feature-store-19cd3kzv5 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: When your data model is the bottleneck: lessons from Medium’s feature store | daily.dev og:description: Medium's principal software engineer explains how the team rebuilt their feature store data model on ScyllaDB to handle 1M operations per second for real-time... og:url: https://daily.dev/posts/when-your-data-model-is-the-bottleneck-lessons-from-medium-s-feature-store-19cd3kzv5 og:image: https://api.daily.dev/og/posts/19cD3kzV5.png og:image:alt: When your data model is the bottleneck: lessons from Medium’s feature store og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # When your data model is the bottleneck: lessons from Medium’s feature store **[The New Stack](https://daily.dev/sources/newstack)** · 10 min read · 1 upvotes · 0 comments ## Summary Medium's principal software engineer explains how the team rebuilt their feature store data model on ScyllaDB to handle 1M operations per second for real-time recommendations. The original relational model split data across two tables, requiring N+1 queries and heavy ALLOW_FILTERING use — with 90% of rows read being discarded. The new list-based model consolidates all items for a given entity into a single partition, enabling single-query reads. Key design decisions include using a clustering key that combines timestamp with an MD5 hash of the value (to handle same-timestamp events), native TTL for automatic storage management, and a local secondary index for value-based deletes. A benchmark comparing ScyllaDB and DynamoDB on production data showed ScyllaDB with significantly more predictable tail latencies: P99 for GetListItems was ~30ms for ScyllaDB vs. up to 220ms for DynamoDB. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [ScyllaDB cut Sprig’s read latency 4X after Redis and ClickHouse hit a wall](https://daily.dev/posts/scylladb-cut-sprig-s-read-latency-4x-after-redis-and-clickhouse-hit-a-wall-ya5cnipcg) · The New Stack · 0 upvotes · 0 comments - [ScyllaDB’s New Cloud Challenges DynamoDB Cost, Performance](https://daily.dev/posts/scylladb-s-new-cloud-challenges-dynamodb-cost-performance-8h1srdkl8) · The New Stack · 8 upvotes · 0 comments - [Lessons Learned From Real-World NoSQL Database Migrations](https://daily.dev/posts/lessons-learned-from-real-world-nosql-database-migrations-xxd0vkngy) · The New Stack · 1 upvotes · 0 comments - [We Compared ScyllaDB and Memcached and… ScyllaDB Lost](https://daily.dev/posts/we-compared-scylladb-and-memcached-and-scylladb-lost-wk3zaixb3) · P99 Conf · 7 upvotes · 0 comments - [DynamoDB crash course: part 2 – data model](https://daily.dev/posts/dynamodb-crash-course-part-2-data-model-67qh0m1qw) · Planet Python · 0 upvotes · 0 comments --- Tags: [#aws-dynamodb](https://daily.dev/tags/aws-dynamodb), [#recommendation-systems](https://daily.dev/tags/recommendation-systems), [#scylladb](https://daily.dev/tags/scylladb) [View this post on daily.dev](https://daily.dev/posts/when-your-data-model-is-the-bottleneck-lessons-from-medium-s-feature-store-19cd3kzv5) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"When your data model is the bottleneck: lessons from Medium’s feature store","url":"https://daily.dev/posts/when-your-data-model-is-the-bottleneck-lessons-from-medium-s-feature-store-19cd3kzv5","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/when-your-data-model-is-the-bottleneck-lessons-from-medium-s-feature-store-19cd3kzv5"},"datePublished":"2026-06-09T14:47:16.752Z","dateModified":"2026-06-09T14:47:44.040Z","description":"Medium's principal software engineer explains how the team rebuilt their feature store data model on ScyllaDB to handle 1M operations per second for real-time...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c9751dd0b8b698d6397fe3d0a66cbb52?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/c9751dd0b8b698d6397fe3d0a66cbb52?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/when-your-data-model-is-the-bottleneck-lessons-from-medium-s-feature-store-19cd3kzv5","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"aws-dynamodb,recommendation-systems,scylladb","timeRequired":"PT10M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"When your data model is the bottleneck: lessons from Medium’s feature store"}]} ```