StarRocks at Fresha: Carving Streams into Rock скачать в хорошем качестве

StarRocks at Fresha: Carving Streams into Rock

We sat down with Anton Borisov (Fresha) — Principal Data Architect — and Sida Shen (CelerData) to walk through Fresha’s real-time analytics platform, the first StarRocks production deployment in the UK. It’s running under real operational load today and continues to expand as more real-time use cases move onto the system. In this webinar, Anton and Sida explain why the platform was built, the constraints they were working within (freshness, concurrency, join-heavy queries, and cost), and why StarRocks became the strategic choice that let them move quickly without fragmenting the architecture. The discussion follows the system from early design decisions to day-to-day production behavior: what scaled, what didn’t, and which trade-offs turned out to matter most. -------------------------------------------------------------------------------------------------------------------- Timestamps 00:00 Intro & Agenda 00:58 What Is Fresha? 01:51 Analytics At Fresha: Customer-Facing Dashboards, Real-Time Expectations, And Core Constraints 06:47 First Attempt To Solve The Problem: Scaling With Snowflake + CDC + dbt 08:41 Why Snowflake Fell Short: Freshness, Concurrency, And Cost Under Real-Time Load 10:26 The Scope Of The Problem: Bookings Volume, CDC Scale, QPS, And Join-Heavy Queries 11:29 The New Architecture/Stack To Meet Real-Time SLAs: CDC → Kafka → Flink/Spark → Lake + Serving Systems 14:22 What Is StarRocks And Why It Became The Strategic Choice For Real-Time + Extensible Analytics 16:45 Efficient CBO And Join Performance (Less Need For Denormalization) 17:52 Comfortable Ops And Integration: Multiple Ingestion Options And MySQL Compatibility 19:08 Elastic Scaling With Stateless Compute Nodes And Cache Considerations 21:07 Query Federation Across Lakehouse And Other Systems 21:52 Different Caching Optimizations: Compute Cache + External Metadata Caching 23:26 Why These Features Matter: Balance Freshness, Performance, And Cost Without Rebuilding The Stack 24:27 Data Paths: Ingestion Spine, Historical Lane, Real-Time Lane, And Text Search Lane 26:00 Data Tiers / Freshness Tiers: Hot, Warm, And Deep History (Cost vs Latency Trade-Offs) 27:47 Case Studies - Homepage Analytics: Replacing Postgres-Backed Dashboards With StarRocks For Low-Latency Aggregations 30:16 Results: Tail Latency And Timeouts Eliminated (Up To ~12s/Timeouts → ~300ms) With ~under1 Minute Freshness 33:43 Case Studies - Payment Logs Case: Multi-DB Joins, Why Flink Pre-Work Was Needed, And How They Got Under SLA 36:02 Results: Legacy Report (under 1 minute) Reduced To ~4 seconds (Then Tuned Further) 37:28 Roadmaps: Scaling Adoption (Workload Isolation), Better Streaming Substrate, And Stronger Governance 40:41 Q&A 41:49 Though it is mysql front end, i assume storage is iceberg? 43:22 Are you using flink rocksdb to managed state for transformation? 44:04 How you are superior to apache pinot aka startree? 46:07 What’s your typically monthly cost for using starrocks? 47:11 Have you compared the different ways to connect kafka to starrocks (kafka connect vs flink vs routine/stream load, etc.) 48:26 You said “3k requests per second”: what kind of requests are those (dashboards/frontend or overall)? 49:31 Why did you choose to use paimon as a table type and why not only use iceberg? 53:49 What is the size of your largest table in terms of rows and bytes, and what cluster configuration are you utilising? 54:46 How does elastic scaling affect the query response time and query queues? 56:23 What’s your use case for having historical and realtime lanes? 57:30 Do you see a data size limit for starrocks (multi-tb tables)? 59:05 What are the roadblocks you encountered migrating to starrocks compared to your snowflake setup? 1:02:00 Is starrocks written in rust? --------------------------------------------------------------------------------------------------------------------- Learn more at https://celerdata.com/ Connect with us: LinkedIn:   / celerdata   Twitter:   / celerdata   CelerData Website: https://celerdata.com/ StarRocks GitHub: https://github.com/StarRocks/StarRocks StarRocks Website: https://www.starrocks.io/ Slack: https://starrocks.io/redirecting-to-s... #DataAnalytics #ApachePolaris #DataEngineering #RealTimeAnalytics #RealTimeData #OLAP #DataAnalyst #DataEngineer #DataInfrastructure #databaseprogramming