HTAP with Azure Cosmos DB: Hybrid Transaction & Analytical Processing
summary
Claims a cut in total cost of ownership by keeping analytics inside Cosmos DB while leaving OLTP latencies unchanged, implemented as a decoupled log‑structured segmented store that writes parquet base format files (~300–400 MB targets), generates invalidation files to track old versions and provide snapshot isolation, flushes increments about every 30 seconds, compacts segments when >50% invalid, physically copies files for geo add‑region, and ships today with point‑in‑time and partitioning previews.
AI-written summary · about
chapters (9)
- 0:00Start
- 2:35Goals
- 7:06Log structured storage
- 20:37Invalidations
- 24:18Freshness optimizations
- 25:39Garbage Collection
- 30:30Partition Splits
- 36:58Roadmap
- 39:33Learnings
related talks
29mHTAP vs OLTP/OLAP
49mAnalytics in SQL Server and Azure SQL Database
01h 01mHTAP Workloads: Challenges and Solutions
01h 13mHTAP made live — from the engineering perspective
01h 02mAzure Cosmos DB - Low Latency and High Availability at Planet Scale
28mArchitecture patterns of Azure Cosmos DB
14mRetrofitting High Availability Mechanism to Tame Hybrid Transaction/Analytical Processing
18mIntelligent Resource Optimization: Mastering Azure Cosmos DB for Enterprise Workloads