OLAP on Your Cassandra Data with Arrow, Flight SQL, ADBC, and DuckDB
summary
Demonstrates converting Cassandra SSTable snapshots to Parquet and running OLAP via an Arrow/Flight SQL stack with DuckDB on Graviton, deployed in Kubernetes with TLS/auth and S3 storage. Exports handle tombstones and duplicates by creating views that select the max-timestamp per column, and the prototype reads Parquet directly (local NVMe or S3). Performance claim: a TPC‑H SF1000 query reads about 6 billion rows and returns four result rows in ~10s on an R7gd instance. Built as a proof‑of‑concept: SSTable→Parquet, Flight SQL + ADBC + DuckDB. Concurrency is limited and remains a noted caveat.
AI-written summary · about
chapters (7)
- 0:00Introduction to OLAP
- 4:09Solution architecture
- 6:07Data engine and connectivity
- 8:53Implementation details
- 11:24Live demonstration
- 16:13Getting started and code
- 22:08Q&A session
related talks
40mDuckDB: Harnessing in-process analytics for data science and beyond
10mUsing DuckDBArrowFlight to Power a Feature Store
36mEmbrace the chaos: Composable data systems with fewer asterisks
42mColumnar Storage: Redefining Data Management for the Modern Era
40mDuckDB beyond the notebook
01h 02mAnalytics for not-so-big data with DuckDB
24mLeveraging the Apache Flight Python Client and InfluxDB
22mAccelerating Analytics with Apache Arrow