SASI, Cassandra on the full text search ride!
summary
SASI adds full text search to Cassandra by creating an SSTable Attached Secondary Index with custom on-disk structures (modified suffix trees) so queries do an index lookup then fetch rows, not Lucene scoring. Demo uses a 100,000-album dataset and shows substring, prefix and stemmed token searches via LIKE %term% and analyzers; it integrates with Spark predicate-pushdown for local index lookups. Benchmarks on 13 bare machines with billions of rows show reasonable latencies but no global ordering or scoring; use Cassandra 3.5.
AI-written summary · about
chapters (9)
- 0:00Introduction and speaker background
- 1:52What is SASI and its origins
- 3:21Demo: Creating full text indexes
- 8:07How distributed secondary indexes work
- 14:56Implementation details and lifecycle
- 17:51Query planner and predicate optimization
- 20:23Benchmark results and performance
- 24:59Availability and future enhancements
- 27:08Production use and Q&A
related talks
28mSponsored Session: AI-Powered Semantic Search with Cassandra: Today and Tomorrow
29mFast and Flexible JSON Retrieval in Cassandra Using SAI
31mStorage Attached Indexes for Cassandra (CEP-7)
01h 01mOK S3
33mMaking Cassandra Easier for AI Developers
11mCassandra Vector Search Performance Tuning
29mCassandra - an introduction
36mSearch on the fly: how to lighten your Big Data