https://github.com/Stratio/cassandra-lucene-index

Stratio’s Cassandra Lucene Index

Stratio’s Cassandra Lucene Index, derived from Stratio Cassandra, is a plugin for Apache Cassandra that extends its index functionality to provide near real time search such as ElasticSearch or Solr, including full text search capabilities and free multivariable, geospatial and bitemporal search. It is achieved through an Apache Lucene based implementation of Cassandra secondary indexes, where each node of the cluster indexes its own data. Stratio’s Cassandra indexes are one of the core modules on which Stratio’s BigData platform is based.

Index relevance searches allow you to retrieve the n more relevant results satisfying a search. The coordinator node sends the search to each node in the cluster, each node returns its n best results and then the coordinator combines these partial results and gives you the n best of them, avoiding full scan. You can also base the sorting in a combination of fields.

Any cell in the tables can be indexed, including those in the primary key as well as collections. Wide rows are also supported. You can scan token/key ranges, apply additional CQL3 clauses and page on the filtered results.

Index filtered searches are a powerful help when analyzing the data stored in Cassandra with MapReduce frameworks as Apache Hadoop or, even better, Apache Spark. Adding Lucene filters in the jobs input can dramatically reduce the amount of data to be processed, avoiding full scan.

The following benchmark result can give you an idea about the expected performance when combining Lucene indexes with Spark. We do successive queries requesting from the 1% to 100% of the stored data. We can see a high performance for the index for the queries requesting strongly filtered data. However, the performance decays in less restrictive queries. As the number of records returned by the query increases, we reach a point where the index becomes slower than the full scan. So, the decision to use indexes in your Spark jobs depends on the query selectivity. The trade-off between both approaches depends on the particular use case. Generally, combining Lucene indexes with Spark is recommended for jobs retrieving no more than the 25% of the stored data.

This project is not intended to replace Apache Cassandra denormalized tables, inverted indexes, and/or secondary indexes. It is just a tool to perform some kind of queries which are really hard to be addressed using Apache Cassandra out of the box features, filling the gap between real-time and analytics.

More detailed information is available at Stratio’s Cassandra Lucene Index documentation.

Features

Lucene search technology integration into Cassandra provides:

Stratio’s Cassandra Lucene Index and its integration with Lucene search technology provides:

• Full text search (language-aware analysis, wildcard, fuzzy, regexp)
• Boolean search (and, or, not)
• Sorting by relevance, column value, and distance
• Geospatial indexing (points, lines, polygons and their multiparts)
• Geospatial transformations (bounding box, buffer, centroid, convex hull, union, difference, intersection)
• Geospatial operations (intersects, contains, is within)
• Bitemporal search (valid and transaction time durations)
• CQL complex types (list, set, map, tuple and UDT)
• CQL user defined functions (UDF)
• CQL paging, even with sorted searches
• Columns with TTL
• Third-party CQL-based drivers compatibility
• Spark and Hadoop compatibility

Not yet supported:

• Thrift API
• Legacy compact storage option
• Indexing counter columns
• Indexing static columns
• Other partitioners than Murmur3

Requirements

• Cassandra (identified by the three first numbers of the plugin version)
• Java >= 1.8 (OpenJDK and Sun have been tested)
• Maven >= 3.0