Cosmos was driven by the necessity to sort, filter and count large swaths of data queried from an Accumulo instance. Cosmos currently accepts columnar data sets without a guaranteed schema. Because of this, and the support for multiple users to write concurrently to one Accumulo instance using Cosmos, this software scales well given sufficient Accumulo nodes for the concurrent client write threads.

Build the cosmos-core jar using mvn package and place it on the Accumulo classpath in addition to your application. Your client code must then implement the transformation from your data model to the MultimapQueryResult. From here, you can treat your records uniformly using the Cosmos API.

Cosmos supports the followign features:

1. Fetch records by unique identifier
2. Fetch records by column, ascending or descending
3. Fetch records by value in a column
4. Count unique occurrences of a value in a column (SQL groupby)

For each call above, result sets can also be paginated by Cosmos to alleviate large result sets. As such, each operation lazily creates the result set so as to avoid excessive Java heap usage.

Cosmos requires a functioning Accumulo instance and a ZooKeeper instance (can be the same as the ZooKeeper instance used by the Accumulo installation).

In keeping with the Apache Hadoop ecosphere, Cosmos is licensed with the Apache Software License.

Copyright 2013 Josh Elser

Josh Elser -- SRA International