Skip to content

jwhitbeck/dendrite

master
Switch branches/tags
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
cli
 
 
 
 
 
 
www
 
 
 
 
 
 
 
 
 
 
 
 

Dendrite

Dendrite is a library for querying large datasets on a single host at near-interactive speeds.

It attempts to be:

  • simple: there is no configuration, no services to run, and reads are as simple as opening a file;
  • fast: there are few bottlenecks and reads will usually make good use of all available CPU cores;
  • compact: the file size is typically 30-40% lower than the equivalent compressed JSON;
  • flexible: it supports the same rich set of nested data structures as EDN;
  • write once, read often: optimizations are run at write-time to ensure fast read-time performance.

The current implementation is in Java but only exposes a Clojure API. In the future, I would like to expose a clean Java interface and build a C implementation for non-JVM code.

This code has been in used in production for over a year. It has been successfully used both as a building block in large ETL systems and for ad-hoc data-science studies. However, prior to the 1.0 release, no effort will be made at preserving backwards compatibility of APIs or binary compatibility of files.

Dendrite implements the record shredding and assembly ideas from Google's Dremel paper [1]. Querying for only small parts of the stored records can be up to several orders of magnitude faster than fully deserializing each record and pulling out the desired information. Furthermore, this library also borrows many ideas from the Parquet project, an implementation of the Dremel file format for Hadoop. Unlike Parquet, Dendrite is not tied to any particular ecosystem and is designed to be a small library with no external dependencies.

Status update (March 23, 2018): For personal reasons, I haven't been able to work on this project in the past two years. However, I have been accumulating ideas for the next iteration.

Build Status

Documentation

Work-in-progress documentation and benchmarks are available at dendrite.tech.

Roadmap to 1.0

  • Improve writer performance.
  • Cleanly separate clojure and java code.
  • Expose a good Java API
  • Preserve presence/absence of record keys
  • Add indexing

References

  1. Sergey Melnik, Andrey Gubarev, Jing Jing Long, Geoffrey Romer, Shiva Shivakumar, Matt Tolton, Theo Vassilakis. Dremel: Interactive Analysis of Web-Scale Datasets. In Proc. VLDB, 2010

License

Copyright © 2013-2017 John Whitbeck

Distributed under the Eclipse Public License, the same as Clojure.

About

Dendrite is a library for querying large datasets on a single host at near-interactive speeds.

Resources

License

Stars

Watchers

Forks

Packages

No packages published