Leela

Introduction

Leela started as a time-series engine, very similar to graphite ¹ but with less features, tailored for our needs. We found it hard to scale graphite at the time and figure that using cassandra would make it a lot easier to scale up and cope with our load. Back then we were collecting mostly server metrics like cpu, memory and disk.

After a while we figure we also wanted to store more information about our datacenter. Like the relationships a machine or switch may have. For instance, in what switch a given network card is connected to.

Leela then evolved into a property-graph engine. It continue to be a time-series storage but we now could create vertexes and connects them to each other.

This allowed us to store complex data, like the relationships we have in our datacenter:

      o cpu-usage [time-series]
      |  
      |  o hwinfo [json-data]
      |  |
   +---------+     +-----+            +--------+
   | machine |-----| nic |----------->| switch |
   +---------+     +-----+            +--------+
                                          |
           +------------+    +-----+      |
           | datacenter |<---| rak |<-----+
           +------------+    +-----+

It is an ever-evolving system and we are continuously working on it. New features are planned and we are working hard to improve the documentation as fast as we can.

The Documentation

This section describes how the documentation has been organized. There are three major sections, Development, Operations and Users. Each section covers a given aspect of the system from that perspective.

Orthogonal to those sections there are the various modules that comprise Leela.

client libraries	backend modules	frontend
libleela	warpgrep	warpserver
libleela-ruby	blackbox	webleela
libleela-python

All client libraries are built on top of libleela. Libleela is written in C and the other languages creates a module on top of that library.

Blackbox and warpgrep are internal modules, the user usually don't need to interact with them. They are used by warpserver. The former provides access to the storage engine whereas the later allows users to monitor queries in real-time.

Warpserver and webleela are the frontends. They provide a ZMQ interface and an HTTP interface, respectively.

Development

The audience is software developers of ones interested in contributing or learning about Leela internals.

Environment

Operations

The audience is system administrators or ones interested in installing, maintaining and tuning a Leela cluster.

User Guide

For the ones interested in using a Leela cluster.

http://graphite.readthedocs.org ↩