strum

Autotune

Introduction

Autotune is a Cloud and Container aware diagnostics that uses Machine Learning to help figure out issues with your infrastructure.

Quick Start

Download Autotune

Autotune outputs its data in JSON to the command line. Run the following command:

sudo autotune output

Autotune primarily runs as a daemon which regularily pushes diagnostics to a central server. Acksin runs Autotune Cloud providing this capability. You can get the configuration on the Acksin Console or you can check out config.json.template for agent configuration. We will open source the server side in the future.

Run the following:

sudo autotune agent config.json

Getting Started & Documentation

All documentation is on the Autotune website.

Developing Autotune

Autotune's command line portion is primarily written in Go whereas the Machine Learning is written in Python. We will go over how to code for each part.

The code is split into a couple different sections:

• CLI Tool: Collects stats from the System and Containers.
• Mental Models: Take System stats and creates models for AI. AWS Lambda code.
• Tensorflow AI: Uses Mental Models to generate train AI for various tasks.
• Console: ReactJS Frontend App used on STRUM Cloud
• Server: This component is not yet open sourced.

Primary Dependencies

One of the primary dependencies of Autotune is the ProcFS Library we use. Any code that needs to read from ProcFS should go there and we will primarily code there for things. In the future we will have similar dependencies for SysFS. In addition to that we use the Go libraries provided by the Cloud providers.

Goals

Autotune is based around the work of John Boyd with his Observe Orient Decide and Act paradigm, also called the OODA loop.

UNIX has traditionally been very much about composition of tools which works exceedingly well when you have a single machine with multiple services. However, we are now in the era of Linux as applicance. Linux is now just a single layer with one or two apps being the main users of the operating system. Furthermore, as we go up the stack with Containers and maintain clusters instead of individual machines we need to know how one service affects the others.

Situational awareness

Autotune's goal is to be situationally aware about the Containers, the System, the Cluster and the Cloud so that it can help you make effective decisions. By keeping track of this various information about the cluster we can help point you to potential issues. Autotune is not trying to replace application level inautotuneentation such as Graphite and Datadog. Our tools are at the system level.

Mental Models

Mental Models are how a system and a cluster should behave such that there is minimal operational issues. Mental Models are kernel changes as well as various feature columns that are used to train the Machine Learning Algorithms.

This data is contained in the ai/mental_models directory.

License

Copyright (C) 2016 Acksin hey@acksin.com

This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0. If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/.