IRC: #acksin on Freenode
Acksin is a Cloud and Container aware diagnostics and tuning tool. It uses Machine Learning to find optimizations in your infrastructure. The goal is to make your servers more performant, reduce the amount you spend on servers, and help reduce the environmental footprint.
Acksin outputs its data in JSON to the command line. Run the following command:
sudo acksin output
Acksin primarily runs as a daemon which regularily pushes diagnostics to a central server. Acksin runs a service called Acksin Console 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 near future.
Run the following:
sudo acksin agent config.json
All documentation is on the Acksin website.
Acksin's command line portion is primarily written in Go whereas the Machine Learning is written in Python. The code is split into a couple different sections:
- Command Line Tool: Collects stats from the System and Containers.
- Mental Models: Take System stats and creates models for AI. Currently this is a program that runs on AWS Lambda.
- Tensorflow AI: We use the output generated from the Mental Models to create train AI for the various tasks.
- Console: ReactJS Frontend App used on Acksin Cloud.
- Server: This component is not yet open sourced. This is will be a Go server which will be built into the command line.
The primary dependency of Acksin is the ProcFS Library we use. Any code that needs to read from ProcFS should go there. Most of the Command Line App is a wrapper for that library. In the future we will have similar dependencies for SysFS. In addition to that we use the Go libraries provided by the Cloud providers.
Acksin has several components that need to be deployed to make a complete system. This includes the Command Line Tool, Mental Models, AI, Console, and Server.
To build the command line tool run the following:
make deps
make build
Currently to run the Mental Models you need to install the Serverless Framework as they run on AWS Lambda. We will split this out so it doesn't rely on Lambda in the near future.
Make sure to create a ai/mental_models/serverless.env.yaml file.
Example:
vars: null
stages:
dev:
vars: null
regions:
us-west-2:
prod:
vars: null
regions:
us-west-2:
To deploy this function run.
cd ai
make deps
make dev
We love contributors to the project. Please check out the CONTRIBUTING.md file.
Acksin 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. We need to understand the entire system.
Acksin'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. Acksin is not trying to replace application level instrumentation and monitoring services such as Graphite and Datadog. Acksin is being built to augment those services.
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.
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/.