Skip to content

Latest commit

 

History

History
248 lines (163 loc) · 10.7 KB

README.rst

File metadata and controls

248 lines (163 loc) · 10.7 KB
Raw

Madcore CLI

Deep Learning & Machine Intelligence Infrastructure Controller

https://travis-ci.org/madcore-ai/cli.svg?branch=master

What is Madcore?

Madcore is a CLI tool for deployment and auto-configuration of data mining and analytics microservices. It is a Kubernetes-based unmodified KOPS/Minikube installation manager. However, single point of truth is preserved as a unified yaml file called "clusterfile". Clusterfile controls generic aspects of provisioning, deployment, scale and configuration. All KOPS and Kubernetes templates are then populated from input clusterfile.

Install

Mac & Linux install form terminal.

pip install madcore

Development Install

If you want to contribute to MadcoreCLI, add new k8s modules or create your own clusterfiles you should clone the repo, create python virtual environment (make sure package is installed) to isolate installations from the rest of the system, activate the virtual env, perform requirements install, and create a symlink so you can call MadcoreCLI from any folder on the system. For below example we assume both cloned repo and new virtual environment will end up in ~/git folder in your user space and full path is /Users/peter/git Obviously you can adjust this to match your system.

mkdir ~/git && cd ~/git
virtualenv venv_cli && source venv_cli/bin/activate
git clone [email protected]:madcore-ai/cli.git
pip install -r ~git/cli/requirements.txt
ln -s /Users/peter/git/cli/bin/madcore.sh /usr/local/bin/madcore

Minikube Environment Prerequisites

  • Virtual Box
  • Minikube 1.9
  • Local PC 16GB of Ram (minikube is set at 8GB by default)

AWS Environment Prerequisites

  • VPC in AWS (you will need id)
  • Internet Gateway attached to VPC
  • S3 Storage bucket for KOPS settings

Provision Locally on Minikube

Minikube is great but obviously limited by specs of your local machine. You can comment out elements of minikube.yaml to suit your needs. Then when you're ready start provisioning. When you're done, run second command to map minikube.local hostname to your new setup.

madcore --provision minikube.yaml
madcore --mini-hostname

Provision in AWS

Currently Madcore is tested on Mac and Linux only. We are working on exposing clusterfiles and templates in a better way. By default they install with the python project files location similar to this lib/python2.7/site-packages/madcore

madcore --provision demo.yaml

AWS POST-INSTALL:

  • Create <yourdomain.com> A record and point it to ingress IP (ingress horizontal scaling above 500MB/s is described in another doc)
  • Create wildcard CNAME *.<yourdomain.com> and point it to your above hostname (will automate this eventually)
  • Create Security Group in your VPC and whitelist your access IP's, attach it to ingress node (will automate this eventually)

Madcore Data Mining & Deep Learning Ecosystem

Functionality is grouped into instance groups (physically) and into namespaces (logically). Each software deployed here belongs to their respective owners. We do not interfere in containers but make sure that we find best containers for deployment in Kubernetes.

Goal of Madcore is to abstract deployment and configuration of data processing elements and have it available in working state out-of-the-box. This way anyone can start work on their actual problem and not spend time on deployment and configuration of common toolsets.

usage: ./madcore.py [-h]
                    [-p CLUSTERFILE | -c CLUSTERFILE | --destroy | --kops-update | --kops-validate | --kubectl-use-context | --mini-hostname | --get-attr ATTR | --install-core | --install-elk | --install-neo4j | --install-kafka | --install-flink]

Madcore CLI 1.9.15 - (c) 2016-2018 Madcore Ltd <https://madcore.ai>

optional arguments:
  -h, --help            show this help message and exit
  -p CLUSTERFILE, --provision CLUSTERFILE
                        provision based on <cllusterfile>
  -c CLUSTERFILE, --clusterfile CLUSTERFILE
                        set default clusterfile to input <clusterfile>
  --destroy             destroy infrastructure
  --kops-update         kops update
  --kops-validate       kopds validate
  --kubectl-use-context
                        kubectl use context
  --mini-hostname       set minikube hostname (will sudo)
  --get-attr ATTR       get atribute
  --install-core        install core of Madcore
  --install-elk         install elk
  --install-neo4j       install neo4j
  --install-kafka       install apache kafka
  --install-flink       install apache flink

Deploy Core

Installation of core elements is a single command. Filenames in range of 100-200. You can comment out any of those installs. By commenting corresponding lines in your aws clusterfile. Registry and metrics elements are optional. You probably want to leave dashboard and ingress setup as everything else maps to it.

madcore --install-core
https://asciinema.org/a/179330.png
Core Stack Description
dashboard Kubernetes Dashboard
nfs NFS 4.1 for utilized for Kubernetes persistent volume claims (StatefulSets)
registry2 (optional) docker registry v2
influxdb InfluxDB for Heapster data
heapster Kubernetes metrics collector
grafana Grafana Dashboard pointed at InfluxDB for kube metrics
haproxy-ingress HAProxy ingress (route external traffic and map to kube services)
ingress-default default container reporting 404 when hitting anything but mapped endpoints
ingress echo echo container to test ingress alive

Deploy neo4j

Neo4j and Dashboard is in the template file space of 9220-9229. Deploy using command below. Few second later you will have a working dashboard and single pod engine configuration ready to start your tests. Thi deployment is installed onto standard nodes instancegroup. This deployment lives its own neo4j namespace. It's easy to remove it when you don't require it anymore. It using standard neo4j:3.1.4-enterprise containers from docker hub maintainer by neo4j team. It is exposed through ingress and mapped through its own subodmain neo4j.<yourdomain.com>

madcore --install-neo4j
Neo4J Stack Description
engine Enterprise: neo4j:3.1.4-enterprise (subject to EULA)
ui Dashboard

Deploy kafka

Kafka and Dashboard is in the template file space of 9240-9249. Deploy using command below. Few second later you will have a working dashboard and single pod engine configuration ready to start your tests. Thi deployment is installed onto standard nodes instancegroup. This deployment lives its own kafka namespace. It's easy to remove it when you don't require it anymore. It is exposed through ingress and mapped through its own subodmain kafka.<yourdomain.com> for Yahoo kafka dashboard and kafka.<yourdomain.com>/rest for Mailgun Pixy rest ui (grpc is listening internally but not exposed outside)

madcore --install-kafka
Kafka Stack Containers
zookeeper solsson/kafka:1.0.1
kafka solsson/kafka:1.0.1
kafka-manager solsson/kafka-manager
kafka-pixy mailgun/kafka-pixy

Deploy Elasticsearch / FluentD / Kibana

Famous trio optimized for Kubernetes. Elasticsearch exposed through ingress as well as Kibana. Internally FluentD DaemonSets are deployed to ALL nodes and collect all logs from pods stdout along with kubernetes logs and aggregate in ElasticSearch. Deploy this when you have a need. There is a dedicated instance group for ELK so it doesn't collide with any of your other applications.

madcore --install-elk
Kafka Stack Containers
elasticsearch docker.elastic.co/elasticsearch/elasticsearch-oss:6.0.0
fluentd fluent/fluentd-kubernetes-daemonset:v0.12.33-elasticsearch
kibana docker.elastic.co/kibana/kibana-oss:6.0.0

Deploy Apache Flink Cluster

Apache Flink is an open source stream processing framework developed by the Apache Software Foundation. The core of Apache Flink is a distributed streaming dataflow engine written in Java and Scala

madcore --install-flink
Flink Stack Description
jobmanager Flink Job Manager
jobmanager-ui Flink Web Ui
taskmanager Flink Task Manager (Horizontally Scaling)

Chat with us on Gitter

If you want to try Madcore, make sure you join us on Gitter. We are now focused on building Machine Learning and Ai plugins as well as building Ingress listeners for social media and queueing mechanisms in Spark and Kafka. All based on Kubernetes. Chat with us now: https://gitter.im/madcore-ai/core

Mailing List

Visit https://madcore.ai to sign up for weekly newsletter on Machine Learning and AI simulations that are now possible with Madcore

Credits

We will be adding a formal Credits file into this project. For now just want to make clear that all registered brands/products remain property of their respective owners.

License

Madcore Project is distributed on MIT License (c) 2016-2017 Madcore Ltd (London, UK) https://madcore.ai