Demo Instagram object recognition service on DCOS/Marathon

Given an Instagram username, return the recognized objects from the user's latest image posts.

• microservice architecture based on DCOS/Marathon
• TensorFlow + ImageNet model as deep learning backend

Architecture

Front api service

• Provide Web UI to accept an Instagram username as input. It then send requests to the backend recognition service.
• Routed with marathon-lb

Backend recognition service

• Uses TensorFlow to classify the input image with pre-trained ImageNet model.
• Routed using DCOS VIP

Deployment

setup

Clone the repo

git clone https://github.com/xiaoganghan/mesos-instagram-image-recognition.git

Install and boot dcos-vagrant

Follow the instructions DC/OS-vagrant to install DC/OS with Vagrant.

suppose your DC/OS-vagrant directory is dcos-vagrant

cp mesos-instagram-image-recognition/ops/VagrantConfig.yaml dcos-vagrant/VagrantConfig.yaml
cd dcos-vagrant
vagrant up m1 a1 p1 boot 

Install DC/OS Command Line Interface (CLI)

Follow the instruction DC/OS CLI to install DC/OS CLI

Install marathon-lb

cd mesos-instagram-image-recognition/ops
dcos package install --options=marathon-lb.json marathon-lb --yes

Deploy Backend recognition services (3 instances)

Install Docker

Install the Docker software by following the Docker Installation

Creat a Docker Hub account

Create a Docker Hub account to store the docker images created in the following step, assume your docker account name is 'DockerAccount'

build docker image, test locally, and push it to dockerhub

cd mesos-instagram-image-recognition/backend

build a docker image called 'mesos-instagram-detect-backend' use sudo under ubuntu if 'Cannot connect to the Docker daemon, Is the docker daemon running on this host?' error shown

docker build -t mesos-instagram-detect-backend .   

find out the id of image 'mesos-instagram-detect-backend'

docker images

tag the image

docker tag [image_id] DockerAccount/mesos-instagram-detect-backend:latest

run locally to test

docker run -it -p 5000:5000 --rm  DockerAccount/mesos-instagram-detect-backend

open url in browser to test

If correct, the following results should be shown in the browser

{ objects: [ "stage (score = 0.02232)", "wool, woolen, woollen (score = 0.01913)", "mitten (score = 0.01680)", "balloon (score = 0.01405)", "cloak (score = 0.01377)" ] }

push to Docker Hub

docker push DockerAccount/mesos-instagram-detect-backend:latest

deploy backend to DCOS

cd mesos-instagram-image-recognition/ops

Using Web UI: create application using ops/launch_backend.json or,

Using DC/OS CLI:

dcos marathon app add launch_backend.json

deploy front server (2 instances)

build docker image, test locally, and push it to dockerhub

cd mesos-instagram-image-recognition/front
docker build -t DockerAccount/mesos-instagram-detect-frontend:latest .

run locally to test

docker run -it -p 5001:5001 --rm  DockerAccount/mesos-instagram-detect-frontend

find out the id of image 'mesos-instagram-detect-frontend'

docker images

tag the image

docker tag [image_id] DockerAccount/mesos-instagram-detect-frontend:latest

push to Docker Hub

docker push DockerAccount/mesos-instagram-detect-frontend:latest

deploy frontend to DCOS

cd mesos-instagram-image-recognition/ops

Using Web UI: create application using ops/launch_front.json or, Using DC/OS CLI:

dcos marathon app add launch_front.json

verify it works

open http://instagramdemo.com in your browser

references

• marathon-lb settings
  • https://github.com/dcos/dcos-vagrant/blob/e9c0d6ade775debf0889b69b67f87cfdbf1aec9b/examples/oinker/README.md
• VIP settings
  • https://github.com/dcos/dcos-vagrant/tree/master/examples/minuteman
  • https://github.com/dcos/dcos-docs/pull/199/files