Easy deployment of Deep Learning predictions as functions thanks to OpenFaas awesome projet :-)

List of current functions :

• darknet

More to come...

You can create a single-host Docker Swarm on your laptop with a single command. You don't need any additional software to Docker 17.05 or greater. You can also run these commands on a Linux VM or cloud host.

This is how you initialize your master node:

# docker swarm init

If you have more than one IP address you may need to pass a string like --advertise-addr eth0 to this command.

Take a note of the join token

• Join any workers you need

Log into your worker node and type in the output from docker swarm init on the master. If you've lost this info then type in docker swarm join-token worker and then enter that on the worker.

It's also important to pass the --advertise-addr string to any hosts which have a public IP address.

Note: check whether you need to enable firewall rules for the Docker Swarm ports listed here.

• Connect to a swarm manager
• Clone this repository

git clone https://github.com/yogeek/deeplearning-as-a-function.git

• chmod u+x install.sh
• Execute the install script :

./install.sh

Your functions are available !

Classify an image with darknet :

• Get an image

wget https://upload.wikimedia.org/wikipedia/commons/thumb/1/19/%C3%81guila_calva.jpg/1200px-%C3%81guila_calva.jpg -O eagle.jpg

• Call the function from any node

curl localhost:8080/function/darknet --data-binary @eagle.jpg

or with faas-cli

cat eagle.jpg | faas-cli invoke darknet

To use in production mode, follow the guide to lock-down OpenFaas with a reverse proxy like Traefik for example.

The following script will :

• Ask for a password
• Add Traefik reverse proxy in the stack
• Deploy the secured stack

chmod u+x install.sh
./install-secure.sh

Test :

curl -X POST http://localhost/function/darknet --data-binary @eagle.jpg
401 : Acces Denied

curl -u user:password -X POST http://localhost/function/darknet --data-binary @eagle.jpg
OK !