This project is intended to bring auto service staling to Docker Swarm. This script uses prometheus paired with cadvisor metrics to determine cpu usage. It then uses a manager node to determine if a service wants to be autoscaled and uses a manager node to scale the service.

Currently the project only uses cpu to autoscale. If cpu usage reaches 85% the service will scale up, if it reaches 25% it will scale down.

1. You can deploy prometheus, cadvisor, and docker-swarm-autoscaler by running docker stack deploy -c swarm-autoscaler-stack.yml autoscaler from the root of this repo.

• You can also utilize an already deploy prometheus and cadvisor by specifying the PROMETHEUS_URL in docker-swarm-autoscaler environment. swarm-autoscaler-stack.yml shows an example of this.
• docker-swarm-autoscale needs a placement contstraint to deploy to a manager. swarm-autoscaler-stack.yml shows an example of this.

2. For services you want to autoscale you will need a deploy label swarm.autoscaler=true.

deploy:
  labels:
    - "swarm.autoscaler=true"

This is best paired with resource constraints limits. This is also under the deploy key.

deploy:
  resources:
    reservations:
      cpus: '0.25'
      memory: 512M
    limits:
      cpus: '0.50'

You can deploy a test app with the following commands below. Helloworld is initially only 1 replica. The autoscaler will scale to the minimum 3 replicas.

1. docker stack deploy -c swarm-autoscaler-stack.yml autoscaler
2. docker stack deploy -c helloworld.yml hello