A lightweight Docker container to drain EC2 Spot instances from your ECS cluster
On start up, this Docker container will check if the EC2 it is running on is a spot instance or not. If not, it will remain running, but will sleep forever. If it is a spot instance, it will try to read some config from ECS about itself, then check for EC2 termination every 5 seconds, forever. If termination is detected, it will attempt to notify ECS that the EC2 instance should be put into a "DRAINING" state.
The container is based off Amazon's official amazon/aws-cli container. The only dependency is jq, and the script is written in bash, keeping the memory footprint extremely low (~2MB in my testing).
Environment variables for additional configuration:
| Key | Default | Description |
|---|---|---|
ECS_AGENT |
172.17.0.1:51678 |
Used to overwrite the default IP/host/port of the ecs-agent |
EXIT_IF_NOT_SPOT |
false |
If set true, script will exit if the instance-life-cycle is set to normal, and not spot. This is usually not preferred, as your container scheduler may consider this to be an error, and try to relaunch. |
SHOW_OK |
false |
If set true, script will print "OK" for every 5 second loop. |