AWS user data script for setting up a consul cluster using all instances in a given ECS

user-data.sh

#!/bin/bash -xe
S3_BUCKET=foo
EXPECTED_SERVERS=2

yum install -y aws-cli
aws s3 cp s3://$S3_BUCKET/ecs.config /etc/ecs/ecs.config
aws s3 cp s3://$S3_BUCKET/docker.config /etc/sysconfig/docker

run_consul_agent() {
    peers=""

    if [ "$1" ]
    then
        peers="-retry-join $1"
    fi

    docker run -d \
        --name=consul-agent \
        -e 'CONSUL_LOCAL_CONFIG={"leave_on_terminate": true}' \
        -p $EC2_IP:8300:8300 \
        -p $EC2_IP:8301:8301 \
        -p $EC2_IP:8301:8301/udp \
        -p $EC2_IP:8302:8302 \
        -p $EC2_IP:8302:8302/udp \
        -p $EC2_IP:8400:8400 \
        -p $EC2_IP:8500:8500 \
        -p $DOCKER_BRIDGE:8500:8500 \
        -p $DOCKER_BRIDGE:53:8600 \
        -p $DOCKER_BRIDGE:53:8600/udp \
        consul:v0.6.4 consul agent \
                -data-dir=/consul/data \
                -node=$HOSTNAME \
                -client 0.0.0.0 \
                -advertise $EC2_IP $peers
}

run_consul_server() {
    peers=""

    if [ "$1" ]
    then
        peers="-retry-join $1"
    fi

    docker run -d \
        --name=consul-server \
        -e 'CONSUL_LOCAL_CONFIG={"leave_on_terminate": true}' \
        -p $EC2_IP:8300:8300 \
        -p $EC2_IP:8301:8301 \
        -p $EC2_IP:8301:8301/udp \
        -p $EC2_IP:8302:8302 \
        -p $EC2_IP:8302:8302/udp \
        -p $EC2_IP:8400:8400 \
        -p $EC2_IP:8500:8500 \
        -p $DOCKER_BRIDGE:8500:8500 \
        -p $DOCKER_BRIDGE:53:8600 \
        -p $DOCKER_BRIDGE:53:8600/udp \
        consul:v0.6.4 consul agent \
                -server \
                -bootstrap-expect=$EXPECTED_SERVERS \
                -data-dir=/consul/data \
                -node=$HOSTNAME \
                -client 0.0.0.0 \
                -advertise $EC2_IP $peers
}

# Reconfigure Docker
DOCKER_BRIDGE=$(ifconfig docker0 | sed -n -e 's/^ \+inet addr:\([0-9.]\+\).\+$/\1/p')
sed -i "s/PLACEHOLDER_DOCKER_BRIDGE/${DOCKER_BRIDGE}/g" "/etc/sysconfig/docker"

# Reload Docker with new config
service docker reload

# Import cluster name as ECS_CLUSTER
source /etc/ecs/ecs.config

# Get other instance data from AWS meta-data
EC2_IP=$(curl http://169.254.169.254/latest/meta-data/local-ipv4)
EC2_ZONE=$(curl -s http://169.254.169.254/latest/meta-data/placement/availability-zone)
EC2_REGION=$(echo $EC2_ZONE | sed -n -e 's/\([a-z]\+-[a-z]\+-[0-9]\+\).\+/\1/p')

# Fetch cluster instance ARNs
INSTANCE_ARNS=$(aws ecs list-container-instances \
    --region $EC2_REGION \
    --cluster $ECS_CLUSTER \
    --query 'containerInstanceArns' \
    --output text | sed 's/\t/ /g')
echo "INSTANCE_ARNS: $INSTANCE_ARNS"

# Fetch cluster instance IDs
if [ "$INSTANCE_ARNS" ]
then
    INSTANCE_IDS=$(aws ecs describe-container-instances \
        --region $EC2_REGION \
        --cluster $ECS_CLUSTER \
        --container-instances $INSTANCE_ARNS \
        --query "containerInstances[*].ec2InstanceId" \
        --output text | sed 's/\t/ /g')
fi
echo "INSTANCE_IDS: $INSTANCE_IDS"

# Fetch cluster instance IPs
if [ "$INSTANCE_IDS" ]
then
    CLUSTER_IPS=$(aws ec2 describe-instances \
        --region $EC2_REGION \
        --instance-ids $INSTANCE_IDS \
        --query "Reservations[*].Instances[*].PrivateIpAddress" \
        --output text | sed 's/\t/ /g')
fi
echo "CLUSTER_IPS: $CLUSTER_IPS"

# Run consul
if [ "$CLUSTER_IPS" ]
then
    CONSUL_CONTAINER_ID=$(run_consul_agent "$CLUSTER_IPS")
    echo "CONSUL_CONTAINER_ID: $CONSUL_CONTAINER_ID"

    sleep 2

    CONSUL_INFO=$(docker exec $CONSUL_CONTAINER_ID consul info)
    echo "CONSUL_INFO: $CONSUL_INFO"

    IS_SERVER=$(echo $CONSUL_INFO | sed -n -e 's/.\+server = \(true\|false\).\+/\1/p')
    echo "IS_SERVER: $IS_SERVER"

    KNOWN_SERVERS=$(echo $CONSUL_INFO | sed -n -e 's/.\+known_servers = \([0-9]\+\).\+/\1/p')
    echo "KNOWN_SERVERS: $KNOWN_SERVERS"

    if [ "$KNOWN_SERVERS" -lt "$EXPECTED_SERVERS" ]
    then
        docker stop $CONSUL_CONTAINER_ID
        docker rm $CONSUL_CONTAINER_ID
        run_consul_server "$CLUSTER_IPS"
    fi
else
    echo "NO KNOWN SERVERS"
    run_consul_server
fi