AWS CoreOS cluster provisioning with Terraform
- Overview
- Setup AWS credentials
- Install tools
- Quick start
- Customization
- Build multi-node cluster
- Manage individual platform resources
- Technical notes
This is a practical implementation of [CoreOS cluster architectures ] (https://coreos.com/os/docs/latest/cluster-architectures.html) built on AWS. The cluster follows CoreOS production cluster model that contains an autoscaling etcd cluster, and an autoscaling worker cluster for hosted containers. You can optionally add an admiral cluster for shared services such as CI, private docker registry, logging and monitoring, etc.
The entire infrastructure is managed by Terraform.
For other type of Unix cluster, see a similar repo aws-linux-cluster.
Go to AWS Console.
- Signup AWS account if you don't already have one. The default EC2 instances created by this tool is covered by AWS Free Tier (https://aws.amazon.com/free/) service.
- Create a group
coreos-cluster
withAdministratorAccess
policy. - Create a user
coreos-cluster
and Download the user credentials. - Add user
coreos-cluster
to groupcoreos-cluster
.
If you use Vagrant, you can skip this section and go to Quick Start section.
Instructions for install tools on MacOS:
-
Install Terraform
$ brew update $ brew install terraform
or
$ mkdir -p ~/bin/terraform $ cd ~/bin/terraform $ curl -L -O https://dl.bintray.com/mitchellh/terraform/terraform_0.6.0_darwin_amd64.zip $ unzip terraform_0.6.0_darwin_amd64.zip
-
Install Jq
$ brew install jq
-
Install AWS CLI
$ brew install awscli
or
$ sudo easy_install pip $ sudo pip install --upgrade awscli
For other platforms, follow the tool links and instructions on tool sites.
Install AWS CLI:
```
$ brew install awscli
```
or
```
$ sudo easy_install pip
$ sudo pip install --upgrade awscli
```
$ git clone https://github.com/xuwang/aws-terraform.git
$ cd aws-terraform
If you use Vagrant, instead of install tools on your host machine, there is Vagranetfile for a Ubuntu box with all the necessary tools installed:
$ vagrant up
$ vagrant ssh
$ cd aws-terraform
$ aws configure --profile coreos-cluster
Use the downloaded aws user credentials when prompted.
The above command will create a coreos-cluster profile authentication section in ~/.aws/config and ~/.aws/credentials files. The build process bellow will automatically configure Terraform AWS provider credentials using this profile.
This default build will create one etcd node and one worker node cluster in a VPC, with application buckets for data, necessary iam roles, polices, keypairs and keys. The instance type for the nodes is t2.micro. You can review the configuration and make changes if needed. See Customization for details.
$ make
... build steps info ...
... at last, shows the worker's ip:
worker public ips: 52.27.156.202
...
$ make show
...
module.etcd.aws_autoscaling_group.etcd:
id = etcd
availability_zones.# = 3
availability_zones.2050015877 = us-west-2c
availability_zones.221770259 = us-west-2b
availability_zones.2487133097 = us-west-2a
default_cooldown = 300
desired_capacity = 1
force_delete = true
health_check_grace_period = 0
health_check_type = EC2
launch_configuration = terraform-4wjntqyn7rbfld5qa4qj6s3tie
load_balancers.# = 0
max_size = 9
min_size = 1
name = etcd
tag.# = 1
....
$ ssh -A [email protected]
CoreOS beta (723.3.0)
[email protected] ~ $ fleetctl list-machines
MACHINE IP METADATA
289a6ba7... 10.0.1.141 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=etcd2
320bd4ac... 10.0.5.50 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=worker
$ make destroy_all
This will destroy ALL resources created by this project.
$ vagrant destroy
-
The default values for VPC, ec2 instance profile, policies, keys, autoscaling group, lanuch configurations etc., can be override in resources/terraform/module-.tf` files.
-
AWS profile and cluster name are defined at the top of Makefile:
AWS_PROFILE := coreos-cluster CLUSTER_NAME := coreos-cluster
These can also be customized to match your AWS profile and cluster name.
The number of etcd nodes and worker nodes are defined in resource/terraform/module-etcd.tf and resource/terraform/module-worker.tf
Change the cluster_desired_capacity in the file to build multi-nodes etcd/worker cluster, for example, change to 3:
cluster_desired_capacity = 3
Note: etcd minimum, maximum and cluster_desired_capacity should be the same and in odd number, e.g. 3, 5, 9
You should also change the aws_instance_type
from micro
to medium
or large
if heavy docker containers to be hosted on the nodes:
image_type = "t2.medium"
root_volume_size = 12
docker_volume_size = 120
To build:
$ make all
... build steps info ...
... at last, shows the worker's ip:
worker public ips: 52.26.32.57 52.10.147.7 52.27.156.202
...
Login to a worker node:
$ ssh -A [email protected]
CoreOS beta (723.3.0)
[email protected] ~ $ etcdctl cluster-health
cluster is healthy
member 34d5239c565aa4f6 is healthy
member 5d6f4a5f10a44465 is healthy
member ab930e93b1d5946c is healthy
core@ip-10-0-1-92 ~ $ etcdctl member list
34d5239c565aa4f6: name=i-65e333ac peerURLs=http://10.0.1.92:2380 clientURLs=http://10.0.1.92:2379
5d6f4a5f10a44465: name=i-cd40d405 peerURLs=http://10.0.1.185:2380 clientURLs=http://10.0.1.185:2379
ab930e93b1d5946c: name=i-ecfa0d1a peerURLs=http://10.0.1.45:2380 clientURLs=http://10.0.1.45:2379
[email protected] ~ $ fleetctl list-machines
MACHINE IP METADATA
0d16eb52... 10.0.1.92 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=etcd2
d320718e... 10.0.1.185 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=etcd2
f0bea88e... 10.0.1.45 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=etcd2
0cb636ac... 10.0.5.4 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=worker
4acc8d6e... 10.0.5.112 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=worker
fa9f4ea7... 10.0.5.140 env=coreos-cluster,platform=ec2,provider=aws,region=us-west-2,role=worker
You can create individual resources and the automated-scripts will create resources automatically based on dependencies.
$ make help
Usage: make (<resource> | destroy_<resource> | plan_<resource> | refresh_<resource> | show | graph )
Available resources: vpc s3 route53 iam etcd worker
For example: make worker # to show what resources are planned for worker
Currently defined resources:
Resource | Description |
---|---|
vpc | VPC, gateway, and subnets |
s3 | S3 buckets |
iam | Setup a deployment user and deployment keys |
route53 | Setup public and private hosted zones on Route53 DNS service |
elb | Setup application ELBs |
etcd | Setup ETCD2 cluster |
worker | Setup application docker hosting cluster |
admiral | Central service cluster (Jenkins, fleet-ui, monitoring, logging, etc) |
dockerhub | Private docker registry cluster |
rds | RDS servers |
cloudtrail | Setup AWS CloudTrail |
To build the cluster step by step:
$ make init
$ make vpc
$ make etcd
$ make worker
Make commands can be re-run. If a resource already exists, it just refreshes the terraform status.
This will create a build/ directory, copy all terraform files to the build dir, and execute correspondent terraform cmd to build the resource on AWS.
To destroy a resource:
$ make destroy_<resource>
- Etcd cluster is on an autoscaling group. It should be set with a fixed, odd number (1,3,5..), and cluster_desired_capacity=min_size=max_size.
- Cluster discovery is managed with dockerage/etcd-aws-cluster image. etcd cluster is formed by self-discover through its auto-scaling group and then an etcd initial cluster is updated automatically to s3://AWS-ACCOUNT-CLUSTER-NAME-cloudinit/etcd/initial-cluster s3 bucket. Worker nodes join the cluster by downloading the etcd initial-cluster file from the s3 bucket during their bootstrap.
- AWS resources are defined in resources and modules directories. The build process will copy all resource files from resources to a build directory. The terraform actions are performed under build, which is ignored in .gitignore. The original Terraform files in the repo are kept intact.
- Makefiles and shell scripts are used to give us more flexibility on tasks Terraform leftover. This provides stream-lined build automation.
- All nodes use a common bootstrap shell script as user-data, which downloads initial-cluster file and nodes specific cloud-config.yaml to configure the node. If cloud-config changes, no need to rebuild an instance. Just reboot it to pick up the change.
- CoreOS AMI is generated on the fly to keep it up-to-data. Default channel can be changed in Makefile.
- Terraform auto-generated launch configuration name and CBD feature are used to allow launch configuration update on a live autoscaling group, however, running ec2 instances in the autoscaling group has to be recycled outside of the Terraform management to pick up the new LC.
- For a production system, the security groups defined in etcd, worker, and admiral module should be carefully reviewed and tightened.
Once the tools are installed run
$ aws configure
AWS Access Key ID [None]: <YOUR ACCESS KEY>
AWS Secret Access Key [None]: <YOUR SECRET KEY>
Default region name [None]: <YOUR PREFRED REGION>
Default output format [None]:
When prompted for the access and secret keys, enter the ones you saved earlier. Set the default region to your prefered one (based on the area you are located in) and the output format can be left as default.
Here is list of AWS Regions: http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/using-regions-availability-zones.html#concepts-available-regions
You will now need to create a file in this directory called terraform.tfvars with contents like this:
access_key = "YOUR ACCESS KEY"
secret_key = "YOUR SECRET KEY"
allowed_network = "YOUR NETWORK CIDR"
Choose a region that has at-least 3 availability zones! as otherwise the script will fail! e.g.
us-west-2
For complete list please take a look here: http://www.stelligent.com/cloud/list-all-the-availability-zones/
# For get-ami.sh
COREOS_UPDATE_CHANNE=stable
AWS_REGION=us-west-2
VM_TYPE=hvm
- add possibility that user can specify the number of AZ his region has so, that script doesn't fail. user should be able to setup in region which has currently 2 AZs only!
- bastion host -- create default users which can be used to get into the environment