smilodon

Smilodon manages attachment of EBS and ENI pairs in AWS EC2 auto scaling groups.

Think about zookeeper, etcd or similar datastores, where a data volume and IP address have to always go together. Achieving that can be tricky, especially if you want to take advantage of EC2 auto scaling groups.

Getting Started

• First you need to create a number of EBS and ENI resources and tag them with NodeID. NodeID tag value can be anything as long as you have a matching EBS and ENI pair with the same NodeID tag value.

• Secondly, create an autoscaling group with a number of instances with smilodon provisioned. You can have more instances in auto scaling group than you have EBS+ENI pairs.

When smilodon starts it will try to find a matching EBS+ENI pair and attach them. EBS gets attached first and ENI second. You can tell smilodon to create a file system and mount it for convenience.

Smilodon does not know how to start a service, but that work is in progress right now.

If your distro uses systemd, then you can easily tell your service unit to watch for when a specific mount point is ready and start the service unit.

Required AWS Permissions

You have two options here.

• Create an IAM user and provide smilodon with AWS API credentials via environment variables.

• Define an IAM instance role and reference that in the LaunchConfiguration.

I recommend the latter, but regardless which option you pick, you need to allow the following IAM permissions (below could be a lot more granular and specific):

Resource: "*"
Action:
  - ec2:DescribeInstances
  - ec2:DescribeTags
  - ec2:DescribeNetworkInterfaces
  - ec2:DescribeVolumes
  - ec2:AttachVolume
  - ec2:AttachNetworkInterface
  - ec2:DetachNetworkInterface
  - ec2:ModifyNetworkInterfaceAttribute

Configuration

Configuration is done using command line flags - smilodon --help.

Filtering AWS Resources

It is very likely that you have many EBS volumes and ENI devices in your AWS account.

When you create your EBS+ENI resources, it makes sense to tag them with a service name or something sensible, so that smilodon attaches correct resources.

Only NodeID tag is required and the rest can be arbitrary. For example, you could tag your resources with the following tags:

NodeID=[0-5]
Env=development
Service=etcd
Project=<any value>

To tell smilodon to only look for resources with the above tags, do this:

smilodon --filters='tag:Env=development,tag:Service=etcd,tag-key=Project'

As you can see above, last filter matches on any value of tag Project. You can also filter on a bunch of other AWS specific filters.