Images

Changing the image for an instance group

You can choose a different AMI for an instance group.

If you kops edit ig nodes, you should see an image member of the spec.

Various syntaxes are available:

  • ami-abcdef specifies an AMI by id directly.
  • <owner>/<name> specifies an AMI by its owner and Name properties

The ami spec is precise, but AMIs vary by region. So it is often more convenient to use the <owner>/<name>specifier, if equivalent images have been copied to various regions with the same name.

For example, to use Ubuntu 16.04, you could specify:

image: 099720109477/ubuntu/images/hvm-ssd/ubuntu-xenial-16.04-amd64-server-20180405

You can find the name for an image using e.g. aws ec2 describe-images --image-id ami-493f2f29

(Please note that ubuntu is currently undergoing validation testing with k8s - use at your own risk!)

If you are creating a new cluster you can use the --image flag when running kops create cluster,which should be easier than editing your instance groups.

In addition, we support a few-well known aliases for the owner:

  • kope.io => 383156758163
  • redhat.com => 309956199498
  • coreos.com => 595879546273
  • amazon.com => 137112412989

Debian

A Debian image with a custom kubernetes kernel is the primary (default) platform for kops.

We run a Debian Jessie image, with a 4.4 (stable series) kernel that is built with kubernetes-specific settings.

The tooling used to build these images is open source:

The latest image name is kept in the stable channel manifest,but an example is kope.io/k8s-1.4-debian-jessie-amd64-hvm-ebs-2016-10-21. This means to look for an image publishedby kope.io, (which is a well-known alias to account 383156758163), with the namek8s-1.4-debian-jessie-amd64-hvm-ebs-2016-10-21. By using a name instead of an AMI, we can reference an imageirrespective of the region in which it is located.

kops should also now work on stock Debian 9 (Stretch) images. Stock Debian 8 (Jessie) images are not recommended,as they typically do not have a suitable kernel and kernel options configured.

Ubuntu

Ubuntu is not the default platform, but is believed to be entirely functional.

Ubuntu 16.04 or later is required (we require systemd).

For example, to use Ubuntu 16.04, you could specify:

image: 099720109477/ubuntu/images/hvm-ssd/ubuntu-xenial-16.04-amd64-server-20180405

You can find the name for an image by first consulting Ubuntu’s image finder,and then using e.g. aws ec2 describe-images --image-id ami-493f2f29

CentOS

CentOS7 support is still experimental, but should work. Please report any issues.

The following steps are known:

Be aware of the following limitations:

  • CentOS 7.2 is the recommended minimum version
  • CentOS7 AMIs are running an older kernel than we prefer to run elsewhere

RHEL7

RHEL7 support is still experimental, but should work. Please report any issues.

The following steps are known:

  • Redhat AMIs can be found using aws ec2 describe-images --region=us-east-1 --owner=309956199498 --filters Name=virtualization-type,Values=hvm
  • You can specify the name using the redhat.com owner alias, for example redhat.com/RHEL-7.2_HVM-20161025-x86_64-1-Hourly2-GP2

Be aware of the following limitations:

  • RHEL 7.2 is the recommended minimum version
  • RHEL7 AMIs are running an older kernel than we prefer to run elsewhere

CoreOS

CoreOS has been tested enough to be considered ready for production with kops, but if you encounter any problem please report it to us.

The following steps are known:

  • The latest stable CoreOS AMI can be found using:
    1. aws ec2 describe-images --region=us-east-1 --owner=595879546273 \
    2.   --filters "Name=virtualization-type,Values=hvm" "Name=name,Values=CoreOS-stable*" \
    3.   --query 'sort_by(Images,&CreationDate)[-1].{id:ImageLocation}'

Also, you can obtain the “AMI ID” from CoreOS web page too. They publish their AMI’s using a json file at https://coreos.com/dist/aws/aws-stable.json. Using some scripting and a “json” parser (like jq) you can obtain the AMI ID from a specific availability zone:

  1. curl -s https://coreos.com/dist/aws/aws-stable.json | jq -r '.["us-east-1"].hvm'
  2. "ami-32705b49"
  • You can specify the name using the coreos.com owner alias, for example coreos.com/CoreOS-stable-1409.8.0-hvm or leave it at 595879546273/CoreOS-stable-1409.8.0-hvm if you prefer to do so.

As part of our documentation, you will find a practical exercise using CoreOS with KOPS. See the file “coreos-kops-tests-multimaster.md” in the “examples” directory. This exercise covers not only using kops with CoreOS, but also a practical view of KOPS with a multi-master kubernetes setup.

Note: SSH username for CoreOS based instances will be core

Amazon Linux 2

Amazon Linux 2 support is still experimental, but should work. Please report any issues.

The following steps are known:

  • The latest Amazon Linux 2 AMI can be found using:
    1. aws ec2 describe-images --region=us-east-1 --owner=137112412989 \
    2.   --filters "Name=name,Values=amzn2-ami-hvm-2*-gp2" \
    3.   --query 'sort_by(Images,&CreationDate)[-1].{name:Name}'
  • You can specify the name using the amazon.com owner alias, for example amazon.com/amzn2-ami-hvm-2.0.20180622.1-x86_64-gp2

Be aware of the following limitations:

  • Amazon Linux 2 LTS is the recommended minimum version, a previous version called just “Amazon Linux AMI” is not supported.

Note: SSH username for Amazon Linux 2 based instances will be ec2-user