How to expand Kubernetes node instance profiles created by kops? - amazon-web-services

I'm running a Kubernetes cluster on AWS, managed with kops. Now I want to run external-dns, which requires additional permissions in the nodes instance role. My question is: what is the best way to make these changes?
I could edit the role manually in AWS, but I want to automate my setup. I could also edit the role through the API (using the CLI, Cloudformation, Terraform, etc), but then I have a two-phase setup which seems fragmented and inelegant. Ideally I'd want to tell kops about my additional needs, and have it manage those with the ones it manages itself. Is there any way to do this?

Related

deriving re-useable terraform configuration from kops

I manage an established AWS ECS application with terraform. The terraform also manages all other aspects of each of 4 AWS environments, including VPCs, subnets, bastion hosts, RDS databases, security groups and so on.
We manage our 4 environments by putting all the common configuration in modules which are parameterised with variables derived from the environment specific terraform files.
Now, we are trying to migrate to using Kubernetes instead of Amazon ECS for container orchestration and I am trying to do this incrementally rather than with a big bang approach. In particular, I'd like to use terraform to provision the Kubernetes cluster and link it to the other AWS resources.
What I'd initially hoped to do was capture the terraform output from kops create cluster, generalise it by parameterising it with environment specific variables and then use this one kubernetes module across all 4 environments.
However, I now realise this isn't going to work because the k8s nodes and masters all reference the kops state bucket (in s3) and it seems like I am going to have clone that bucket and rewrite the files contained therein. This seems like a rather fragile way to manage the kubernetes environment - if I recreate the terraform environment, the related state kops state bucket is going to be inconsistent with the AWS environment.
It seems to me that kops generated terraform may be useful for managing a single instance of an environment, but it isn't easily applied to multiple environments - you effectively need one kops generated terraform per environment and there is noway to reuse the terraform to establish a new environment - for this you must fall back from a declarative approach and resort to an imperative kops create cluster command.
Am I missing a good way to manage the definition of multiple similar kubernetes environments with a single terraform module?
I'm not sure how you reached either conclusions, Terraform (which will cause more trouble than it will ever solve, that's definitely a tool to get rid of asap), or having to duplicate S3 buckets.
I'm pretty sure you'd be interested in kops's cluster template feature.
You won't need to generate, hack, and launch (and debug...) Terraform, and kops templates are just as easy if not significantly easier (and more specific...) to maintain than Terraform.
When kops releases new versions, you won't have to re-generate and re-hack your Terraform scripts either!
Hope this helps!

What exactly does EKS do if CloudFormation is needed?

What does AWS' Elastic Kubernetes Service (EKS) do exactly if so much configuration is needed in CloudFormation which is (yet) another AWS service?
I followed the AWS EKS Getting Started in the docs at (https://docs.aws.amazon.com/eks/latest/userguide/eks-ug.pdf) where it seems CloudFormation knowledge is heavily required to run EKS.
Am I mistaken or something?
So in addition to learning the Kubernetes .yaml manifest definitions, to run k8s on EKS, AWS expects you to learn their CloudFormation .yaml configuration manifests as well (which are all PascalCase as opposed to k8s' camelCase i might add)?
I understand that EKS does some management of latest version of k8s and control plane, and is "secure by default" but other than that?
Why wouldn't I just run k8s on AWS using kops then, and deal with the slightly outdated k8s versions?
Or am I supposed to do EKS + CloudFormation + kops at which point GKE looks like a really tempting alternative?
Update:
At this point I'm really thinking EKS is just a thin wrapper over CloudFormation after searching on EKS in detail and how it is so reliant on CloudFormation manifests.
Likely a business response to the alarming popularity of k8s, GKE in general with no substance to back the service.
Hopefully this helps save the time of anyone evaluating the half-baked service that is EKS.
To run Kubernetes on AWS you have basically 2 options:
using kops, it will create Master nodes + workers node under the hood, in plain EC2 machines
EKS + Cloudformation workers stack (you can use also Terraform as an alternative to deploy the workers, or eksctl, that will create both the EKS cluster and the workers. I recommend you to follow this workshop)
EKS alone provides only the master nodes of a kubernetes cluster, in a highly available setup. You still need to add the worker nodes, where your containers will be created.
I tried both kops and EKS + Workers, and I ended up using EKS, because I found it easier to setup and maintain and more fault-tolerant.
I feel the same difficulties earlier, and none of article could give me requirement in a glance for things that need to be done. Lot of people just recommend using eksctl which in my opinion will create a bloated and hard to manage kind of CloudFormation.
Basically both EKS is just a wrapper of Kubernetes, there's some points of integration between Kubernetes and AWS that still need to be done manually.
I've wrote an article that hope could help you understand all the process that need to be inplaces
EKS is the managed control plane for kubernetes , while Cloud-formation is a infrastructure templating service .
Instead of EKS you can run and manage the control plane(master nodes) on top of EC2 machines if you want to optimize for costs.For using EKS you have to pay for the underlying infra(EC2+networking..) and managed service fee(EKS price) .
Cloud-formation provides a nice interface to template and automate your infrastructure.You may use terraform in place of CF

kubernetes: kops and IAMFullAccess policy

According to documentation of both kops and aws, the dedicated kops user needs IAMFullAccess permission to operate properly.
Why is this permission needed?
Is there a way to avoid (i.e. restrict) this, given that it is a bit too intrusive to create a user with such a permission?
edit: one could assume that the specific permission is needed to attach the respective roles to the master(s) and node(s) instances;
therefore perhaps the question / challenge becomes how to:
not use IAMFullAccess
sync with the node creation / bootstrapping process and attach the above roles; (perhaps create a cluster on pre-configured instances? - no idea if kops provides for that)
As far as I understand kops design, it's meant to be end to end tool for provisioning you with k8s clusters. If you want to provision your nodes separately and deploy k8s on them I would suggest to use other tool, such as kubespray or kubeadm:
https://github.com/kubernetes-incubator/kubespray
https://kubernetes.io/docs/setup/independent/create-cluster-kubeadm/

The best way to add post configuration to an ECS Instance

I wonder what is the best way to add a post config step after instance creation when instance are automatically created by an ECS Cluster.
It seems there is no way to add user-data to ECS instance ?
Note : the instance are created automatically by the ECS Cluster itself.
EDIT:
When using ECS, you configure a Cluster. While configuring the cluster you select instance type and other stuff (ssh key, ...) but there is nowhere to give some user-data to the instances that will be created by ECS. So the question is how to do some post-configuration on instances automatically created with ECS.
When using the management console, it's more of a wizard that creates everything needed for you, including the instances using the Amazon Linux ECS optimized AMI, and doesn't give you a whole lot of control beyond that.
To get more fine-grained control, you would have to use another method of creating your cluster, such as the AWS CLI or CloudFormation. These methods allow you (or require you, actually) to create each piece at a time.
Example:
$ aws ecs create-cluster --cluster-name MyEcsCluster
The above command creates you a cluster, and cluster only. You would still have to create an ECS task definition, ECS service—although you could still use the management console for those—and (here's the real answer to your question) the EC2 instances which you want to attach to the cluster (either individually or through an Auto Scaling group). You could create instances from the Amazon Linux ECS optimized AMI, but also add user-data at that time to further configure them (you would also probably use the user-data in this scenario to create the /etc/ecs/ecs.config file to make sure it attaches to the ECS cluster you've created, e.g. echo "ECS_CLUSTER=MyEcsCluster" > /etc/ecs/ecs.config).
The short answer is, it's a more work to gain that sort of flexibility, but it is doable.
Edit: Thinking about it further, you could likely use the management console wizards to create everything once, then manually terminate the instances it created for the cluster (or, rather, delete the Auto Scaling group that creates them) and add your own. This would save you some work.

Boot strapping AWS auto scale instances

We are discussing at a client how to boot strap auto scale AWS instances. Essentially, a instance comes up with hardly anything on it. It has a generic startup script that asks somewhere "what am I supposed to do next?"
I'm thinking we can use amazon tags, and have the instance itself ask AWS using awscli tool set to find out it's role. This could give puppet info, environment info (dev/stage/prod for example) and so on. This should be doable with just the DescribeTags privilege. I'm facing resistance however.
I am looking for suggestions on how a fresh AWS instance can find out about it's own purpose, whether from AWS or perhaps from a service broker of some sort.
EC2 instances offer a feature called User Data meant to solve this problem. User Data executes a shell script to perform provisioning functions on new instances. A typical pattern is to use the User Data to download or clone a configuration management source repository, such as Chef, Puppet, or Ansible, and run it locally on the box to perform more complete provisioning.
As #e-j-brennan states, it's also common to prebundle an AMI that has already been provisioned. This approach is faster since no provisioning needs to happen at boot time, but is perhaps less flexible since the instance isn't customized.
You may also be interested in instance metadata, which exposes some data such as network details and tags via a URL path accessible only to the instance itself.
An instance doesn't have to come up with 'hardly anything on it' though. You can/should build your own custom AMI (Amazon machine image), with any and all software you need to have running on it, and when you need to auto-scale an instance, you boot it from the AMI you previously created and saved.
http://docs.aws.amazon.com/gettingstarted/latest/wah-linux/getting-started-create-custom-ami.html
I would recommend to use AWS Beanstalk for creating specific instances, this makes it easier since it will create the AutoScaling groups and Launch Configurations (Bootup code) which you can edit later. Also you only pay for EC2 instances and you can manage most of the things from Beanstalk console.