caos/orbos

ORBOS - GitOps everything

orbos's Issues

Insert validating phase after adapt phase and before ensuring

Only migrate on api version changes

Structured secrets.yml

myorbmycluster_kubeconfig:
  encoding: Base64
  encryption: AES256
  value: xyz
myorbmyclustermyprovider_bootstrapkey:
  encoding: Base64
  encryption: AES256
  value: xyz
myorbmyclustermyprovider_maintenancekey:
  encoding: Base64
  encryption: AES256
  value: xyz
myorbmyclustermyprovider_maintenancekey_pub:
  encoding: Base64
  encryption: AES256
  value: xyz

becomes:

kind: orbiter.caos.ch/Secrets
version: v0
spec:
  myorb:
    mycluster:
      kubeconfig:
        encoding: Base64
        encryption: AES256
        value: xyz
      myprovider:
        bootstrapkey:
          encoding: Base64
          encryption: AES256
          value: xyz
        maintenancekey:
          encoding: Base64
          encryption: AES256
          value: xyz
        maintenancekey_pub:
          encoding: Base64
          encryption: AES256
          value: xyz

Chaos Monkey

Recreate infrop issues

Consistent Terminology: Remove

=> ubiquitous language

Node Agent API Communication

Selinux integration for orbiter and boom

Remove deps properties

Load Balancer and VIP Observability

Upgrade to Helm v3

Upgrade from helm v2 to helm v3, for the aspects of using it tillerless and having the positive aspects of rollbacks per release.

The state should then be persistently managed in a git repository.

Too many boom-tokens

I think with each update of the Service Account we create a new boom-token.

This creates a lot of secrets and should be fixed 😁

Pass Orbiter version in Node Agents desired state

Software is not downgradeable

yum install kubelet=1.16.4 does not overwrite installed package kubelet=1.17.0

Only install nginx for pools that serve load balancing

Move specs from map[string]interface{} to interface{}

Dashboard for network plugins

we should provide a dashboard for calico in the first place

We should test security best practices in k8s

I think it would be best to remediate as much as possible in an automated manor.
So we could run something like kube-bench in our test pipeline and then remediate most of the issues.

Target should be that the kubeadm deployment is hardened automatically without a lot of customization from customers.

@thesephirot @eliobischof @stebenz inputs?

Create additional flows

We should also collect log files of k8s and our tooling

• Kube-System (API Server, Controller Manager, Etcd, kube-proxy)
• Orbiter
• Boom
• Kubelet logs (systemd on node)
• Node-Agent logs (systemd on node)

@eliobischof additional inputs?

Theme/Design Devops-Brand

Rename "compute" to "machine"

Stop trying to parse null maintenancekeys

Replace ignite as it doesn't support rebooting

Refactor docs

• Create better to use guides regarding k8s
• Document the API

Orbiter should no interfere with cordon nodes

Orbiter removes the cordon flag from a specific node in each iteration.

IMHO this should not behave like this 😁

Reenable concurrency

Stop Keepalived and Nginx when destroying node

orbctl shall have a function "list secrets"

to ease the download of a kubeconfig, orbctl shall have the option to list the secrets in its config.

Use envoy for load balancing

Orbiter 0.12.5 is not reconciling boom

In our testcluster orbiter should apply boom 0.9.13 as defined in the orbiter.yml.
However we still have 0.9.10 which was the last version applied.

This might be the root cause for caos/boom#48

Side note this does not affect orbiter version 0.12.2

Dashboards:

(quickwin)
ambassador
cert-manager
argo

(the long run)
orbiter
boom

Improve Resilience

• PodDisruptionPolicies
• Taints
• Node Selectors
• Pod Priority and Preemption

Stop managing node agent installations in clusters

orbctl ps, orbctl exec

Change sshd listen address to node IP

Cache Images for used helm charts

To better support boom, we should cache and security scan the used images and substitute them.

FirewallD is not running

Publish development artifacts

The on node Loadbalancer should use proxy protocol for worker nodes

Implement GCE Provider with LB

aws provider implementation

AWS can be used as Static Provider to test static implementations @ customer sites.

In addition we will implement full AWS functionality.

Refactor Pipeline

I think we can split test and release & push from each other

Automate end-to-end testing

Add comments to API structs

Add comments to API structs so that crd definition has descriptions

Deploy boom

Understand Mission Control files

We should document our security defaults and processes

Master node not properly setup with role `master`

Sometimes when setting up a 3 node master cluster, the third node does not have the proper role master applied in kubectl get nodes. Also the taints are missing.

@eliobischof a customer project is related to this.

Change takeoff command and implement a destroy function

To make cluster lifecycle easy we should implement a destroy flag and change the behaviour off takeoff.

This relates to a customer issue.

Proxy Protocol Support

Prio 1

• Proxy Protocol Support between NGINX and Ambassador

Prio 2

• Proxy Protocol Support between NGINX and externalLB

Additional Infos

https://www.getambassador.io/reference/core/ambassador/

This links to #18