Thoughts on Kubernetes

1 minute read

Just a collection of annoyances on Kubernetes:

kubectl config and multiple namespaces

Seriously! The shear amount of YAML duplication if you have a large number of namespaces and clusters it quite rapidly becomes very annoying, do remember that this is YAML so you could have (ab)used anchors to copy commonly used fragments:

    apiVersion: v1
    - cluster:
      name: cluster1
    - context: &bar-ns
        cluster: cluster1
        namespace: bar-namespace
        user: "cluster1-user"
      name: cluster1-bar-namespace
    - context:
        <<: *bar-ns
        namespace: foo-namespace
      name: cluster1-foo-namespace
    - context:
        <<: *bar-ns
        namespace: kube-system
      name: cluster1-kubesystem
    current-context: cluster1-foo-namespace
    kind: Config
    preferences: {}
    users: []
*         cluster1-foo-namespace   cluster1              foo-namespace
          cluster1-kubesystem      cluster1              kube-system
          cluster1-bar-namespace   cluster1              bar-namespace

Except: as soon as you use kubectl config use-context it goes and helpfully writes back the expanded YAML structure. This makes sense but is remarkably unhelpful as it goes.

You can of course template a number of ~/.kube/CONFIG_FILES and then use the following to switch:

$ export KUBECONFIG=~/.kube/config-a
$ kubectl get pods


Maybe it’s the rapid pace of change within the software, but it’s full of holes, and yes I should raise pull requests for changes to the documentation, but look at the CentOS instructions

Where the repo contains packages for Kubernetes 1.1.0 - from 2015!


Headless services without selectors are very helpful, but unfortunately a TTL of 30 seconds that is currently fixed can be limiting.