k8sflag

README

k8sflag

Flag-style bindings for Kubernetes ConfigMaps.

Rationale

There are a lot of ways to consume ConfigMap entries from a Pod--environment variables, volume mounts and direct API access. There is also a layer of code to write which brings the configuration values into the application code where it's needed. The k8sflag library provides a binding layer that brings ConfigMap values into a Golang application with a familiar, flag-style interface.

Example

kind: ConfigMap
data:
  name: "world"

var name = k8sflag.String("name", "nobody")
fmt.Printf("hello %v", name.Get())
--> hello world

Dynamic Option

Flags can be static or dynamically bound. A dynamic binding will be updated when the underlying ConfigMap is updates. This can be useful for flipping feature flags without redeploying the binary.

kind: ConfigMap
data:
  name: "world"

var name = k8sflag.String("name", "nobody", k8sflag.Dynamic)
fmt.Printf("hello %v", name.Get())
--> hello world

kind: ConfigMap
data:
  name: "mundo"  # changed

fmt.Printf("hello %v", name.Get())
--> hello mundo

Required Option

Flags can also be required. If a configuration value is not present, the flag will immediately panic.

var name = k8sflag.String("missing-property", "", k8sflag.Required)
--> PANIC

Try it out

$ kubectl create -f example/config.yaml
$ kubectl create -f example/pod.yaml

From a node in cluster:

node $ curl http://$POD_IP
--> hello world

Edit config.yaml and change hello.name property from world to mundo:

$ kubectl apply -f example/config.yaml

node $ curl http://$POD_IP
--> hello mundo

Implementation

The current implementation uses a volume mount to provide configuration because volume mounts are dynamically updated and don't require a Kubernetes client. But with a Kuberenetes client, the same interface can be implemented without volume mounting.