A very technical thing that I don’t want to look up again.
If you work with kubernetes and want to expose information about the pod and the machine that the container is running on to the program inside the container you need to pass that information down. This is called the downward-api1 in kubernetes.
The idea behind kubernetes is that you don’t have to care about the underlying hardware, it is an abstraction on top of that. But in practice you want to spot if your failing pods all fail on the same node.
So how do you do it.
Your program lives inside a container. That container lives inside a pod. that pod is often deployed in a control structure such as a ‘deployment’ that lives inside a namespace.
namespace deployment/job/whatever pod container program
According to the documentation2 you can add an environmental variable like this
apiVersion: v1 kind: Pod metadata: name: dapi-envars-fieldref spec: containers: - name: test-container image: registry.k8s.io/busybox command: [ "sh", "-c"] args: - while true; do echo -en '\n'; printenv MY_NODE_NAME MY_POD_NAME MY_POD_NAMESPACE; printenv MY_POD_IP MY_POD_SERVICE_ACCOUNT; sleep 10; done; env: - name: MY_NODE_NAME valueFrom: fieldRef: fieldPath: spec.nodeName
Note that this is a description for a pod. So you have to modify this to make a deployment but when I created this in a deployment I couldn’t see the MY_NODE_NAME env var in the yaml of the pod. So I thought it wasn’t working.
But when you check the logs you see that the env var is correct.
My mistake was that I didn’t check the logs. A deployment doesn’t know where or when the pod will be created, only at runtime are the variables like nodename available.