Does kubectl replace not update environment variables? - environment-variables

Question

I am adding an environment variable to a Kubernetes replication controller spec, but when I update the running RC from the spec, the environment variable isn't added to it. How come?
I update the RC according to the following spec, where the environment variable IRON_PASSWORD gets added since the previous revision, but the running RC isn't updated correspondingly, kubectl replace -f docker/podspecs/web-controller.yaml:
apiVersion: v1
kind: ReplicationController
metadata:
name: web
labels:
app: web
spec:
replicas: 1
selector:
app: web
template:
metadata:
labels:
app: web
spec:
containers:
- name: web
image: quay.io/aknuds1/muzhack
# Always pull latest version of image
imagePullPolicy: Always
env:
- name: APP_URI
value: https://staging.muzhack.com
- name: IRON_PASSWORD
value: password
ports:
- name: http-server
containerPort: 80
imagePullSecrets:
- name: quay.io
After updating the RC according to the spec, it looks like this (kubectl get pod web-scpc3 -o yaml), notice that IRON_PASSWORD is missing:
apiVersion: v1
kind: Pod
metadata:
annotations:
kubernetes.io/created-by: |
{"kind":"SerializedReference","apiVersion":"v1","reference":{"kind":"ReplicationController","namespace":"default","name":"web","uid":"c1c4185f-0867-11e6-b557-42010af000f7","apiVersion":"v1","resourceVersion":"17714"}}
kubernetes.io/limit-ranger: 'LimitRanger plugin set: cpu request for container
web'
creationTimestamp: 2016-04-22T08:54:00Z
generateName: web-
labels:
app: web
name: web-scpc3
namespace: default
resourceVersion: "17844"
selfLink: /api/v1/namespaces/default/pods/web-scpc3
uid: c1c5035f-0867-11e6-b557-42010af000f7
spec:
containers:
- env:
- name: APP_URI
value: https://staging.muzhack.com
image: quay.io/aknuds1/muzhack
imagePullPolicy: Always
name: web
ports:
- containerPort: 80
name: http-server
protocol: TCP
resources:
requests:
cpu: 100m
terminationMessagePath: /dev/termination-log
volumeMounts:
- mountPath: /var/run/secrets/kubernetes.io/serviceaccount
name: default-token-vfutp
readOnly: true
dnsPolicy: ClusterFirst
imagePullSecrets:
- name: quay.io
nodeName: gke-staging-default-pool-f98acf11-ba7d
restartPolicy: Always
securityContext: {}
serviceAccount: default
serviceAccountName: default
terminationGracePeriodSeconds: 30
volumes:
- name: default-token-vfutp
secret:
secretName: default-token-vfutp
status:
conditions:
- lastProbeTime: null
lastTransitionTime: 2016-04-22T09:00:49Z
message: 'containers with unready status: [web]'
reason: ContainersNotReady
status: "False"
type: Ready
containerStatuses:
- containerID: docker://dae22acb9f236433389ac0c51b730423ef9159d0c0e12770a322c70201fb7e2a
image: quay.io/aknuds1/muzhack
imageID: docker://8fef42c3eba5abe59c853e9ba811b3e9f10617a257396f48e564e3206e0e1103
lastState:
terminated:
containerID: docker://dae22acb9f236433389ac0c51b730423ef9159d0c0e12770a322c70201fb7e2a
exitCode: 1
finishedAt: 2016-04-22T09:00:48Z
reason: Error
startedAt: 2016-04-22T09:00:46Z
name: web
ready: false
restartCount: 6
state:
waiting:
message: Back-off 5m0s restarting failed container=web pod=web-scpc3_default(c1c5035f-0867-11e6-b557-42010af000f7)
reason: CrashLoopBackOff
hostIP: 10.132.0.3
phase: Running
podIP: 10.32.0.3
startTime: 2016-04-22T08:54:00Z

Answer 1

Replacing the ReplicationController object does not actually recreate the underlying pods, so the pods keep the spec from the previous configuration of the RC until they need to be recreated. If you delete the running pod, the new one that gets created to replace it will have the new environment variable.
This is what the kubectl rolling update command is for, and a part of the reason why the Deployment type was added to Kubernetes 1.2.