envFrom set by Service Binding Operator is overwritten by odo when there are automounting ConfigMaps or Secrets

[rm3l]

/kind bug

Originally posted by @rm3l in #6698 (review)

I quickly wanted to check how this would behave with Service Binding, which, I remembered might also alter the Pod spec to inject envFrom using a secret:

odo init --devfile nodejs --starter nodejs-starter --name automount-volumes-nodejs
odo add binding --service edb-cloud-native-postgresql.Cluster.postgresql.k8s.enterprisedb.io \
  --name automount-volumes-nodejs-edb-cloud-native-postgresql \
  --service-namespace shared-services \
  --bind-as-files=false

If I don't have any automounting configmap/secret, it works as expected:

$ kubectl get pod automount-volumes-nodejs-app-68d4c554b-kldjk -o yaml
[...]
    envFrom:                                                                                                                                                                                                        
    - secretRef:                                                                                                                                                                                                    
        name: automount-volumes-nodejs-edb-cloud-native-postgresql-fb57c24f

$ kubectl exec -it automount-volumes-nodejs-app-68d4c554b-kldjk -- /bin/sh -c 'env | grep CLUSTER | sort'
Alias tip: k exec -it automount-volumes-nodejs-app-68d4c554b-kldjk -- /bin/sh -c 'env | grep CLUSTER | sort'
CLUSTER_CA.CRT=-----BEGIN CERTIFICATE-----
CLUSTER_CA.KEY=-----BEGIN EC PRIVATE KEY-----
CLUSTER_CLUSTERIP=10.96.151.42
CLUSTER_DATABASE=app
CLUSTER_HOST=edb-cloud-native-postgresql-rw
CLUSTER_PASSWORD=3IxAlTt9AUeVNvPS1TVrptKCUxUMXWIkE37cQgq3orNfZcbKHX4JJOHV8pqD0xK6
CLUSTER_PGPASS=edb-cloud-native-postgresql-rw:5432:*:postgres:3IxAlTt9AUeVNvPS1TVrptKCUxUMXWIkE37cQgq3orNfZcbKHX4JJOHV8pqD0xK6
CLUSTER_PROVIDER=enterprisedb
CLUSTER_TLS.CRT=-----BEGIN CERTIFICATE-----
CLUSTER_TLS.KEY=-----BEGIN EC PRIVATE KEY-----
CLUSTER_TYPE=postgresql
CLUSTER_USERNAME=postgres

If I stop the previous Dev session and apply some automounting configmap/secret (I applied the resources you used in the tests of this PR), it looks like the environment variables expected from the ServiceBinding are no longer injected.

$ kubectl apply -f /path/to/odo/tests/examples/manifests/config-automount/
$ odo dev
$ kubectl get pod automount-volumes-nodejs-app-67f467566c-gjwxx -o yaml
[...]
    envFrom:                                                                                                                                                                                                        
    - secretRef:                                                                                                                                                                                                    
        name: automount-env-secret                                                                                                                                                                                  
    - configMapRef:                                                                                                                                                                                                 
        name: automount-env-configmap 

$ kubectl exec -it automount-volumes-nodejs-app-67f467566c-gjwxx -- /bin/sh -c 'env | grep CLUSTER | sort' 
$ 

---

Comment from @feloy (#6698 (comment))

I'm pretty sure the service binding operator is using an Update command to set the envfrom field. Because of using Update and not Apply, the values set by the operator are overwritten by odo.

$ kubectl get deployments.apps automount-volumes-nodejs-app -o yaml --show-managed-fields
[...]
  managedFields:
  - apiVersion: apps/v1
    fieldsType: FieldsV1
    fieldsV1:
      f:spec:
        f:template:
          f:spec:
            f:containers:
              k:{"name":"runtime"}:
                f:envFrom: {}
    manager: manager
    operation: Update
    time: "2023-04-17T09:53:51Z"

[github-actions]

A friendly reminder that this issue had no activity for 90 days. Stale issues will be closed after an additional 30 days of inactivity.

[github-actions]

This issue was closed because it has been inactive for 30 days since being marked as stale.

[rm3l]

/priority Low