Postgres-operator: Quickstart: How to access acid-minimal-cluster (Minikube) from host?

[hjacobs]

I just tried out the quickstart (introduced with #289), but I don't know how to access the new "acid-minimal-cluster" PostgreSQL cluster:

$ kubectl create -f manifests/configmap.yaml  # configuration
configmap "postgres-operator" created
$ kubectl create -f manifests/operator-service-account-rbac.yaml  # identity and permissions
serviceaccount "zalando-postgres-operator" created
clusterrole "zalando-postgres-operator" created
clusterrolebinding "zalando-postgres-operator" created
$ kubectl create -f manifests/postgres-operator.yaml  # deployment
deployment "postgres-operator" created
$ kubectl get pod
NAME                                 READY     STATUS    RESTARTS   AGE
postgres-operator-6474c766b4-tgp82   1/1       Running   0          8s
$ kubectl create -f manifests/minimal-postgres-manifest.yaml 
postgresql "acid-minimal-cluster" created
$ kubectl get pod
NAME                                 READY     STATUS    RESTARTS   AGE
acid-minimal-cluster-0               0/1       Pending   0          2s
postgres-operator-6474c766b4-tgp82   1/1       Running   0          17s
$ kubectl get pod
NAME                                 READY     STATUS    RESTARTS   AGE
acid-minimal-cluster-0               1/1       Running   0          20s
acid-minimal-cluster-1               1/1       Running   0          6s
postgres-operator-6474c766b4-tgp82   1/1       Running   0          35s
$ kubectl get svc
NAME                        TYPE           CLUSTER-IP       EXTERNAL-IP   PORT(S)          AGE
acid-minimal-cluster        LoadBalancer   10.102.211.199   <pending>     5432:32114/TCP   25s
acid-minimal-cluster-repl   ClusterIP      10.106.83.140    <none>        5432/TCP         25s
kubernetes                  ClusterIP      10.96.0.1        <none>        443/TCP          1m
$ minikube service acid-minimal-cluster
Waiting, endpoint for service is not ready yet...

The service will never become ready as it's of type LoadBalancer.

[Jan-M]

FYI: We are looking into this right now.

[sdudoladov]

@hjacobs thanks for finding this out !

Overall, this behavior seems to be minikube-specific and relates to the recent upgrades that made minikube use Kubernetes 1.10+

It should be fully resolved once a Spilo image with a Patroni >= v1.4.4 is out because of that:

Kubernetes 1.10.0+ started returning Endpoints.subsets set to None instead of [].

Kudos to @mgomezch for remembering this bug just in time !

There are other configuration issues here, but these I've already fixed and am currently preparing a PR

[sdudoladov]

@hjacobs Can you please confirm the updated quickstart works ? Just do

git clone https://github.com/zalando-incubator/postgres-operator.git
cd postgres-operator
git checkout --track origin/fix-conf-for-minikube

and then these instruction should work.

The service will never become ready as it's of type LoadBalancer.

Clarification: we actually need service of such type here for it to be accessible outside of minikube. Such services run fine on minikube.

[hjacobs]

@zerg-junior first observation, one might need to wait some seconds:

kubectl create -f manifests/minimal-postgres-manifest.yaml 
error: unable to recognize "manifests/minimal-postgres-manifest.yaml": no matches for acid.zalan.do/, Kind=postgresql

Apart from this minor problem (retrying works), the instructions are fine and I could successfully connect to the database :smile:

[sdudoladov]

one might need to wait some seconds:

Yes, that's true. The run_operator_locally script actually retries until timeout; I am just unsure how to indicate such behavior in quickstart beyond a simple comment.

And I totally overlooked all these issues just because I was assuming that if pods successfully start with Patroni, than everyhing is fine; had no idea that Patroni failed to create endpoints. More end-user testing for me in the future )))

[sdudoladov]

Addressed by #301

[zanoubia]

Not only minikube. I have the same problem with a three nodes cluster ! any ideas ?