Skip to content

Latest commit

 

History

History
167 lines (130 loc) · 5.83 KB

manual-installation.md

File metadata and controls

167 lines (130 loc) · 5.83 KB
Raw

Manual installation

Install the Operator manually

Note: If you are running on Google Kubernetes Engine (GKE), please check this first.

Register the Bookkeeper cluster custom resource definition (CRD).

$ kubectl create -f deploy/crds/crd.yaml

Create the operator role, role binding and service account.

$ kubectl create -f deploy/role.yaml
$ kubectl create -f deploy/role_binding.yaml
$ kubectl create -f deploy/service_account.yaml

Install the operator.

$ kubectl create -f deploy/operator.yaml

Finally create a ConfigMap which contains the list of supported upgrade paths for the Bookkeeper cluster.

$ kubectl create -f deploy/version_map.yaml

Install the Operator in Test Mode

The operator can be deployed in test mode by providing the argument -test inside the operator.yaml file in the following way.

containers:
  - name: bookkeeper-operator
    image: pravega/bookkeeper-operator:latest
    ports:
    - containerPort: 60000
      name: metrics
    command:
    - bookkeeper-operator
    imagePullPolicy: Always
    args: [-test]

For more details check this

Install the Bookkeeper cluster manually

Note that the Bookkeeper cluster must be installed in the same namespace as the Zookeeper cluster.

If the BookKeeper cluster is expected to work with Pravega, we need to create a ConfigMap which needs to have the following values

KEY VALUE
PRAVEGA_CLUSTER_NAME Name of Pravega Cluster using this BookKeeper Cluster
WAIT_FOR Zookeeper URL

To create this ConfigMap.

$ kubectl create -f deploy/config_map.yaml

The name of this ConfigMap needs to be mentioned in the field envVars present in the BookKeeper Spec. For more details about this ConfigMap refer to this.

Once all these have been installed, you can use the following YAML template to install a small development Bookkeeper Cluster. Create a bookkeeper.yaml file with the following content.

apiVersion: "bookkeeper.pravega.io/v1alpha1"
kind: "BookkeeperCluster"
metadata:
  name: "pravega-bk"
spec:
  version: 0.7.0
  zookeeperUri: [ZOOKEEPER_HOST]:2181
  envVars: bk-config-map
  replicas: 3
  image:
    repository: pravega/bookkeeper
    pullPolicy: IfNotPresent

where:

  • [ZOOKEEPER_HOST] is the Zookeeper service endpoint of your Zookeeper deployment (e.g. zookeeper-client:2181). It expects the zookeeper service URL in the given format <service-name>:<port-number>

Check out other sample CR files in the example directory.

Deploy the Bookkeeper cluster.

$ kubectl create -f bookkeeper.yaml

Verify that the cluster instances and its components are being created.

$ kubectl get bk
NAME         VERSION   DESIRED MEMBERS    READY MEMBERS      AGE
pravega-bk   0.7.0     3                  0                  25s

Uninstall the Bookkeeper cluster manually

$ kubectl delete -f bookkeeper.yaml

Once the Bookkeeper cluster has been deleted, make sure to check that the zookeeper metadata has been cleaned up before proceeding with the deletion of the operator. This can be confirmed with the presence of the following log message in the operator logs.

zookeeper metadata deleted

However, if the operator fails to delete this metadata from zookeeper, you will instead find the following log message in the operator logs.

failed to cleanup <bookkeeper-cluster-name> metadata from zookeeper (znode path: /pravega/<pravega-cluster-name>): <error-msg>

The operator additionally sends out a ZKMETA_CLEANUP_ERROR event to notify the user about this failure. The user can check this event by doing kubectl get events. The following is the sample describe output of the event that is generated by the operator in such a case

Name:             ZKMETA_CLEANUP_ERROR-nn6sd
Namespace:        default
Labels:           app=bookkeeper-cluster
                  bookkeeper_cluster=pravega-bk
Annotations:      <none>
API Version:      v1
Event Time:       <nil>
First Timestamp:  2020-04-27T16:53:34Z
Involved Object:
  API Version:   app.k8s.io/v1beta1
  Kind:          Application
  Name:          bookkeeper-cluster
  Namespace:     default
Kind:            Event
Last Timestamp:  2020-04-27T16:53:34Z
Message:         failed to cleanup pravega-bk metadata from zookeeper (znode path: /pravega/pravega): failed to delete zookeeper znodes for (pravega-bk): failed to connect to zookeeper: lookup zookeeper-client on 10.100.200.2:53: no such host
Metadata:
  Creation Timestamp:  2020-04-27T16:53:34Z
  Generate Name:       ZKMETA_CLEANUP_ERROR-
  Resource Version:    864342
  Self Link:           /api/v1/namespaces/default/events/ZKMETA_CLEANUP_ERROR-nn6sd
  UID:                 5b4c3f80-36b5-43e6-b417-7992bc309218
Reason:                ZK Metadata Cleanup Failed
Reporting Component:   bookkeeper-operator
Reporting Instance:    bookkeeper-operator-6769886978-xsjx6
Source:
Type:    Error
Events:  <none>

In case the operator fails to delete the zookeeper metadata, the user is expected to manually delete the metadata from zookeeper prior to reinstall.

Uninstall the Operator manually

Note that the Bookkeeper clusters managed by the Bookkeeper operator will NOT be deleted even if the operator is uninstalled. To delete all clusters, delete all cluster CR objects before uninstalling the operator.

$ kubectl delete -f deploy