Terminating a Datacenter within an Existing DSE Cluster

DataStax Mission Control is current in Private Preview. It is subject to the beta agreement executed between you and DataStax. DataStax Mission Control is not intended for production use, has not been certified for production workloads, and might contain bugs and other functional issues. There is no guarantee that DataStax Mission Control will ever become generally available. DataStax Mission Control is provided on an “AS IS” basis, without warranty or indemnity of any kind.

If you are interested in trying out DataStax Mission Control please contact your DataStax account team.

Use DataStax Mission Control to terminate one or more datacenters. Do this from an existing cluster, one at a time. With multiple datacenters, select the order in which to terminate based on DataStax Enterprise (DSE) rules. Refer to the DSE upgrading planning guide detailing the upgrade priorities. For instance, terminate DSE Analytics datacenters first, taking into account whether the nodes use DSE Hadoop or Spark. Second is the termination of DSE Graph or transactional datacenters, followed by datacenters running DSE Search nodes.

DataStax Mission Control manages the system keyspaces replication during the termination process without altering user-defined keyspaces. For instance, if a datacenter selected for termination still has user keyspaces replicated to it, then DataStax Mission Control blocks the termination until the keyspaces are manually altered. This is a safety measure to prevent unintended removal of data.

Prerequisites

The kubectl CLI tool.
Kubeconfig file` or context pointing to a Control Plane Kubernetes cluster.

Example

Within an existing multi-datacenter Kubernetes cluster are two datacenters (DC1 in west region and DC2 in east region), each with 3 nodes. The decision is made to reduce costs and run from a single west region datacenter.

Workflow of user and operators

User updates the replication strategy on user-defined keyspaces to remove references on the east region datacenter being terminated. User must wait to update east region keyspaces in use until those keyspaces are dormant.

DataStax Mission Control operators automatically, and by default, update all the system keyspaces to terminate the east region datacenter and then terminate its nodes (pods).

DataStax Mission Control issues an error if there are keyspaces actively using the datacenter in the east region that is targeted for termination.

User submits the updated DSECluster to the Control Plane Kubernetes cluster.
Cluster-level operator picks up the modification and automatically update keyspace replication settings on system keyspaces.
Cluster-level operator deletes datacenter-level resources in the Kubernetes cluster where the nodes are to be terminated.

DC-level operator picks up datacenter-level resource changes and deletes native Kubernetes objects representing the DSE nodes.

If a user-defined keyspace is still replicating to the DC that is targeted for termination then the operation FAILS. By design all user-defined keyspaces MUST NOT reference the DC to be terminated.

Procedure

Modify the existing DSECluster YAML (demo-dse.yaml) in the Control Plane Kubernetes cluster, updating the spec.datacenters list so that it no longer references the datacenter dc1 targeted for termination. In this example, the following lines are deleted:

spec:
  datacenters:
  - metadata:
       name: dc1
    k8sContext: east
    size: 3
    racks:
    - name: rack1
      nodeAffinityLabels:
      topology.kubernetes.io/zone: us-east1-c
    - name: rack2
      nodeAffinityLabels:
        topology.kubernetes.io/zone: us-east1-b
    - name: rack3
      nodeAffinityLabels:
        topology.kubernetes.io/zone: us-east1-d

Submit this modification to the Kubernetes Control Plane cluster with the following command:
```
kubectl apply -f demo-dse.cassandratask.yaml
```
The following keyspaces are updated:
- system_traces
- system_distributed
- system_auth
- dse_leases
- dse_perf
- dse_security
Monitor the termination operation progress in the Control Plane cluster with this command:
```
kubectl get k8ssandracluster demo
```
In the event that any user-defined keyspaces are still replicating to datacenter targeted for termination, the kubectl command returns an error such as:
```
NAME   ERROR
demo   cannot decommission DC dc1: keyspace ks1 still has replicas on it
```
To rectify this error, the replication of all user-defined keyspaces must be manually updated to remove references to the datacenter being terminated.
```
ALTER KEYSPACE ks1 WITH replication = {'class': 'NetworkTopologyStrategy', 'west': 3};
```
Monitor the termination progress by checking the status of the datacenter to be terminated in the east cluster. This example uses the following command:
```
kubectl get cassandradatacenter dc1 -o yaml
```
Here is a partial sample of the results:
```
status:
  cassandraOperatorProgress: Updating
  conditions:
  - lastTransitionTime: "2022-10-24T02:43:20Z"
    message: ""
    reason: ""
    status: "True"
    type: Healthy
  - lastTransitionTime: "2022-10-24T02:43:20Z"
    message: ""
    reason: ""
    status: "False"
    type: Stopped
...
```
The sample output indicates that one DSE node is online and one is not at this point in the monitoring. The CassandraDatacenter dc1 is terminated when DC-level operators set all of the Decommission conditions:status to "False". The nodeStatuses map is also updated.

The DC-level operators must terminate each node (pod) in the datacenter before the datacenter itself is terminated.

Monitor the DSE logs with this command:

kubectl logs demo-dc1-rack3-sts-0 -c server-system-logger

where demo-dc1-rack3-sts-0 is the StatefulSet designation of the ordinal index of the node in a rack.

Sample output:

INFO  [pool-17-thread-1] 2022-10-27 17:13:09,717  StorageService.java:2143 - LEAVING: sleeping 30000 ms for batch processing and pending range setup
...
INFO  [pool-17-thread-1] 2022-10-27 17:13:39,770  Gossiper.java:1301 - InetAddress /10.100.5.15 is now DOWN
INFO  [pool-17-thread-1] 2022-10-27 17:13:39,788  StorageService.java:4968 - Announcing that I have left the ring for 30000ms

Upon successful completion of the east datacenter termination operation, users now run only in the west region datacenter.