Remove a Node and Rebalance

Couchbase Server allows a cluster node to be removed, and the remaining nodes rebalanced.

Understanding Removal and Rebalance

The rebalance operation distributes active and replica vBuckets across available cluster nodes in optimal fashion. This allows the best possible data-availability to be maintained after nodes have been added or removed. Examples of using rebalance after node-addition have already been provided, in Add a Node and Rebalance and Join a Cluster and Rebalance. In particular, reference was made in both locations to the data initially resident on a single node being replicated and distributed, across two nodes, following node-addition.

Nodes can also be rebalanced following removal: this is demonstrated in the current section. When a node is flagged for removal (as opposed to failover), new replica vBuckets will be created on the remaining nodes, as node-removal occurs. This has the effect of reducing available memory for the specified number of replicas.

Examples on This Page

The examples in the subsections below show how to remove the same node from the same two-node cluster; using the UI, the CLI, and the REST API respectively. The examples assume:

  • A two-node cluster already exists; as at the conclusion of List Cluster Nodes.

  • The cluster has the Full Administrator username of Administrator, and password of password.

Remove a Node with the UI

Proceed as follows:

  1. Access the Servers screen of Couchbase Web Console, by means of the Servers tab in the left-hand navigation bar. The screen appears as follows:

    twoNodeClusterAfterRebalanceCompressedView
  2. Left-click on the row for node 10.142.181.102. The row expands vertically, as follows:

    twoNodeClusterAfterRebalance
  3. To initiate removal, left-click on the Remove button, at the lower left of the row:

    removeButton

    The Confirm Server Removal dialog appears:

    confirmServerRemoval

    Left-click on the Remove Server confirmation button. The Servers screen reappears as follows:

    twoNodeClusterFollowingRemoval

    This indicates that node 10.142.181.102 has been flagged for removal, and is still taking traffic. A rebalance must be performed to complete removal.

  4. Left-click on the Rebalance button, at the upper right:

    rebalanceButton

    Rebalancing now occurs. A progress bar is provided towards the top, summarizing overall progress. Each row also has its own progress bar, indicating the workload accomplished by each node:

    rebalanceProgressBarFollowingRemoval

    Following the rebalance, the Servers screen confirms that a single node remains:

    singleNodeClusterBeforeAddNode

    Note that all 31.1 K of Items (from the travel-sample bucket) are again solely located on 10.142.181.101, with no replicas (since at least two nodes are required for replication to occur).

Remove a Node with the CLI

To remove a node from a cluster and perform the necessary rebalance, use the rebalance command with the --server-remove option:

couchbase-cli rebalance -c 10.142.181.102:8091 \
--username Administrator \
--password password --server-remove 10.142.180.102:8091

This initiates the rebalance process. As it continues, progress is shown as console output:

Rebalancing
Bucket: 01/01 (travel-sample)             0 docs remaining
[================================                 ] 31.67%

For more information, see the command reference for rebalance.

Remove a Node with the REST API

To remove a node from a cluster with the REST API, and rebalance the remaining nodes, use the /controller/rebalance URI. This requires that all known nodes be specified, and that the nodes to be ejected also be specified:

curl  -u Administrator:password -v -X POST \
http://10.142.181.101:8091/controller/rebalance \
-d 'ejectedNodes=ns_1%4010.142.181.102' \
-d 'knownNodes=ns_1%4010.142.181.101%2Cns_1%4010.142.181.102'

The command returns no output.

Next Steps

Nodes can be failed over, so that an unhealthy or responsive node can be removed from the cluster without application-access being affected. See Fail Nodes Over.