Switching Between Hot and Cold Data

The key is to allocate index data storage by node labels and index allocation policies.

• Node labels:
  • Data node (hot): Stores real-time data (hot data) by default and supports high-concurrency read/write requests.
  • Cold data node (cold): Stores historical data (cold data). Cold data nodes use less expensive hardware and deliver lower query performance than regular data nodes.
• Index allocation: You can configure index templates or directly specify index names to allocate specific indexes to data nodes or cold data nodes.

Procedure: Enable cold data nodes, and configure an index template or specific index settings to allocate index data storage. The cluster automatically allocates data based on your settings.

1. Check whether cold data nodes are enabled in the target cluster.
   1. Log in to the CSS management console.
   2. In the navigation pane on the left, choose Clusters > Elasticsearch.
   3. In the cluster list, click the name of the target cluster. The cluster information page is displayed.
   4. On the Overview tab, check whether the Node Information area contains cold data node information.
      • If there is information about cold data nodes, the cluster has cold data nodes. Go to the next step.
      • If not, the cluster does not have cold data nodes. Add cold data nodes for the cluster and then go to the next step. For details about how to add cold data nodes, see Adding New Node Types.

2. Log in to the Kibana console.
   1. On the cluster information page, click Kibana in the upper-right corner to log in to Kibana.
   2. In the left navigation pane, choose Dev Tools.

      The left part of the console is the command input box, and the triangle icon in its upper-right corner is the execution button. The right part shows the execution result.

3. On the Kibana page, configure index allocation policies.

   You can configure index templates or directly specify index names to allocate specific indexes to data nodes or cold data nodes.

   • Configuring an index template

     Configure an index template to allocate matching indexes to cold or hot data nodes. For example, run the following command to store newly created indexes whose name starts with myindex to cold data nodes. (The command varies slightly depending on the Elasticsearch version.)

     • For Elasticsearch 6.x or later:

       PUT _template/{template_name}
       {
         "order": 1,
         "index_patterns": "myindex*",
         "settings": {
           "refresh_interval": "30s",
           "number_of_shards": "3",
           "number_of_replicas": "0",
           "routing.allocation.require.box_type": "cold"
         }
       }
       
       

     • For Elasticsearch earlier than 6.x:

       PUT _template/{template_name}
       {
           "order": 1,
           "template": "myindex*",
           "settings": {
               "index": {
                   "refresh_interval": "30s",
                   "number_of_shards": "3",
                   "number_of_replicas": "0",
                   "routing.allocation.require.box_type": "cold"
               }
           }
       }
       
       

     Table 1 Key parameters

     Parameter	Type	Description
     index_patterns	String	Defines the index name matching pattern. The wildcard (*) can be used to match multiple indexes. For example, myindex* indicates all indexes whose name starts with myindex.
     routing.allocation.require.box_type	String	Specifies the data node type for shard allocation. The value can be: cold: Shards are allocated to cold data nodes. hot: Shards are allocated to regular data nodes. If this parameter is not specified, shards will be evenly and randomly distributed among cold and hot data nodes.

   • Configuring specific indexes

     To change the node type for an existing index, run the following command:

     PUT {index_name}/_settings   
     { 
             "index.routing.allocation.require.box_type": "cold"
     }
     
     

     Table 2 Parameter description

     Parameter	Type	Description
     index_name	String	Specifies one or more indexes. Single index: Enter the index name, for example, my_index. Multiple indexes: Enter multiple index names and use a comma (,) to separate them, for example, my_index1,my_index2. Wildcard: Use the wildcard (*) to match multiple indexes. For example, myindex* indicates all indexes whose name starts with myindex.
     routing.allocation.require.box_type	String	Specifies the data node type for shard allocation. The value can be: cold: Shards are allocated to cold data nodes. hot: Shards are allocated to regular data nodes. If this parameter is not specified, shards will be evenly and randomly distributed among cold and hot data nodes.

4. Verify the switchover between hot and cold storage.

   Run the following command to check the distribution of index shards:

   GET _cat/shards/{index_name}?v
   
   

   When the data volume is large, the switchover may take a long time. There may be an intermediate state where the data of an index resides on both cold data nodes and data nodes.

   As shown in the following figure, all shards of the myindex index are stored on the cold data node css-e668-ess-cold-esn-1-1.

   index    shard prirep state       docs  store ip             node
   myindex 1     p      STARTED 14085446 17.8gb 192.168.91.188 css-e668-ess- cold -esn-1-1
   myindex 2     p      STARTED 14094005 17.9gb 192.168.91.188 css-e668-ess- cold -esn-1-1
   myindex 0     p      STARTED 14094742 17.8gb 192.168.91.188 css-e668-ess- cold -esn-1-1
   
   

5. Roll back the cold-hot switchover configuration.

   To cancel the cold-hot switchover configuration, run the following command:

   PUT {index_name}/_settings    
   {
           "index.routing.allocation.require.box_type": null
   }
   
   

   After the rollback, index data will be evenly and randomly allocated to both cold data nodes and data nodes.

• You can scale cold data nodes by adding or reducing nodes or their storage capacity, or adding cold data nodes dynamically. For details, see Managing Cluster Changes.
• If your cluster does not have cold data nodes but you wish to cut storage costs, you can try using decoupled storage and compute. For details, see Configuring Decoupled Storage and Compute.