Enhancing the Data Import Performance of OpenSearch Clusters

Overview

Developed by CSS, data import performance enhancement effectively improves data import performance and reduces write rejections through bulk route optimization, text index acceleration, and word segmentation acceleration. You are advised to use this feature for clusters that contain a large number of index shards and text indexes, or have a high import throughput.

Constraints

Only Elasticsearch 7.10.2 and OpenSearch 1.3.6 clusters support data import performance enhancement.

Prerequisites

The cluster whose data import performance you want to enhance is in the Available state.

Bulk Route Optimization

You can specify the index.bulk_routing configuration item to enable bulk route optimization. You can use it to reduce the number of requests that need to be internally forwarded. For a cluster that has a large number of shards, this setting can improve write performance and reduce bulk rejection.

After bulk route optimization is enabled (that is, index.bulk_routing is set to pack or ocal_pack), data writing is no longer routed based on _id, and routing-related functionality may be affected. For example, ID-based GET requests may fail.

1. Choose Clusters in the navigation pane. On the Clusters page, select an available cluster, and click Access Kibana in the Operation column.
2. In the navigation tree on the left on the Kibana console, choose Dev Tools.
3. On the Dev Tools page, run the following command to enable bulk route optimization:

   PUT my_index 
   {
     "settings": { 
       "index.bulk_routing": "local_pack"
     } 
   }

   The options for the index.bulk_routing configuration item are as follows:

   • default: Use the default routing mechanism. Records in a bulk request are split and then routed separately.
   • pack: All data of a single bulk request is randomly routed to the same shard.
   • ocal_pack: The data of a single bulk request is routed to a local shard of the data node that has received the bulk request. If the node does not contain the corresponding index shard, the data is randomly routed to another node that contains the index shard. This solution depends on the random distribution of client bulk requests and the balanced distribution of primary shards.

Bulk Aggregation Optimization

You can specify the index.aggr_perf_batch_size configuration item to enable or disable bulk aggregation optimization. When bulk aggregation optimization is enabled, documents in a bulk request are written in batches, rather than individually. This helps to reduce the overheads of memory requests, lock requests, and other calls, improving data import performance.

1. Choose Clusters in the navigation pane. On the Clusters page, select an available cluster, and click Access Kibana in the Operation column.
2. In the navigation tree on the left on the Kibana console, choose Dev Tools.
3. On the Dev Tools page, run the following command to enable bulk aggregation optimization:

   PUT my_index 
   {
     "settings": { 
       "index.aggr_perf_batch_size": "128"
     } 
   }

   The value range of the index.aggr_perf_batch_size configuration item is [1, Integer.MAX_VALUE]. The default value is 1, indicating that bulk aggregation optimization is disabled. If the value is greater than 1, bulk aggregation optimization is enabled, the value of MIN(bulk_doc_size, aggr_perf_batch_size) indicates the bulk size.

Text Index Acceleration

You can configure index.native_speed_up to enable or disable text index acceleration. This setting optimizes the indexing process and memory access to accelerate index building for text fields (text and keyword). When text index acceleration is enabled, you can configure index.native_analyzer to enable word segmentation acceleration as well. For texts that need common word segmentation, you can use the analyzer to accelerate word segmentation.

• Word segmentation acceleration can be enabled (set index.native_analyzer to true) only when text index acceleration is enabled (set index.native_speed_up to true). Otherwise, word segmentation acceleration will not take effect.
• Text index acceleration cannot be enabled for indexes that contain nested fields.

1. Choose Clusters in the navigation pane. On the Clusters page, select an available cluster, and click Access Kibana in the Operation column.
2. In the navigation tree on the left on the Kibana console, choose Dev Tools.
3. On the Dev Tools page, run the following command to enable text index acceleration:

   PUT my_index 
   {
     "settings": {
       "index.native_speed_up": true,
       "index.native_analyzer": true
     }
   }

   The values of index.native_speed_up and index.native_analyzer are true or false. The default value is false.

Index Merge Task Optimization

Enabling the three optimization options above will increase the number of index merge tasks in the cluster. You can configure index.merge.scheduler.max_thread_count to reduce the impact of index merge tasks on data import performance. The optimization is about increasing the number of shard merging threads and thereby alleviate the throttling of data import.

1. Choose Clusters in the navigation pane. On the Clusters page, select an available cluster, and click Access Kibana in the Operation column.
2. In the navigation tree on the left on the Kibana console, choose Dev Tools.
3. On the Dev Tools page, run the following command to start index merge task optimization:

   PUT my_index 
   {
     "settings": {
       "index.merge.scheduler.max_thread_count": 8
     }
   }

   The value range of index.merge.scheduler.max_thread_count is [1, node.processors/2]. The default value is 4, and the recommended value is 8.