[BUG] Index setting partition size is ignored if routing num shard setting is not specified, leading to ingestion to single shard

[MilanCoder]

Describe the bug

Hi,

We came up with an issue in opensearch 2.11 where while ingesting data into an index, all the data was getting routed to single shard even if we specify the partition size to more than 1 as value, it seems the data was only getting distributed to multiple shards if we specify the routing num shards setting value to equal to num of shards.

Setting used without routing num shards setting :

` "settings": {
      "number_of_shards": 20,
      "index.routing_partition_size": 3,
      "number_of_replicas": 0,
      "refresh_interval": "300s"
  },`

Setting used with routing num shards setting which works as expected.

 "settings": {
      "number_of_shards": 20,
      "index.routing_partition_size": 3,
      "number_of_replicas": 0,
      "refresh_interval": "300s",
      "number_of_routing_shards":20
  },

Related component

No response

To Reproduce

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior

It should work without specifying routing num shards setting. The data should get distributed across shards.

Additional Details

Plugins
Please list all plugins currently enabled.

Screenshots
If applicable, add screenshots to help explain your problem.

Host/Environment (please complete the following information):

• OS: [e.g. iOS]
• Version [e.g. 22]

Additional context
Add any other context about the problem here.

[shwetathareja]

index.routing_partition_size setting is used to control the no. of shards, custom routing can be assigned to instead of 1 shard by default.

It is not mandatory to specify number_of_routing_shards and by default, it uses number_of_shards.

OpenSearch/server/src/main/java/org/opensearch/cluster/metadata/IndexMetadata.java

Lines 284 to 287 in 9bef705

	public static final Setting<Integer> INDEX_NUMBER_OF_ROUTING_SHARDS_SETTING = Setting.intSetting(
	"index.number_of_routing_shards",
	INDEX_NUMBER_OF_SHARDS_SETTING,
	1,

Can you share exact request you are using to create the index and share the bulk request for ingestion as well.

[vikasvb90]

Setting used with routing num shards setting which works as expected.

In this section, values of both number of shards and routing shards are same which means that routing factor is 1. Meaning, that further shrink or split cannot be done on this index. By doing this we are essentially disabling this feature and due to which you are not seeing this issue.

[gaobinlong]

This bug exists for a very long time since the introduction of the setting number_of_routing_shards and then the change of its default value, that change breaks the computation of shard id when routing_partition_size is greater than 1, for now the default value of number_of_routing_shards is the largest number which is less than 1024 and is obtained by continuous multiplication number_of_shards by 2, see

OpenSearch/server/src/main/java/org/opensearch/cluster/metadata/MetadataCreateIndexService.java

Line 1727 in e306d51

public static int calculateNumRoutingShards(int numShards, Version indexVersionCreated) {

. We can check the default value of number_of_routing_shards by calling GET cluster state API:

GET _cluster/state/metadata

 "indices": {
      "test-1": {
        "version": 359,
        "mapping_version": 1,
        "settings_version": 1,
        "aliases_version": 2,
        "routing_num_shards": 100,
....

For this case, the workaround is to set number_of_routing_shards to number_of_shards explicitly, that's the only way that routing_partition_size which is greater than 1 can work well, but the bad effect is that the index cannot be split anymore. If we want to fix this, one option is to set number_of_routing_shards to number_of_shards if routing_partition_size is greater than 1 and forbid setting number_of_routing_shards to other values at index creation time.

[krisfreedain]

[Catch All Triage - 1, 2]

[skimbrel-figma]

I just encountered this in 2.17 after thinking I was totally losing my mind trying to set up routing partitions for the first time. Please prioritize fixing this implementation, or at least document the fact that the configuration does not do what it says on its own!