[BUG] Cannot make use of default_model_id in neural_sparse query type

[jdnvn]

What is the bug?

Receive an error when omitting the model_id param in a nested neural_sparse query after configuring a default_model_id for the index. This seems to only be for nested queries, I cannot reproduce it for non-nested queries.

How can one reproduce the bug?

Create an index called my_index

PUT /my_index

{
	"settings": {
		"index": {"knn": True},
		"number_of_shards": 1,
		"number_of_replicas": 1,
		"analysis": {
			"analyzer": {
				"default": {
					"type": "standard"
				}
			}
		}
	},
	"mappings": {
		"properties": {
			"id": {"type": "keyword"},
			"chunks": {
				"type": "nested",
				"properties": {
					"chunk_id": {"type": "keyword"},
					"chunked_content": {"type": "text"},
					"chunked_content_embedding": {"type": "rank_features"},
				}
			},
		}
	}
}

Update cluster settings

PUT /_cluster/settings
{
	"persistent": {
		"plugins": {
			"ml_commons": {
				"allow_registering_model_via_url": "true",
				"only_run_on_ml_node": "false",
				"model_access_control_enabled": "true",
				"native_memory_threshold": "99"
			}
		}
	}
}

Create the neural sparse model

POST /_plugins/_ml/model_groups/_register
{
	"name": "my_model_group",
	"description": "Models for search",
}

POST /_plugins/_ml/models/_register?deploy=true
{
	"name": "neural-sparse/opensearch-neural-sparse-encoding-v1",
	"version": "1.0.1",
	"model_group_id": <model_group_id>
	"description": "This is a neural sparse encoding model: It transfers text into sparse vector, and then extract nonzero index and value to entry and weights. It serves in both ingestion and search.",
	"model_format": "TORCH_SCRIPT",
	"function_name": "SPARSE_ENCODING",
	"model_content_size_in_bytes": 492184214,
	"model_content_hash_value": "d1ebaa26615090bdb0195a62b180afd2a8524c68c5d406a11ad787267f515ea8",
	"url": "https://artifacts.opensearch.org/models/ml-models/amazon/neural-sparse/opensearch-neural-sparse-encoding-v1/1.0.1/torch_script/neural-sparse_opensearch-neural-sparse-encoding-v1-1.0.1-torch_script.zip",
	"created_time": 1696913667239
}

Create the search pipeline with a neural query enricher

PUT /_search/pipeline/my_pipeline
{
	"request_processors": [
		{
			"neural_query_enricher": {
				"default_model_id": <model_id>
			}
		}
	]
}

Update the index settings with the default pipeline

PUT /my_index/_settings
{
  "index.search.default_pipeline" : "my_pipeline"
}

Search the index

POST /_search

{
	"query": {
		"function_score": {
			"query": {
				"bool": {
					"should": [
						{
							"nested": {
								"path": "chunks",
								"score_mode": "max",
								"query": {
									"bool": {
										"must": [
											{
												"match": {
													"chunks.chunked_content": {
														"query": "contract"
													}	
												}	
											},
											{
												"neural_sparse": {
													"chunks.chunked_content_embedding": {
														"query_text": "contract" # NO MODEL ID!
													}
												}
											}
										]
									}
								},
								"inner_hits": {
									"_source": [
										"chunks.chunk_id",
										"chunks.chunked_content"
									]
								}
							}
						}
					]
				}
			},
			"score_mode": "sum",
			"min_score": 0.0
		}
	},
	"_source": {
		"excludes": [
			"_index",
			"chunks.chunked_content_embedding",
			"chunks.chunked_content",
			"chunks.chunk_id"
		]
	},
	"size": 100,
	"explain": "true"
}

Receive error:

{

	"error": {
		"root_cause": [
			{
				"type": "illegal_argument_exception",
				"reason": "query_text and model_id cannot be null"
			}
		],
		"type": "illegal_argument_exception",
		"reason": "query_text and model_id cannot be null"
	},
	"status": 400
}

What is the expected behavior?

There is no error and the default_model_id configured on the search pipeline is used to embed the query.

What is your host/environment?

MacOS Sonoma 14.3, Docker 4.28.0, OpenSearch 2.16.0

Do you have any screenshots?

N/A

Do you have any additional context?

Might be due to this conditional not taking the default model id into account

[yuye-aws]

@zhichao-aws Do you know whether opensearch-project/OpenSearch#14739 can fix this?

[zhichao-aws]

Hi @jdnvn the root cause is the compound query visitor logics, and we have added the logics in nested query opensearch-project/OpenSearch#14739. However, function score query is also compound query but we haven't add the visitor logics in it yet https://github.com/opensearch-project/OpenSearch/blob/main/server/src/main/java/org/opensearch/index/query/functionscore/FunctionScoreQueryBuilder.java

This seems to only be for nested queries, I cannot reproduce it for non-nested queries.

Can you try this query? it should also throw exception

{
	"query": {
		"function_score": {
			"query": {
				"neural_sparse": {
                    "chunks.chunked_content_embedding": {
                        "query_text": "contract" # NO MODEL ID!
                    }
                }
			}
		}
	}
}

[jdnvn]

@yuye-aws @zhichao-aws thank you both for the quick responses! This makes sense, and I did get an exception from that query. I opened an issue in OpenSearch core and a PR to fix it. I'm going to close this as a duplicate.