[BUG] Embeddings pipeline with remote model using AWS Bedrock Connector throws error when indexing or simulate

[t49tran]

What is the bug?

I am using OpenSearch 2.19 / 2.18.

A connector to AWS Amazon Bedrock has been created with the below configuration, following the blueprint here.

After creation, when indexing data or simulate the pipeline, an error is thrown, indicating some issues with data mapping between the connector and the processor.

The error:

{
  "docs": [
    {
      "error": {
        "root_cause": [
          {
            "type": "status_exception",
            "reason": """Error validating input schema: Validation failed: [$: required property 'parameters' not found] for instance: {"algorithm":"REMOTE","text_docs":["Orange table"],"return_bytes":false,"return_number":true,"target_response":["sentence_embedding"]} with schema: {
    "type": "object",
    "properties": {
        "parameters": {
            "type": "object",
            "properties": {
                "inputText": {
                    "type": "string"
                },
                "inputImage": {
                    "type": "string"
                }
            }
        }
    },
    "required": [
        "parameters"
    ]
}"""
          }
        ],
        "type": "status_exception",
        "reason": """Error validating input schema: Validation failed: [$: required property 'parameters' not found] for instance: {"algorithm":"REMOTE","text_docs":["Orange table"],"return_bytes":false,"return_number":true,"target_response":["sentence_embedding"]} with schema: {
    "type": "object",
    "properties": {
        "parameters": {
            "type": "object",
            "properties": {
                "inputText": {
                    "type": "string"
                },
                "inputImage": {
                    "type": "string"
                }
            }
        }
    },
    "required": [
        "parameters"
    ]
}"""
      }
    }
  ]
}

How can one reproduce the bug?

POST /_plugins/_ml/connectors/_create
{
  "name": "Amazon Bedrock Connector: multi-modal embedding",
  "description": "The connector to bedrock Titan multi-modal embedding model",
  "version": 1,
  "protocol": "aws_sigv4",
  "parameters": {
    "region": "",
    "service_name": "bedrock",
    "model": "amazon.titan-embed-image-v1",
    "input_docs_processed_step_size": 2
  },
  "credential": {
    "access_key": "",
    "secret_key": "",
    "session_token": ""
  },
  "actions": [
    {
      "action_type": "predict",
      "method": "POST",
      "url": "https://bedrock-runtime.${parameters.region}.amazonaws.com/model/${parameters.model}/invoke",
      "headers": {
        "content-type": "application/json",
        "x-amz-content-sha256": "required"
      },
      "request_body": "{\"inputText\": \"${parameters.inputText:-null}\", \"inputImage\": \"${parameters.inputImage:-null}\"}",
      "pre_process_function": "connector.pre_process.bedrock.multimodal_embedding",
      "post_process_function": "connector.post_process.bedrock.embedding"
    }
  ]
}

This connector is used to create a model for a pipeline with text embedding processor. Pipeline:

PUT /_ingest/pipeline/aws-multimodal-pipeline
{
  "description": "A text/image using aws bedrock embedding pipeline",
  "processors": [
    {
      "text_image_embedding": {
        "model_id": "TzMSQpUBEsYL3wz3lnAk",
        "embedding": "vector_embedding",
        "field_map": {
          "text": "description",
          "image": "image_binary"
        }
      }
    }
  ]
}

An index was then created:

PUT /doc-multimodal
{
  "settings": {
    "index.knn": true,
    "default_pipeline": "aws-multimodal-pipeline",
    "number_of_shards": 2
  },
  "mappings": {
    "properties": {
      "vector_embedding": {
        "type": "knn_vector",
        "dimension": 1024,
        "method": {
          "name": "hnsw",
          "engine": "lucene",
          "parameters": {}
        }
      },
      "description": {
        "type": "text"
      },
      "image_binary": {
        "type": "binary"
      }
    }
  }
}

Pipeline simulation request:

POST /_ingest/pipeline/aws-multimodal-pipeline/_simulate
{
  "docs": [
    {
      "_index": "testindex1",
      "_id": "1",
      "_source":{
         "description": "Orange table"
      }
    }
  ]
}

What is the expected behavior?

The simulation/pipeline should run successfully.

What is your host/environment?

Ubuntu 24.4, Opensearch docker image 2.18 and 2.19

[weijia-aws]

Hi @t49tran, I just followed your steps, but I didn't receive such error, the simulate and ingest API worked for me. I'm using the Opensearch docker image 2.19 on Mac OS 15.3.1.

Also I don't see your model registration step, maybe something went wrong such as incorrect connector id?

[t49tran]

Hi @weijia-aws, thanks for looking into this. I have encountered this problem in 2.18 and at first updated to 2.19 didn't resolve it.

I ran this on local ubuntu.

After a clean installation with 2.19 the problem has been resolved, maybe I made some mistakes before.

I think this issue can now be closed.