Troubleshooting

19/10/2026

GraphRAG 1.3.0

This section covers how to address issues related to:

an unsupported or invalid embedding model.
a mismatch between the embedding model's dimensions and those configured in the vector store.
empty results returned from the Components service.

In this case a status 400 will be returned. The response may look like this:

{
  "timestamp": "2026-06-19T17:11:53.944238406+03:00",
  "status": 400,
  "error": "Bad Request",
  "message": "The model ID 'cohere.embed-english-light-v3' is not supported or invalid.",
  "path": "/vector/search"
}

The Vector Search workflow will also fail in GraphRAG workflows. If this failure occurs within the Vector Search tool, only the MCP tool call will fail, and the AI agent will most likely continue by attempting a different tool.

MCP tool calls can fail frequently, this is not unusual. The response to such failures is entirely determined by the AI agent. In most cases, if the question remains unanswered and the agent has not exhausted all available tool options, it will attempt another tool or retry the same tool with different parameters. However, if failures occur repeatedly, or if no other appropriate tools are available, the AI agent will conclude with a generic response such as "I cannot answer the question."

Caution

Ensure that the vector embedding model used in the Components service is identical to the model used when ingesting documents into the vector store.

When the incoming vector dimension does not match the dimension specified in the index mapping (for example, 1536 instead of 1024), the vector store rejects the query with a mapping validation error.

The following is an example error response from the components service:

{
  "error": {
    "root_cause": [
      {
        "type": "query_shard_exception",
        "reason": "failed to create query: The query vector has a different number of dimensions [1536] than the document vectors [1024].",
        "index_uuid": "9G2KlJjvTsiqkUX0gKvB1Q",
        "index": "your_index_vectors_v2"
      }
    ]
  }
}

Errors in GraphRAG workflows related to the Components service typically result from incorrect parameters in the Configuration workflow.

Misconfigured vectorSearchUrl

If the vectorSearchUrl parameter is misconfigured, the "Step - Vector Search - Async" and "Vector Search Tool" will fail at the "Call graphrag-components API" step. The GraphRAG workflow will return an error similar to the following:

  
"error": {
  
  
"message": "getaddrinfo EAI_AGAIN graphrag-componentss",
  
  
  
"name": "Error",
  
  
  
"stack": "Error: getaddrinfo EAI_AGAIN graphrag-componentss\n 
at AxiosError.from 
(/usr/local/lib/node_modules/n8n/node_modules/.pnpm/axios@1.16.0/node_modules/axios/lib/core/AxiosError.js:77:24)\n 
at RedirectableRequest.handleRequestError (/usr/local/lib/node_modules/n8n/node_modules/.pnpm/..

This error will also appear in the logs of the Components service container or pod.

Invalid vectorStoreCustomMappings configuration
If any of the mandatory fields in vectorStoreCustomMappings, specifically payload, source, score or id, reference a JSON path that does not exist in the results returned from the Components service, the "Step - Vector Search - Async" and Vector Search Tool will fail at the Prepare response step. The GraphRAG workflow will return an error similar to the following:
```
[
  {
    "error": "Mandatory vector store results property is either not 
available or null [line 64]"
  }
]
```

The most common reason is that the vector index is empty. You can verify this either through the Elasticsearch or OpenSearch dashboards, or by using a curl query.

Example curl request to check whether an Elasticsearch index is empty:

curl -X GET "http://localhost:9200/my_vector_index/_count" \
     -H "Content-Type: application/json"

Example response when the index is empty:

{
  "count": 0,
  "_shards": {
    "total": 1,
    "successful": 1,
    "skipped": 0,
    "failed": 0
  }
}

Another possible reason is a misconfigured vector field name. If the VECTOR_FIELD_NAME configuration property does not match the actual field name in the index mappings, the search will fail to locate the vector data. Compare the configured value against the vector field name defined in your index mappings to ensure they match.

In this section: