[BUG] Creating a model from a non top level field causes a 'field does not exist' error

[karlney]

What is the bug?
It is not possible to train a model from a field that is not a top level field in the opensearch field mappings.

How can one reproduce the bug?
If I have a knn vector field located at the path vector.faiss.hnsw.inner_ef100_m16_flat that uses the following mappings

...
"mappings": {
     ...
    "properties": {
       ...
         "vector": {
          "dynamic": false,
          "properties": {
            "faiss": {
              "dynamic": false,
              "properties": {
                "hnsw": {
                  "dynamic": false,
                  "properties": {
                    "inner_ef100_m16_flat": {
                      "dimension": 768,
                      "method": {
                        "engine": "faiss",
                        "name": "hnsw",
                        "parameters": {
                          "ef_construction": 100,
                          "encoder": {
                            "name": "flat"
                          },
                          "m": 16
                        },
                        "space_type": "innerproduct"
                      },
                      "type": "knn_vector"
                    },
...

When I try to train a model from that field using the method described in https://opensearch.org/docs/latest/search-plugins/knn/api#usage-5

I get this error message

{"error":{"root_cause":[{"type":"action_request_validation_exception","reason":"Validation Failed: 1: Field \"vector.faiss.hnsw.inner_ef100_m16_flat\" does not exist.;"}],"type":"action_request_validation_exception","reason":"Validation Failed: 1: Field \"vector.faiss.hnsw.inner_ef100_m16_flat\" does not exist.;"},"status":400}

Even though I have confirmed that the field exists and that it has the correct data.

If I change the name of the field to top_vector and put it at the top of the mappings like this, and add the same data as before to the new field, THEN I can train a model.

...
"mappings": {
     ...
    "properties": {
       ...
         "top_vector": {
          "type": "knn_vector"
          ... all the same settings as above.. 
         }
...

I believe the error is in

k-NN/src/main/java/org/opensearch/knn/index/IndexUtil.java

Line 103 in 4d6cc16

Object fieldMapping = properties.get(field);

which makes the assumption that the field must be in the top of the properties hierarchy in the mappings, which is no documented in any documentation as far as I can tell, and even if it was, it seems like a weird limitation to not allow the vector field to be anywhere in the document.

[jmazanec15]

I believe the error is in

k-NN/src/main/java/org/opensearch/knn/index/IndexUtil.java

Line 103 in 4d6cc16

Object fieldMapping = properties.get(field);

@karlney I think you are correct. Let me take a look at this one.