fix(botocore): tag Bedrock embedding InvokeModel calls as embeddings

.changelog/4997.fixed

@@ -0,0 +1 @@
+`opentelemetry-instrumentation-botocore`: tag Bedrock embedding InvokeModel calls as embeddings

instrumentation/opentelemetry-instrumentation-botocore/src/opentelemetry/instrumentation/botocore/extensions/bedrock.py

@@ -100,6 +100,12 @@
 _MODEL_ID_KEY: str = "modelId"
 
 
+def _is_embedding_model(model_id: str) -> bool:
+    if not model_id:
+        return False
+    return "embed" in model_id.rsplit(".", 1)[-1].lower()
+
+
 class _BedrockRuntimeExtension(_AwsSdkExtension):
     """
     This class is an extension for <a
@@ -147,8 +153,10 @@ def _extract_metrics_attributes(self) -> _AttributeMapT:
 
         attributes[GEN_AI_REQUEST_MODEL] = model_id
 
-        # titan in invoke model is a text completion one
-        if "body" in self._call_context.params and "amazon.titan" in model_id:
+        if _is_embedding_model(model_id):
+            attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.EMBEDDINGS.value
+        elif "body" in self._call_context.params and "amazon.titan" in model_id:
+            # titan in invoke model is a text completion one
             attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.TEXT_COMPLETION.value
         else:
             attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.CHAT.value
@@ -164,7 +172,10 @@ def extract_attributes(self, attributes: _AttributeMapT):
         model_id = self._call_context.params.get(_MODEL_ID_KEY)
         if model_id:
             attributes[GEN_AI_REQUEST_MODEL] = model_id
-            attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.CHAT.value
+            if _is_embedding_model(model_id):
+                attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.EMBEDDINGS.value
+            else:
+                attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.CHAT.value
 
             # Converse / ConverseStream
             if inference_config := self._call_context.params.get("inferenceConfig"):
@@ -196,7 +207,9 @@ def extract_attributes(self, attributes: _AttributeMapT):
                 try:
                     request_body = json.loads(body)
 
-                    if "amazon.titan" in model_id:
+                    if _is_embedding_model(model_id):
+                        attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.EMBEDDINGS.value
+                    elif "amazon.titan" in model_id:
                         # titan interface is a text completion one
                         attributes[GEN_AI_OPERATION_NAME] = GenAiOperationNameValues.TEXT_COMPLETION.value
                         self._extract_titan_attributes(attributes, request_body)
@@ -357,7 +370,7 @@ def _get_request_messages(self):
                 # if no messages interface, convert to messages format from generic API
                 if not messages:
                     model_id = self._call_context.params.get(_MODEL_ID_KEY)
-                    if "amazon.titan" in model_id:
+                    if "amazon.titan" in model_id and not _is_embedding_model(model_id):
                         messages = self._get_messages_from_input_text(decoded_body, "inputText")
                     elif "cohere.command-r" in model_id:
                         # chat_history can be converted to messages; for now, just use message
@@ -474,6 +487,8 @@ def _invoke_model_on_success(
             result["body"] = StreamingBody(new_stream, len(body_content))
 
             response_body = json.loads(body_content.decode("utf-8"))
+            if _is_embedding_model(model_id):
+                return
             if "amazon.titan" in model_id:
                 self._handle_amazon_titan_response(span, response_body, instrumentor_context, capture_content)
             elif "amazon.nova" in model_id:

instrumentation/opentelemetry-instrumentation-botocore/tests/cassettes/test_invoke_model_with_embeddings_model[amazon.titan].yaml

@@ -0,0 +1,58 @@
+interactions:
+- request:
+    body: |-
+      {
+        "inputText": "Say this is a test"
+      }
+    headers:
+      Content-Length:
+      - '36'
+      User-Agent:
+      - Boto3/1.35.56 md/Botocore#1.35.56 ua/2.0 os/macos#24.0.0 md/arch#arm64 lang/python#3.13.1
+        md/pyimpl#CPython cfg/retry-mode#legacy Botocore/1.35.56
+      X-Amz-Date:
+      - 20250826T103000Z
+      X-Amzn-Trace-Id:
+      - Root=1-aaaaaaaa-bbbbbbbbbbbbbbbbbbbbbbbb;Parent=cccccccccccccccc;Sampled=1
+      amz-sdk-invocation-id:
+      - 11111111-2222-3333-4444-555555555555
+      amz-sdk-request:
+      - attempt=1
+      authorization:
+      - Bearer test_aws_authorization
+    method: POST
+    uri: https://bedrock-runtime.us-east-1.amazonaws.com/model/amazon.titan-embed-text-v1/invoke
+  response:
+    body:
+      string: |-
+        {
+          "embedding": [
+            0.0135,
+            -0.0272,
+            0.0047,
+            0.0189,
+            -0.0091,
+            0.0023,
+            0.0314,
+            -0.0168
+          ],
+          "inputTextTokenCount": 5
+        }
+    headers:
+      Connection:
+      - keep-alive
+      Content-Type:
+      - application/json
+      Date:
+      - Wed, 26 Aug 2026 10:30:00 GMT
+      Set-Cookie: test_set_cookie
+      X-Amzn-Bedrock-Input-Token-Count:
+      - '5'
+      X-Amzn-Bedrock-Invocation-Latency:
+      - '120'
+      x-amzn-RequestId:
+      - aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee
+    status:
+      code: 200
+      message: OK
+version: 1

instrumentation/opentelemetry-instrumentation-botocore/tests/cassettes/test_invoke_model_with_embeddings_model[cohere.embed].yaml

@@ -0,0 +1,67 @@
+interactions:
+- request:
+    body: |-
+      {
+        "texts": [
+          "Say this is a test"
+        ],
+        "input_type": "search_document"
+      }
+    headers:
+      Content-Length:
+      - '76'
+      User-Agent:
+      - Boto3/1.35.56 md/Botocore#1.35.56 ua/2.0 os/macos#24.0.0 md/arch#arm64 lang/python#3.13.1
+        md/pyimpl#CPython cfg/retry-mode#legacy Botocore/1.35.56
+      X-Amz-Date:
+      - 20250826T103001Z
+      X-Amzn-Trace-Id:
+      - Root=1-ffffffff-eeeeeeeeeeeeeeeeeeeeeeee;Parent=dddddddddddddddd;Sampled=1
+      amz-sdk-invocation-id:
+      - 66666666-7777-8888-9999-000000000000
+      amz-sdk-request:
+      - attempt=1
+      authorization:
+      - Bearer test_aws_authorization
+    method: POST
+    uri: https://bedrock-runtime.us-east-1.amazonaws.com/model/cohere.embed-v4%3A0/invoke
+  response:
+    body:
+      string: |-
+        {
+          "id": "836a33cc-61ec-4e65-afaf-c4628171a315",
+          "embeddings": [
+            [
+              0.0135,
+              -0.0272,
+              0.0047,
+              0.0189,
+              -0.0091,
+              0.0023,
+              0.0314,
+              -0.0168
+            ]
+          ],
+          "response_type": "embeddings_floats",
+          "texts": [
+            "Say this is a test"
+          ]
+        }
+    headers:
+      Connection:
+      - keep-alive
+      Content-Type:
+      - application/json
+      Date:
+      - Wed, 26 Aug 2026 10:30:01 GMT
+      Set-Cookie: test_set_cookie
+      X-Amzn-Bedrock-Input-Token-Count:
+      - '5'
+      X-Amzn-Bedrock-Invocation-Latency:
+      - '95'
+      x-amzn-RequestId:
+      - 836a33cc-61ec-4e65-afaf-c4628171a315
+    status:
+      code: 200
+      message: OK
+version: 1

instrumentation/opentelemetry-instrumentation-botocore/tests/test_botocore_bedrock.py

@@ -1269,6 +1269,68 @@ def get_model_name_from_family(llm_model):
     return llm_model_name[llm_model]
 
 
+def get_embeddings_model_name_from_family(model_family):
+    return {
+        "amazon.titan": "amazon.titan-embed-text-v1",
+        "cohere.embed": "cohere.embed-v4:0",
+    }[model_family]
+
+
+def get_invoke_embeddings_body(llm_model):
+    if "cohere.embed" in llm_model:
+        return json.dumps(
+            {
+                "texts": ["Say this is a test"],
+                "input_type": "search_document",
+            }
+        )
+    if "amazon.titan" in llm_model:
+        return json.dumps({"inputText": "Say this is a test"})
+    raise ValueError(f"No embeddings config for {llm_model}")
+
+
+@pytest.mark.parametrize(
+    "model_family",
+    [
+        "amazon.titan",
+        "cohere.embed",
+    ],
+)
+@pytest.mark.vcr()
+def test_invoke_model_with_embeddings_model(
+    span_exporter,
+    log_exporter,
+    bedrock_runtime_client,
+    instrument_with_content,
+    model_family,
+):
+    llm_model_value = get_embeddings_model_name_from_family(model_family)
+    body = get_invoke_embeddings_body(llm_model_value)
+    response = bedrock_runtime_client.invoke_model(
+        body=body,
+        modelId=llm_model_value,
+    )
+
+    response_body = json.loads(response["body"].read())
+    assert response_body
+    if model_family == "amazon.titan":
+        assert "embedding" in response_body
+    elif model_family == "cohere.embed":
+        assert "embeddings" in response_body
+    else:
+        pytest.xfail(f"model family not handled: {model_family}")
+
+    (span,) = span_exporter.get_finished_spans()
+    assert_stream_completion_attributes(
+        span,
+        llm_model_value,
+        operation_name="embeddings",
+    )
+
+    logs = log_exporter.get_finished_logs()
+    assert len(logs) == 0
+
+
 @pytest.mark.parametrize(
     "model_family",
     [