fix: plumb LLM token counts into rlsapi Splunk telemetry

total_llm_tokens was hardcoded to 0 in the rlsapi Splunk event builder
despite token counting being implemented via extract_token_usage().
Add input_tokens and output_tokens to InferenceEventData and pass
actual counts from the endpoint handler.

Ref: RSPEED-2857
Signed-off-by: Major Hayden <major@redhat.com>

Author: major

src/app/endpoints/rlsapi_v1.py

@@ -353,8 +353,22 @@ def _queue_splunk_event(  # pylint: disable=too-many-arguments,too-many-position
     response_text: str,
     inference_time: float,
     sourcetype: str,
+    input_tokens: int = 0,
+    output_tokens: int = 0,
 ) -> None:
-    """Build and queue a Splunk telemetry event for background sending."""
+    """Build and queue a Splunk telemetry event for background sending.
+
+    Args:
+        background_tasks: FastAPI background task manager.
+        infer_request: Original rlsapi v1 inference request.
+        request: FastAPI request object used to resolve identity context.
+        request_id: Unique identifier for the request.
+        response_text: Response text to include in the telemetry event.
+        inference_time: Request processing duration in seconds.
+        sourcetype: Splunk sourcetype to use when sending the event.
+        input_tokens: Number of prompt tokens consumed by the LLM call.
+        output_tokens: Number of completion tokens produced by the LLM call.
+    """
     org_id, system_id = _get_rh_identity_context(request)
     systeminfo = infer_request.context.systeminfo
 
@@ -370,6 +384,8 @@ def _queue_splunk_event(  # pylint: disable=too-many-arguments,too-many-position
         system_os=systeminfo.os,
         system_version=systeminfo.version,
         system_arch=systeminfo.arch,
+        input_tokens=input_tokens,
+        output_tokens=output_tokens,
     )
 
     event = build_inference_event(event_data)
@@ -754,6 +770,8 @@ async def infer_endpoint(  # pylint: disable=R0914
         response_text,
         inference_time,
         "infer_with_llm",
+        input_tokens=token_usage.input_tokens,
+        output_tokens=token_usage.output_tokens,
     )
 
     logger.info("Completed rlsapi v1 /infer request %s", request_id)

src/observability/formats/rlsapi.py

@@ -26,6 +26,8 @@ class InferenceEventData:  # pylint: disable=too-many-instance-attributes
     system_os: str
     system_version: str
     system_arch: str
+    input_tokens: int = 0
+    output_tokens: int = 0
 
 
 def build_inference_event(data: InferenceEventData) -> dict[str, Any]:
@@ -47,8 +49,7 @@ def build_inference_event(data: InferenceEventData) -> dict[str, Any]:
         "deployment": configuration.deployment_environment,
         "org_id": data.org_id,
         "system_id": data.system_id,
-        # Token counting not yet implemented in lightspeed-stack; rlsapi uses 0 as default
-        "total_llm_tokens": 0,
+        "total_llm_tokens": data.input_tokens + data.output_tokens,
         "request_id": data.request_id,
         "cla_version": data.cla_version,
         "system_os": data.system_os,

tests/integration/endpoints/test_rlsapi_v1_integration.py

@@ -115,6 +115,10 @@ def _setup_responses_mock(
     """Set up responses.create mock with the given response text."""
     mock_response = mocker.Mock()
     mock_response.output = [_create_mock_response_output(mocker, response_text)]
+    mock_usage = mocker.Mock()
+    mock_usage.input_tokens = 10
+    mock_usage.output_tokens = 5
+    mock_response.usage = mock_usage
 
     mock_responses = mocker.Mock()
     mock_responses.create = mocker.AsyncMock(return_value=mock_response)
@@ -303,6 +307,10 @@ async def test_rlsapi_v1_infer_fallback_response_empty_output(
 
     mock_response = mocker.Mock()
     mock_response.output = []
+    mock_usage = mocker.Mock()
+    mock_usage.input_tokens = 10
+    mock_usage.output_tokens = 5
+    mock_response.usage = mock_usage
 
     mock_responses = mocker.Mock()
     mock_responses.create = mocker.AsyncMock(return_value=mock_response)
@@ -342,6 +350,10 @@ async def test_rlsapi_v1_infer_input_source_combination(
 
     mock_response = mocker.Mock()
     mock_response.output = [_create_mock_response_output(mocker, "response text")]
+    mock_usage = mocker.Mock()
+    mock_usage.input_tokens = 10
+    mock_usage.output_tokens = 5
+    mock_response.usage = mock_usage
 
     mock_responses = mocker.Mock()
     mock_responses.create = mocker.AsyncMock(return_value=mock_response)
@@ -401,6 +413,10 @@ async def test_rlsapi_v1_infer_no_mcp_servers_passes_empty_tools(
 
     mock_response = mocker.Mock()
     mock_response.output = [_create_mock_response_output(mocker, "response text")]
+    mock_usage = mocker.Mock()
+    mock_usage.input_tokens = 10
+    mock_usage.output_tokens = 5
+    mock_response.usage = mock_usage
 
     mock_responses = mocker.Mock()
     mock_responses.create = mocker.AsyncMock(return_value=mock_response)
@@ -442,6 +458,10 @@ async def test_rlsapi_v1_infer_mcp_tools_passed_to_llm(
 
     mock_response = mocker.Mock()
     mock_response.output = [_create_mock_response_output(mocker, "enriched response")]
+    mock_usage = mocker.Mock()
+    mock_usage.input_tokens = 10
+    mock_usage.output_tokens = 5
+    mock_response.usage = mock_usage
 
     mock_responses = mocker.Mock()
     mock_responses.create = mocker.AsyncMock(return_value=mock_response)

tests/unit/app/endpoints/test_rlsapi_v1.py

@@ -116,6 +116,10 @@ def mock_llm_response_fixture(mocker: MockerFixture) -> None:
     mock_response.output = [
         _create_mock_response_output(mocker, "This is a test LLM response.")
     ]
+    mock_usage = mocker.Mock()
+    mock_usage.input_tokens = 10
+    mock_usage.output_tokens = 5
+    mock_response.usage = mock_usage
     _setup_responses_mock(mocker, mocker.AsyncMock(return_value=mock_response))
 
 
@@ -124,6 +128,10 @@ def mock_empty_llm_response_fixture(mocker: MockerFixture) -> None:
     """Mock responses.create to return empty output list."""
     mock_response = mocker.Mock()
     mock_response.output = []
+    mock_usage = mocker.Mock()
+    mock_usage.input_tokens = 10
+    mock_usage.output_tokens = 5
+    mock_response.usage = mock_usage
     _setup_responses_mock(mocker, mocker.AsyncMock(return_value=mock_response))
 
 

tests/unit/observability/formats/test_rlsapi.py

@@ -50,6 +50,33 @@ def test_builds_event_with_all_fields(
     assert event["total_llm_tokens"] == 0
 
 
+def test_builds_event_with_token_counts(mocker: MockerFixture) -> None:
+    """Test total_llm_tokens is computed from input and output token counts."""
+    mocker.patch(
+        "observability.formats.rlsapi.configuration"
+    ).deployment_environment = "production"
+
+    data = InferenceEventData(
+        question="test",
+        response="test",
+        inference_time=1.0,
+        model="test-model",
+        org_id="org1",
+        system_id="sys1",
+        request_id="req_1",
+        cla_version="CLA/1.0",
+        system_os="RHEL",
+        system_version="9.4",
+        system_arch="x86_64",
+        input_tokens=150,
+        output_tokens=75,
+    )
+
+    event = build_inference_event(data)
+
+    assert event["total_llm_tokens"] == 225
+
+
 def test_handles_auth_disabled_values(mocker: MockerFixture) -> None:
     """Test event handles auth_disabled placeholder values."""
     data = InferenceEventData(