fix: return all matching metrics in TTFT test helper per review

Address review feedback: test helper now returns all matching metric
data points so tests can assert exactly one data point was recorded.

Author: Nik-Reddy

instrumentation-genai/opentelemetry-instrumentation-openai-v2/tests/test_ttft_metrics.py

@@ -30,15 +30,16 @@
 )
 
 
-def _get_ttft_metric(metric_reader):
+def _get_ttft_data_points(metric_reader):
+    results = []
     metrics = metric_reader.get_metrics_data().resource_metrics
     if not metrics:
-        return None
+        return results
     for scope_metrics in metrics[0].scope_metrics:
         for m in scope_metrics.metrics:
             if m.name == gen_ai_metrics.GEN_AI_SERVER_TIME_TO_FIRST_TOKEN:
-                return m
-    return None
+                results.extend(m.data.data_points)
+    return results
 
 
 def test_streaming_chat_records_ttft_metric(
@@ -55,12 +56,12 @@ def test_streaming_chat_records_ttft_metric(
         for _ in response:
             pass
 
-    ttft_metric = _get_ttft_metric(metric_reader)
-    assert ttft_metric is not None, (
-        "gen_ai.server.time_to_first_token metric should be recorded for streaming"
+    data_points = _get_ttft_data_points(metric_reader)
+    assert len(data_points) == 1, (
+        "expected exactly one TTFT data point for streaming"
     )
 
-    data_point = ttft_metric.data.data_points[0]
+    data_point = data_points[0]
     assert data_point.sum >= 0
     assert data_point.count == 1
     assert data_point.explicit_bounds == _TTFT_BUCKETS
@@ -94,12 +95,12 @@ async def test_async_streaming_chat_records_ttft_metric(
         async for _ in response:
             pass
 
-    ttft_metric = _get_ttft_metric(metric_reader)
-    assert ttft_metric is not None, (
-        "gen_ai.server.time_to_first_token metric should be recorded for async streaming"
+    data_points = _get_ttft_data_points(metric_reader)
+    assert len(data_points) == 1, (
+        "expected exactly one TTFT data point for async streaming"
     )
 
-    data_point = ttft_metric.data.data_points[0]
+    data_point = data_points[0]
     assert data_point.sum >= 0
     assert data_point.count == 1
     assert data_point.explicit_bounds == _TTFT_BUCKETS
@@ -114,8 +115,8 @@ def test_non_streaming_chat_does_not_record_ttft_metric(
             messages=USER_ONLY_PROMPT, model=DEFAULT_MODEL, stream=False
         )
 
-    ttft_metric = _get_ttft_metric(metric_reader)
-    assert ttft_metric is None, (
+    data_points = _get_ttft_data_points(metric_reader)
+    assert len(data_points) == 0, (
         "gen_ai.server.time_to_first_token metric should not be recorded for non-streaming"
     )
 
@@ -150,11 +151,11 @@ def test_streaming_tool_calls_records_ttft_metric(
         for _ in response:
             pass
 
-    ttft_metric = _get_ttft_metric(metric_reader)
-    assert ttft_metric is not None, (
-        "gen_ai.server.time_to_first_token metric should be recorded for streaming tool calls"
+    data_points = _get_ttft_data_points(metric_reader)
+    assert len(data_points) == 1, (
+        "expected exactly one TTFT data point for streaming tool calls"
     )
 
-    data_point = ttft_metric.data.data_points[0]
+    data_point = data_points[0]
     assert data_point.sum >= 0
     assert data_point.count == 1