fix: rename metric to gen_ai.client.operation.time_to_first_chunk

Align metric name with semantic conventions specification:
- Rename gen_ai.client.time_to_first_token to
  gen_ai.client.operation.time_to_first_chunk per semconv registry
- Update bucket boundaries to match spec-defined values
- Rename helper functions from ttft to ttfc throughout
- Update all test assertions to use correct metric name

Addresses reviewer feedback from @lmolkova (client, not server)
and @xrmx (return all data points for assertions).

Author: Nik-Reddy

instrumentation-genai/opentelemetry-instrumentation-openai-v2/src/opentelemetry/instrumentation/openai_v2/instruments.py

@@ -2,12 +2,12 @@
 from opentelemetry.util.genai.instruments import (
     create_duration_histogram,
     create_token_histogram,
-    create_ttft_histogram,
+    create_ttfc_histogram,
 )
 
 
 class Instruments:
     def __init__(self, meter: Meter):
         self.operation_duration_histogram: Histogram = create_duration_histogram(meter)
         self.token_usage_histogram: Histogram = create_token_histogram(meter)
-        self.ttft_histogram: Histogram = create_ttft_histogram(meter)
+        self.ttfc_histogram: Histogram = create_ttfc_histogram(meter)

instrumentation-genai/opentelemetry-instrumentation-openai-v2/src/opentelemetry/instrumentation/openai_v2/patch.py

@@ -926,7 +926,7 @@ def _record_ttft(self):
             common_attributes[ServerAttributes.SERVER_PORT] = (
                 self._request_attributes[ServerAttributes.SERVER_PORT]
             )
-        self._instruments.ttft_histogram.record(
+        self._instruments.ttfc_histogram.record(
             ttft,
             attributes=common_attributes,
         )

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

@@ -8,8 +8,8 @@
     server_attributes as ServerAttributes,
 )
 from opentelemetry.util.genai.instruments import (
-    GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN,
-    GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN_BUCKETS,
+    GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK,
+    _GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK_BUCKETS,
     get_metric_data_points,
 )
 from opentelemetry.util.genai.utils import is_experimental_mode
@@ -29,15 +29,15 @@ def test_streaming_chat_records_ttft_metric(
         for _ in response:
             pass
 
-    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN)
+    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK)
     assert len(data_points) == 1, (
-        "expected exactly one TTFT data point for streaming"
+        "expected exactly one TTFC data point for streaming"
     )
 
     data_point = data_points[0]
     assert data_point.sum >= 0
     assert data_point.count == 1
-    assert data_point.explicit_bounds == tuple(GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN_BUCKETS)
+    assert data_point.explicit_bounds == tuple(_GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK_BUCKETS)
 
     latest_experimental_enabled = is_experimental_mode()
     assert GenAIAttributes.GEN_AI_OPERATION_NAME in data_point.attributes
@@ -68,15 +68,15 @@ async def test_async_streaming_chat_records_ttft_metric(
         async for _ in response:
             pass
 
-    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN)
+    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK)
     assert len(data_points) == 1, (
-        "expected exactly one TTFT data point for async streaming"
+        "expected exactly one TTFC data point for async streaming"
     )
 
     data_point = data_points[0]
     assert data_point.sum >= 0
     assert data_point.count == 1
-    assert data_point.explicit_bounds == tuple(GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN_BUCKETS)
+    assert data_point.explicit_bounds == tuple(_GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK_BUCKETS)
 
 
 def test_non_streaming_chat_does_not_record_ttft_metric(
@@ -88,9 +88,9 @@ def test_non_streaming_chat_does_not_record_ttft_metric(
             messages=USER_ONLY_PROMPT, model=DEFAULT_MODEL, stream=False
         )
 
-    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN)
+    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK)
     assert len(data_points) == 0, (
-        "gen_ai.client.time_to_first_token metric should not be recorded for non-streaming"
+        "gen_ai.client.operation.time_to_first_chunk metric should not be recorded for non-streaming"
     )
 
 
@@ -124,9 +124,9 @@ def test_streaming_tool_calls_records_ttft_metric(
         for _ in response:
             pass
 
-    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN)
+    data_points = get_metric_data_points(metric_reader, GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK)
     assert len(data_points) == 1, (
-        "expected exactly one TTFT data point for streaming tool calls"
+        "expected exactly one TTFC data point for streaming tool calls"
     )
 
     data_point = data_points[0]

util/opentelemetry-util-genai/src/opentelemetry/util/genai/instruments.py

@@ -35,25 +35,25 @@
     67108864,
 ]
 
-GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN = "gen_ai.client.time_to_first_token"
+GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK = (
+    "gen_ai.client.operation.time_to_first_chunk"
+)
 
-GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN_BUCKETS = [
-    0.001,
-    0.005,
+_GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK_BUCKETS = [
     0.01,
     0.02,
     0.04,
-    0.06,
     0.08,
-    0.1,
-    0.25,
-    0.5,
-    0.75,
-    1.0,
-    2.5,
-    5.0,
-    7.5,
-    10.0,
+    0.16,
+    0.32,
+    0.64,
+    1.28,
+    2.56,
+    5.12,
+    10.24,
+    20.48,
+    40.96,
+    81.92,
 ]
 
 
@@ -75,12 +75,12 @@ def create_token_histogram(meter: Meter) -> Histogram:
     )
 
 
-def create_ttft_histogram(meter: Meter) -> Histogram:
+def create_ttfc_histogram(meter: Meter) -> Histogram:
     return meter.create_histogram(
-        name=GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN,
-        description="Time to generate first token for successful responses",
+        name=GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK,
+        description="Time to receive the first chunk in a streaming response",
         unit="s",
-        explicit_bucket_boundaries_advisory=GEN_AI_CLIENT_TIME_TO_FIRST_TOKEN_BUCKETS,
+        explicit_bucket_boundaries_advisory=_GEN_AI_CLIENT_OPERATION_TIME_TO_FIRST_CHUNK_BUCKETS,
     )
 
 

util/opentelemetry-util-genai/src/opentelemetry/util/genai/metrics.py

@@ -17,7 +17,7 @@
 from opentelemetry.util.genai.instruments import (
     create_duration_histogram,
     create_token_histogram,
-    create_ttft_histogram,
+    create_ttfc_histogram,
 )
 from opentelemetry.util.genai.types import LLMInvocation
 from opentelemetry.util.types import AttributeValue
@@ -29,7 +29,7 @@ class InvocationMetricsRecorder:
     def __init__(self, meter: Meter):
         self._duration_histogram: Histogram = create_duration_histogram(meter)
         self._token_histogram: Histogram = create_token_histogram(meter)
-        self._ttft_histogram: Histogram = create_ttft_histogram(meter)
+        self._ttfc_histogram: Histogram = create_ttfc_histogram(meter)
 
     def record(
         self,
@@ -108,7 +108,7 @@ def record(
             )
 
         if invocation.time_to_first_token_s is not None and not error_type:
-            self._ttft_histogram.record(
+            self._ttfc_histogram.record(
                 invocation.time_to_first_token_s,
                 attributes=attributes,
                 context=span_context,