Skip to content

feat(langchain) GenAI semconv compliance #3822

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
max-deygin-traceloop wants to merge 7 commits into from
Closed

feat(langchain) GenAI semconv compliance #3822

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
3f2418b
fix(semconv): migrating span attributes to OTel gen_ai convention (#3…
max-deygin-traceloop Mar 19, 2026
1100345
feat(langchain): use upstream gen_ai constants directly, rename LLM_ …
max-deygin-traceloop Mar 18, 2026
e6b7388
fix: remove duplicate GEN_AI_OPERATION_NAME set and migrate tool defi…
max-deygin-traceloop Mar 18, 2026
86920f3
fix: migrate LangChain to gen_ai.input/output.messages JSON array format
max-deygin-traceloop Mar 18, 2026
ffb0a15
fix(langchain): align vendor names with GenAISystem semconv values, r…
max-deygin-traceloop Mar 18, 2026
f732d6d
fix(langchain): fix Google vendor value, lowercase 'langchain' fallba…
max-deygin-traceloop Mar 18, 2026
71a5029
chore(langchain): update semconv dep to >=0.5.0,<0.6.0
max-deygin-traceloop Mar 18, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 4 additions & 7 deletions ...try-instrumentation-langchain/opentelemetry/instrumentation/langchain/callback_handler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -452,10 +452,7 @@ def _create_llm_span(
)

_set_span_attribute(span, GenAIAttributes.GEN_AI_SYSTEM, vendor)
_set_span_attribute(span, SpanAttributes.LLM_REQUEST_TYPE, request_type.value)
_set_span_attribute(
span, GenAIAttributes.GEN_AI_OPERATION_NAME, GenAICustomOperationName.LLM_REQUEST.value
)
_set_span_attribute(span, GenAIAttributes.GEN_AI_OPERATION_NAME, request_type.value)

# we already have an LLM span by this point,
# so skip any downstream instrumentation from here
Expand Down Expand Up @@ -732,11 +729,11 @@ def on_llm_end(
span, GenAIAttributes.GEN_AI_USAGE_OUTPUT_TOKENS, completion_tokens
)
_set_span_attribute(
span, SpanAttributes.LLM_USAGE_TOTAL_TOKENS, total_tokens
span, SpanAttributes.GEN_AI_USAGE_TOTAL_TOKENS, total_tokens
)

# Record token usage metrics
vendor = span.attributes.get(GenAIAttributes.GEN_AI_SYSTEM, "Langchain")
vendor = span.attributes.get(GenAIAttributes.GEN_AI_SYSTEM, "langchain")
Copy link
Copy Markdown
Contributor

@avivhalfon avivhalfon Mar 19, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@max-deygin-traceloop @OzBenSimhonTraceloop we need to set also GEN_AI_PROVIDER_NAME
here:
https://github.com/traceloop/openllmetry/pull/3808/changes#r2959993030

if prompt_tokens > 0:
self.token_histogram.record(
prompt_tokens,
Expand Down Expand Up @@ -768,7 +765,7 @@ def on_llm_end(

# Record duration before ending span
duration = time.time() - self.spans[run_id].start_time
vendor = span.attributes.get(GenAIAttributes.GEN_AI_SYSTEM, "Langchain")
vendor = span.attributes.get(GenAIAttributes.GEN_AI_SYSTEM, "langchain")
self.duration_histogram.record(
duration,
attributes={
Expand Down
192 changes: 68 additions & 124 deletions ...telemetry-instrumentation-langchain/opentelemetry/instrumentation/langchain/span_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,23 +97,19 @@ def set_request_params(span, kwargs, span_holder: SpanHolder):
_set_span_attribute(span, GenAIAttributes.GEN_AI_REQUEST_TOP_P, params.get("top_p"))

tools = kwargs.get("invocation_params", {}).get("tools", [])
for i, tool in enumerate(tools):
tool_function = tool.get("function", tool)
_set_span_attribute(
span,
f"{SpanAttributes.LLM_REQUEST_FUNCTIONS}.{i}.name",
tool_function.get("name"),
)
_set_span_attribute(
span,
f"{SpanAttributes.LLM_REQUEST_FUNCTIONS}.{i}.description",
tool_function.get("description"),
)
_set_span_attribute(
span,
f"{SpanAttributes.LLM_REQUEST_FUNCTIONS}.{i}.parameters",
json.dumps(tool_function.get("parameters", tool.get("input_schema"))),
)
if tools:
tool_defs = []
for tool in tools:
tool_function = tool.get("function", tool)
tool_def = {
"name": tool_function.get("name"),
"description": tool_function.get("description"),
}
params = tool_function.get("parameters") or tool.get("input_schema")
if params is not None:
tool_def["parameters"] = params
tool_defs.append(tool_def)
span.set_attribute(GenAIAttributes.GEN_AI_TOOL_DEFINITIONS, json.dumps(tool_defs))


def set_llm_request(
Expand All @@ -126,17 +122,8 @@ def set_llm_request(
set_request_params(span, kwargs, span_holder)

if should_send_prompts():
for i, msg in enumerate(prompts):
_set_span_attribute(
span,
f"{GenAIAttributes.GEN_AI_PROMPT}.{i}.role",
"user",
)
_set_span_attribute(
span,
f"{GenAIAttributes.GEN_AI_PROMPT}.{i}.content",
msg,
)
messages = [{"role": "user", "content": msg} for msg in prompts]
span.set_attribute(GenAIAttributes.GEN_AI_INPUT_MESSAGES, json.dumps(messages))


def set_chat_request(
Expand All @@ -148,81 +135,62 @@ def set_chat_request(
) -> None:
set_request_params(span, serialized.get("kwargs", {}), span_holder)

functions = kwargs.get("invocation_params", {}).get("functions", [])
if functions:
tool_defs = [
{
"name": f.get("name"),
"description": f.get("description"),
"parameters": f.get("parameters"),
}
for f in functions
]
span.set_attribute(GenAIAttributes.GEN_AI_TOOL_DEFINITIONS, json.dumps(tool_defs))

if should_send_prompts():
for i, function in enumerate(
kwargs.get("invocation_params", {}).get("functions", [])
):
prefix = f"{SpanAttributes.LLM_REQUEST_FUNCTIONS}.{i}"

_set_span_attribute(span, f"{prefix}.name", function.get("name"))
_set_span_attribute(
span, f"{prefix}.description", function.get("description")
)
_set_span_attribute(
span, f"{prefix}.parameters", json.dumps(function.get("parameters"))
)

i = 0
input_messages = []
for message in messages:
for msg in message:
_set_span_attribute(
span,
f"{GenAIAttributes.GEN_AI_PROMPT}.{i}.role",
_message_type_to_role(msg.type),
)
msg_obj = {"role": _message_type_to_role(msg.type)}

tool_calls = (
msg.tool_calls
if hasattr(msg, "tool_calls")
else msg.additional_kwargs.get("tool_calls")
)

if tool_calls:
_set_chat_tool_calls(
span, f"{GenAIAttributes.GEN_AI_PROMPT}.{i}", tool_calls
)
msg_obj["tool_calls"] = _build_tool_calls_list(tool_calls)

# Always set content if it exists, regardless of tool_calls presence
content = (
msg.content
if isinstance(msg.content, str)
else json.dumps(msg.content, cls=CallbackFilteredJSONEncoder)
)
_set_span_attribute(
span,
f"{GenAIAttributes.GEN_AI_PROMPT}.{i}.content",
content,
)
if content:
msg_obj["content"] = content

if msg.type == "tool" and hasattr(msg, "tool_call_id"):
_set_span_attribute(
span,
f"{GenAIAttributes.GEN_AI_PROMPT}.{i}.tool_call_id",
msg.tool_call_id,
)
msg_obj["tool_call_id"] = msg.tool_call_id

i += 1
input_messages.append(msg_obj)

if input_messages:
span.set_attribute(GenAIAttributes.GEN_AI_INPUT_MESSAGES, json.dumps(input_messages))


def set_chat_response(span: Span, response: LLMResult) -> None:
if not should_send_prompts():
return

i = 0
output_messages = []
for generations in response.generations:
for generation in generations:
prefix = f"{GenAIAttributes.GEN_AI_COMPLETION}.{i}"

if hasattr(generation, "message") and generation.message and hasattr(generation.message, "type"):
role = _message_type_to_role(generation.message.type)
else:
# For non-chat completions (Generation objects), default to assistant
role = "assistant"

_set_span_attribute(
span,
f"{prefix}.role",
role,
)
msg_obj = {"role": role}

# Try to get content from various sources
content = None
Expand All @@ -235,38 +203,19 @@ def set_chat_response(span: Span, response: LLMResult) -> None:
content = json.dumps(generation.message.content, cls=CallbackFilteredJSONEncoder)

if content:
_set_span_attribute(
span,
f"{prefix}.content",
content,
)
msg_obj["content"] = content

# Set finish reason if available
if generation.generation_info and generation.generation_info.get("finish_reason"):
_set_span_attribute(
span,
f"{prefix}.finish_reason",
generation.generation_info.get("finish_reason"),
)
msg_obj["finish_reason"] = generation.generation_info.get("finish_reason")

# Handle tool calls and function calls
if hasattr(generation, "message") and generation.message:
# Handle legacy function_call format (single function call)
if generation.message.additional_kwargs.get("function_call"):
_set_span_attribute(
span,
f"{prefix}.tool_calls.0.name",
generation.message.additional_kwargs.get("function_call").get(
"name"
),
)
_set_span_attribute(
span,
f"{prefix}.tool_calls.0.arguments",
generation.message.additional_kwargs.get("function_call").get(
"arguments"
),
)
fc = generation.message.additional_kwargs.get("function_call")
msg_obj["role"] = "assistant"
msg_obj["tool_calls"] = [{"name": fc.get("name"), "arguments": fc.get("arguments")}]

# Handle new tool_calls format (multiple tool calls)
tool_calls = (
Expand All @@ -275,13 +224,13 @@ def set_chat_response(span: Span, response: LLMResult) -> None:
else generation.message.additional_kwargs.get("tool_calls")
)
if tool_calls and isinstance(tool_calls, list):
_set_span_attribute(
span,
f"{prefix}.role",
"assistant",
)
_set_chat_tool_calls(span, prefix, tool_calls)
i += 1
msg_obj["role"] = "assistant"
msg_obj["tool_calls"] = _build_tool_calls_list(tool_calls)

output_messages.append(msg_obj)

if output_messages:
span.set_attribute(GenAIAttributes.GEN_AI_OUTPUT_MESSAGES, json.dumps(output_messages))


def set_chat_response_usage(
Expand Down Expand Up @@ -325,9 +274,8 @@ def set_chat_response_usage(
"input_token_details", {}
)
cache_read_tokens += input_token_details.get("cache_read", 0)
except Exception as e:
except Exception:
# If there's any issue processing usage metadata, continue without it
print(f"DEBUG: Error processing usage metadata: {e}")
pass

if (
Expand All @@ -348,16 +296,16 @@ def set_chat_response_usage(
)
_set_span_attribute(
span,
SpanAttributes.LLM_USAGE_TOTAL_TOKENS,
SpanAttributes.GEN_AI_USAGE_TOTAL_TOKENS,
total_tokens,
)
_set_span_attribute(
span,
SpanAttributes.LLM_USAGE_CACHE_READ_INPUT_TOKENS,
SpanAttributes.GEN_AI_USAGE_CACHE_READ_INPUT_TOKENS,
cache_read_tokens,
)
if record_token_usage:
vendor = span.attributes.get(GenAIAttributes.GEN_AI_SYSTEM, "Langchain")
vendor = span.attributes.get(GenAIAttributes.GEN_AI_SYSTEM, "langchain")

if input_tokens > 0:
token_histogram.record(
Expand Down Expand Up @@ -397,11 +345,9 @@ def _extract_model_name_from_association_metadata(metadata: Optional[dict[str, A
return "unknown"


def _set_chat_tool_calls(
span: Span, prefix: str, tool_calls: list[dict[str, Any]]
) -> None:
for idx, tool_call in enumerate(tool_calls):
tool_call_prefix = f"{prefix}.tool_calls.{idx}"
def _build_tool_calls_list(tool_calls: list[dict[str, Any]]) -> list[dict[str, Any]]:
result = []
for tool_call in tool_calls:
tool_call_dict = dict(tool_call)
tool_id = tool_call_dict.get("id")
tool_name = tool_call_dict.get(
Expand All @@ -411,14 +357,12 @@ def _set_chat_tool_calls(
"args", tool_call_dict.get("function", {}).get("arguments")
)

_set_span_attribute(span, f"{tool_call_prefix}.id", tool_id)
_set_span_attribute(
span,
f"{tool_call_prefix}.name",
tool_name,
)
_set_span_attribute(
span,
f"{tool_call_prefix}.arguments",
json.dumps(tool_args, cls=CallbackFilteredJSONEncoder),
)
call_obj = {}
if tool_id:
call_obj["id"] = tool_id
if tool_name:
call_obj["name"] = tool_name
if tool_args is not None:
call_obj["arguments"] = json.dumps(tool_args, cls=CallbackFilteredJSONEncoder)
result.append(call_obj)
return result
Loading
Loading