Python: Telemetry and observability follow-up (#833)

* updated telemetry work

* updated telemetry

* slight improvement

* updated tests

* fixes for telemetry

* fixes for mypy

* added settings setup to runner to avoid error

* streamline usage

* updated tests

* updated tests

* further refinement

* fix dumped item for otel

* removed enable_workflow_otel

* final fixes

* final fixes

* updated samples

* removed exporters

* fix tests

* fixed last import'

* fixed devui
This commit is contained in:
Eduard van Valkenburg
2025-09-23 08:21:56 +02:00
committed by GitHub
Unverified
parent f93f16a9ad
commit 2576e7a091
52 changed files with 1625 additions and 1586 deletions
+69
View File
@@ -0,0 +1,69 @@
# Copyright (c) Microsoft. All rights reserved.
from collections.abc import Generator
from typing import Any
from unittest.mock import patch
from opentelemetry.sdk.trace.export import SimpleSpanProcessor, SpanExporter
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from pytest import fixture
@fixture
def enable_otel(request: Any) -> bool:
"""Fixture that returns a boolean indicating if Otel is enabled."""
return request.param if hasattr(request, "param") else True
@fixture
def enable_sensitive_data(request: Any) -> bool:
"""Fixture that returns a boolean indicating if sensitive data is enabled."""
return request.param if hasattr(request, "param") else True
@fixture(autouse=True)
def span_exporter(monkeypatch, enable_otel: bool, enable_sensitive_data: bool) -> Generator[SpanExporter]:
"""Fixture to remove environment variables for OtelSettings."""
env_vars = [
"ENABLE_OTEL",
"ENABLE_SENSITIVE_DATA",
"OTLP_ENDPOINT",
"APPLICATIONINSIGHTS_CONNECTION_STRING",
"APPLICATIONINSIGHTS_LIVE_METRICS",
]
for key in env_vars:
monkeypatch.delenv(key, raising=False) # type: ignore
monkeypatch.setenv("ENABLE_OTEL", str(enable_otel)) # type: ignore
if not enable_otel:
# we overwrite sensitive data for tests
enable_sensitive_data = False
monkeypatch.setenv("ENABLE_SENSITIVE_DATA", str(enable_sensitive_data)) # type: ignore
import importlib
from opentelemetry import trace
import agent_framework.observability as observability
# Reload the module to ensure a clean state for tests, then create a
# fresh OtelSettings instance and patch the module attribute.
importlib.reload(observability)
# recreate otel settings with values from above and no file.
otel = observability.OtelSettings(env_file_path="test.env")
otel.setup_observability()
monkeypatch.setattr(observability, "OTEL_SETTINGS", otel, raising=False) # type: ignore
exporter = InMemorySpanExporter()
with (
patch("agent_framework.observability.OTEL_SETTINGS", otel),
patch("agent_framework.observability.setup_observability"),
):
if enable_otel or enable_sensitive_data:
trace.get_tracer_provider().add_span_processor(
SimpleSpanProcessor(exporter) # type: ignore[func-returns-value]
)
yield exporter
# Clean up
exporter.clear()
@@ -27,7 +27,6 @@ from agent_framework import (
ai_function,
use_function_invocation,
)
from agent_framework.telemetry import OtelSettings, setup_telemetry
if sys.version_info >= (3, 12):
from typing import override # type: ignore
@@ -38,29 +37,6 @@ else:
logger = logging.getLogger(__name__)
@fixture
def enable_otel(request: Any) -> bool:
"""Fixture that returns a boolean indicating if Otel is enabled."""
return request.param if hasattr(request, "param") else True
@fixture
def enable_sensitive_data(request: Any) -> bool:
"""Fixture that returns a boolean indicating if sensitive data is enabled."""
return request.param if hasattr(request, "param") else False
@fixture
def otel_settings(enable_otel: bool, enable_sensitive_data: bool) -> OtelSettings:
"""Fixture to set environment variables for OtelSettings."""
from agent_framework.telemetry import OTEL_SETTINGS
setup_telemetry(enable_otel=enable_otel, enable_sensitive_data=enable_sensitive_data)
return OTEL_SETTINGS
@fixture(scope="function")
def chat_history() -> list[ChatMessage]:
return []
@@ -0,0 +1,469 @@
# Copyright (c) Microsoft. All rights reserved.
import logging
from collections.abc import MutableSequence
from typing import Any
from unittest.mock import MagicMock, Mock, patch
import pytest
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from opentelemetry.semconv_ai import SpanAttributes
from opentelemetry.trace import StatusCode
from agent_framework import (
AGENT_FRAMEWORK_USER_AGENT,
AgentProtocol,
AgentRunResponse,
AgentThread,
BaseChatClient,
ChatMessage,
ChatOptions,
ChatResponse,
ChatResponseUpdate,
Role,
UsageDetails,
prepend_agent_framework_to_user_agent,
)
from agent_framework.exceptions import AgentInitializationError, ChatClientInitializationError
from agent_framework.observability import (
OPEN_TELEMETRY_AGENT_MARKER,
OPEN_TELEMETRY_CHAT_CLIENT_MARKER,
ROLE_EVENT_MAP,
ChatMessageListTimestampFilter,
OtelAttr,
get_function_span,
use_agent_observability,
use_observability,
)
# region Test constants
def test_role_event_map():
"""Test that ROLE_EVENT_MAP contains expected mappings."""
assert ROLE_EVENT_MAP["system"] == OtelAttr.SYSTEM_MESSAGE
assert ROLE_EVENT_MAP["user"] == OtelAttr.USER_MESSAGE
assert ROLE_EVENT_MAP["assistant"] == OtelAttr.ASSISTANT_MESSAGE
assert ROLE_EVENT_MAP["tool"] == OtelAttr.TOOL_MESSAGE
def test_enum_values():
"""Test that OtelAttr enum has expected values."""
assert OtelAttr.OPERATION == "gen_ai.operation.name"
assert SpanAttributes.LLM_SYSTEM == "gen_ai.system"
assert SpanAttributes.LLM_REQUEST_MODEL == "gen_ai.request.model"
assert OtelAttr.CHAT_COMPLETION_OPERATION == "chat"
assert OtelAttr.TOOL_EXECUTION_OPERATION == "execute_tool"
assert OtelAttr.AGENT_INVOKE_OPERATION == "invoke_agent"
# region Test ChatMessageListTimestampFilter
def test_filter_without_index_key():
"""Test filter method when record doesn't have INDEX_KEY."""
log_filter = ChatMessageListTimestampFilter()
record = logging.LogRecord(
name="test", level=logging.INFO, pathname="", lineno=0, msg="test message", args=(), exc_info=None
)
original_created = record.created
result = log_filter.filter(record)
assert result is True
assert record.created == original_created
def test_filter_with_index_key():
"""Test filter method when record has INDEX_KEY."""
log_filter = ChatMessageListTimestampFilter()
record = logging.LogRecord(
name="test", level=logging.INFO, pathname="", lineno=0, msg="test message", args=(), exc_info=None
)
original_created = record.created
# Add the index key
setattr(record, ChatMessageListTimestampFilter.INDEX_KEY, 5)
result = log_filter.filter(record)
assert result is True
# Should increment by 5 microseconds (5 * 1e-6)
assert record.created == original_created + 5 * 1e-6
def test_index_key_constant():
"""Test that INDEX_KEY constant is correctly defined."""
assert ChatMessageListTimestampFilter.INDEX_KEY == "chat_message_index"
# region Test get_function_span
def test_start_span_basic(span_exporter: InMemorySpanExporter):
"""Test starting a span with basic function info."""
# Create a mock function
mock_function = Mock()
mock_function.name = "test_function"
mock_function.description = "Test function description"
attributes = {
OtelAttr.OPERATION: OtelAttr.TOOL_EXECUTION_OPERATION,
OtelAttr.TOOL_NAME: "test_function",
OtelAttr.TOOL_DESCRIPTION: "Test function description",
OtelAttr.TOOL_TYPE: "function",
}
span_exporter.clear()
with get_function_span(attributes) as function_span:
assert function_span is not None
function_span.set_attribute("test_attr", "test_value")
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert span.name == "execute_tool test_function"
assert span.attributes["test_attr"] == "test_value"
assert span.attributes[OtelAttr.OPERATION.value] == OtelAttr.TOOL_EXECUTION_OPERATION
assert span.attributes[OtelAttr.TOOL_NAME] == "test_function"
assert span.attributes[OtelAttr.TOOL_DESCRIPTION] == "Test function description"
def test_start_span_with_tool_call_id(span_exporter: InMemorySpanExporter):
"""Test starting a span with tool_call_id."""
tool_call_id = "test_call_123"
attributes = {
OtelAttr.OPERATION: OtelAttr.TOOL_EXECUTION_OPERATION,
OtelAttr.TOOL_NAME: "test_function",
OtelAttr.TOOL_DESCRIPTION: "Test function",
OtelAttr.TOOL_TYPE: "function",
OtelAttr.TOOL_CALL_ID: tool_call_id,
}
span_exporter.clear()
with get_function_span(attributes) as function_span:
assert function_span is not None
function_span.set_attribute("test_attr", "test_value")
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert span.name == "execute_tool test_function"
assert span.attributes["test_attr"] == "test_value"
assert span.attributes[OtelAttr.TOOL_CALL_ID] == tool_call_id
# Verify all attributes
assert span.attributes[OtelAttr.OPERATION.value] == OtelAttr.TOOL_EXECUTION_OPERATION
assert span.attributes[OtelAttr.TOOL_NAME] == "test_function"
assert span.attributes[OtelAttr.TOOL_DESCRIPTION] == "Test function"
assert span.attributes[OtelAttr.TOOL_TYPE] == "function"
# region Test use_observability decorator
def test_decorator_with_valid_class():
"""Test that decorator works with a valid BaseChatClient-like class."""
# Create a mock class with the required methods
class MockChatClient:
async def get_response(self, messages, **kwargs):
return Mock()
async def get_streaming_response(self, messages, **kwargs):
async def gen():
yield Mock()
return gen()
# Apply the decorator
decorated_class = use_observability(MockChatClient)
assert hasattr(decorated_class, OPEN_TELEMETRY_CHAT_CLIENT_MARKER)
def test_decorator_with_missing_methods():
"""Test that decorator handles classes missing required methods gracefully."""
class MockChatClient:
OTEL_PROVIDER_NAME = "test_provider"
# Apply the decorator - should not raise an error
with pytest.raises(ChatClientInitializationError):
use_observability(MockChatClient)
def test_decorator_with_partial_methods():
"""Test decorator when only one method is present."""
class MockChatClient:
OTEL_PROVIDER_NAME = "test_provider"
async def get_response(self, messages, **kwargs):
return Mock()
with pytest.raises(ChatClientInitializationError):
use_observability(MockChatClient)
# region Test telemetry decorator with mock client
@pytest.fixture
def mock_chat_client():
"""Create a mock chat client for testing."""
class MockChatClient(BaseChatClient):
def service_url(self):
return "https://test.example.com"
async def _inner_get_response(
self, *, messages: MutableSequence[ChatMessage], chat_options: ChatOptions, **kwargs: Any
):
return ChatResponse(
messages=[ChatMessage(role=Role.ASSISTANT, text="Test response")],
usage_details=UsageDetails(input_token_count=10, output_token_count=20),
finish_reason=None,
)
async def _inner_get_streaming_response(
self, *, messages: MutableSequence[ChatMessage], chat_options: ChatOptions, **kwargs: Any
):
yield ChatResponseUpdate(text="Hello", role=Role.ASSISTANT)
yield ChatResponseUpdate(text=" world", role=Role.ASSISTANT)
return MockChatClient
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_chat_client_observability(mock_chat_client, span_exporter: InMemorySpanExporter, enable_sensitive_data):
"""Test that when diagnostics are enabled, telemetry is applied."""
client = use_observability(mock_chat_client)()
messages = [ChatMessage(role=Role.USER, text="Test message")]
span_exporter.clear()
response = await client.get_response(messages=messages, ai_model_id="Test")
assert response is not None
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert span.name == "chat Test"
assert span.attributes[OtelAttr.OPERATION.value] == OtelAttr.CHAT_COMPLETION_OPERATION
assert span.attributes[SpanAttributes.LLM_REQUEST_MODEL] == "Test"
assert span.attributes[OtelAttr.INPUT_TOKENS] == 10
assert span.attributes[OtelAttr.OUTPUT_TOKENS] == 20
if enable_sensitive_data:
assert span.attributes[OtelAttr.INPUT_MESSAGES] is not None
assert span.attributes[OtelAttr.OUTPUT_MESSAGES] is not None
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_chat_client_streaming_observability(
mock_chat_client, span_exporter: InMemorySpanExporter, enable_sensitive_data
):
"""Test streaming telemetry through the use_observability decorator."""
client = use_observability(mock_chat_client)()
messages = [ChatMessage(role=Role.USER, text="Test")]
span_exporter.clear()
# Collect all yielded updates
updates = []
async for update in client.get_streaming_response(messages=messages, ai_model_id="Test"):
updates.append(update)
# Verify we got the expected updates, this shouldn't be dependent on otel
assert len(updates) == 2
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert span.name == "chat Test"
assert span.attributes[OtelAttr.OPERATION.value] == OtelAttr.CHAT_COMPLETION_OPERATION
assert span.attributes[SpanAttributes.LLM_REQUEST_MODEL] == "Test"
if enable_sensitive_data:
assert span.attributes[OtelAttr.INPUT_MESSAGES] is not None
assert span.attributes[OtelAttr.OUTPUT_MESSAGES] is not None
def test_prepend_user_agent_with_none_value():
"""Test prepend user agent with None value in headers."""
headers = {"User-Agent": None}
result = prepend_agent_framework_to_user_agent(headers)
# Should handle None gracefully
assert "User-Agent" in result
assert AGENT_FRAMEWORK_USER_AGENT in str(result["User-Agent"])
# region Test use_agent_observability decorator
def test_agent_decorator_with_valid_class():
"""Test that agent decorator works with a valid ChatAgent-like class."""
# Create a mock class with the required methods
class MockChatClientAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
def __init__(self):
self.id = "test_agent_id"
self.name = "test_agent"
self.display_name = "Test Agent"
self.description = "Test agent description"
async def run(self, messages=None, *, thread=None, **kwargs):
return Mock()
async def run_stream(self, messages=None, *, thread=None, **kwargs):
async def gen():
yield Mock()
return gen()
def get_new_thread(self) -> AgentThread:
return AgentThread()
# Apply the decorator
decorated_class = use_agent_observability(MockChatClientAgent)
assert hasattr(decorated_class, OPEN_TELEMETRY_AGENT_MARKER)
def test_agent_decorator_with_missing_methods():
"""Test that agent decorator handles classes missing required methods gracefully."""
class MockAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
# Apply the decorator - should not raise an error
with pytest.raises(AgentInitializationError):
use_agent_observability(MockAgent)
def test_agent_decorator_with_partial_methods():
"""Test agent decorator when only one method is present."""
from agent_framework.observability import use_agent_observability
class MockAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
def __init__(self):
self.id = "test_agent_id"
self.name = "test_agent"
self.display_name = "Test Agent"
async def run(self, messages=None, *, thread=None, **kwargs):
return Mock()
with pytest.raises(AgentInitializationError):
use_agent_observability(MockAgent)
# region Test agent telemetry decorator with mock agent
@pytest.fixture
def mock_chat_agent():
"""Create a mock chat client agent for testing."""
class MockChatClientAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
def __init__(self):
self.id = "test_agent_id"
self.name = "test_agent"
self.display_name = "Test Agent"
self.description = "Test agent description"
async def run(self, messages=None, *, thread=None, **kwargs):
return AgentRunResponse(
messages=[ChatMessage(role=Role.ASSISTANT, text="Agent response")],
usage_details=UsageDetails(input_token_count=15, output_token_count=25),
response_id="test_response_id",
raw_representation=Mock(finish_reason=Mock(value="stop")),
)
async def run_stream(self, messages=None, *, thread=None, **kwargs):
from agent_framework import AgentRunResponseUpdate
yield AgentRunResponseUpdate(text="Hello", role=Role.ASSISTANT)
yield AgentRunResponseUpdate(text=" from agent", role=Role.ASSISTANT)
return MockChatClientAgent
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_agent_instrumentation_enabled(
mock_chat_agent: AgentProtocol, span_exporter: InMemorySpanExporter, enable_sensitive_data
):
"""Test that when agent diagnostics are enabled, telemetry is applied."""
agent = use_agent_observability(mock_chat_agent)()
span_exporter.clear()
response = await agent.run("Test message")
assert response is not None
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert span.name == "invoke_agent Test Agent"
assert span.attributes[OtelAttr.OPERATION.value] == OtelAttr.AGENT_INVOKE_OPERATION
assert span.attributes[OtelAttr.AGENT_ID] == "test_agent_id"
assert span.attributes[OtelAttr.AGENT_NAME] == "Test Agent"
assert span.attributes[OtelAttr.AGENT_DESCRIPTION] == "Test agent description"
assert span.attributes[SpanAttributes.LLM_REQUEST_MODEL] == "unknown"
assert span.attributes[OtelAttr.INPUT_TOKENS] == 15
assert span.attributes[OtelAttr.OUTPUT_TOKENS] == 25
if enable_sensitive_data:
assert span.attributes[OtelAttr.OUTPUT_MESSAGES] is not None
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_agent_streaming_response_with_diagnostics_enabled_via_decorator(
mock_chat_agent: AgentProtocol, span_exporter: InMemorySpanExporter, enable_sensitive_data
):
"""Test agent streaming telemetry through the use_agent_observability decorator."""
agent = use_agent_observability(mock_chat_agent)()
span_exporter.clear()
updates = []
async for update in agent.run_stream("Test message"):
updates.append(update)
# Verify we got the expected updates
assert len(updates) == 2
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert span.name == "invoke_agent Test Agent"
assert span.attributes[OtelAttr.OPERATION.value] == OtelAttr.AGENT_INVOKE_OPERATION
assert span.attributes[OtelAttr.AGENT_ID] == "test_agent_id"
assert span.attributes[OtelAttr.AGENT_NAME] == "Test Agent"
assert span.attributes[OtelAttr.AGENT_DESCRIPTION] == "Test agent description"
assert span.attributes[SpanAttributes.LLM_REQUEST_MODEL] == "unknown"
if enable_sensitive_data:
assert span.attributes.get(OtelAttr.OUTPUT_MESSAGES) is not None # Streaming, so no usage yet
async def test_agent_run_with_exception_handling(mock_chat_agent: AgentProtocol):
"""Test agent run with exception handling."""
async def run_with_error(self, messages=None, *, thread=None, **kwargs):
raise RuntimeError("Agent run error")
mock_chat_agent.run = run_with_error
agent = use_agent_observability(mock_chat_agent)()
from opentelemetry.trace import Span
with (
patch("agent_framework.observability._get_span") as mock_get_span,
):
mock_span = MagicMock(spec=Span)
# Ensure the patched context manager returns mock_span when entered
mock_get_span.return_value.__enter__.return_value = mock_span
# Should raise the exception and call error handler
with pytest.raises(RuntimeError, match="Agent run error"):
await agent.run("Test message")
# Verify error was recorded
# Check that both error attributes were set on the span
mock_span.set_attribute.assert_called_with(OtelAttr.ERROR_TYPE, "RuntimeError")
mock_span.record_exception.assert_called_once()
mock_span.set_status.assert_called_once_with(
status=StatusCode.ERROR, description=repr(RuntimeError("Agent run error"))
)
@@ -1,44 +1,14 @@
# Copyright (c) Microsoft. All rights reserved.
import logging
from collections.abc import MutableSequence
from typing import Any
from unittest.mock import MagicMock, Mock, patch
import pytest
from opentelemetry.semconv_ai import SpanAttributes
from opentelemetry.trace import StatusCode
from unittest.mock import patch
from agent_framework import (
AgentProtocol,
AgentRunResponse,
AgentThread,
BaseChatClient,
ChatMessage,
ChatOptions,
ChatResponse,
ChatResponseUpdate,
Role,
UsageDetails,
)
from agent_framework.exceptions import AgentInitializationError, ChatClientInitializationError
from agent_framework.telemetry import (
AGENT_FRAMEWORK_USER_AGENT,
OPEN_TELEMETRY_AGENT_MARKER,
OPEN_TELEMETRY_CHAT_CLIENT_MARKER,
ROLE_EVENT_MAP,
USER_AGENT_KEY,
USER_AGENT_TELEMETRY_DISABLED_ENV_VAR,
ChatMessageListTimestampFilter,
OtelAttr,
get_function_span,
prepend_agent_framework_to_user_agent,
use_agent_telemetry,
use_telemetry,
)
from .utils import CopyingMock
# region Test constants
@@ -59,13 +29,13 @@ def test_agent_framework_user_agent_format():
def test_app_info_when_telemetry_enabled():
"""Test that APP_INFO is set when telemetry is enabled."""
with patch("agent_framework.telemetry.IS_TELEMETRY_ENABLED", True):
with patch("agent_framework._telemetry.IS_TELEMETRY_ENABLED", True):
import importlib
import agent_framework.telemetry
import agent_framework._telemetry
importlib.reload(agent_framework.telemetry)
from agent_framework.telemetry import APP_INFO
importlib.reload(agent_framework._telemetry)
from agent_framework import APP_INFO
assert APP_INFO is not None
assert "agent-framework-version" in APP_INFO
@@ -75,7 +45,7 @@ def test_app_info_when_telemetry_enabled():
def test_app_info_when_telemetry_disabled():
"""Test that APP_INFO is None when telemetry is disabled."""
# Test the logic directly since APP_INFO is set at module import time
with patch("agent_framework.telemetry.IS_TELEMETRY_ENABLED", False):
with patch("agent_framework._telemetry.IS_TELEMETRY_ENABLED", False):
# Simulate the module's logic for APP_INFO
test_app_info = (
{
@@ -87,24 +57,6 @@ def test_app_info_when_telemetry_disabled():
assert test_app_info is None
def test_role_event_map():
"""Test that ROLE_EVENT_MAP contains expected mappings."""
assert ROLE_EVENT_MAP["system"] == OtelAttr.SYSTEM_MESSAGE
assert ROLE_EVENT_MAP["user"] == OtelAttr.USER_MESSAGE
assert ROLE_EVENT_MAP["assistant"] == OtelAttr.ASSISTANT_MESSAGE
assert ROLE_EVENT_MAP["tool"] == OtelAttr.TOOL_MESSAGE
def test_enum_values():
"""Test that OtelAttr enum has expected values."""
assert OtelAttr.OPERATION == "gen_ai.operation.name"
assert SpanAttributes.LLM_SYSTEM == "gen_ai.system"
assert SpanAttributes.LLM_REQUEST_MODEL == "gen_ai.request.model"
assert OtelAttr.CHAT_COMPLETION_OPERATION == "chat"
assert OtelAttr.TOOL_EXECUTION_OPERATION == "execute_tool"
assert OtelAttr.AGENT_INVOKE_OPERATION == "invoke_agent"
# region Test prepend_agent_framework_to_user_agent
@@ -144,415 +96,3 @@ def test_modifies_original_dict():
assert result is headers # Same object
assert "User-Agent" in headers
# region Test ChatMessageListTimestampFilter
def test_filter_without_index_key():
"""Test filter method when record doesn't have INDEX_KEY."""
log_filter = ChatMessageListTimestampFilter()
record = logging.LogRecord(
name="test", level=logging.INFO, pathname="", lineno=0, msg="test message", args=(), exc_info=None
)
original_created = record.created
result = log_filter.filter(record)
assert result is True
assert record.created == original_created
def test_filter_with_index_key():
"""Test filter method when record has INDEX_KEY."""
log_filter = ChatMessageListTimestampFilter()
record = logging.LogRecord(
name="test", level=logging.INFO, pathname="", lineno=0, msg="test message", args=(), exc_info=None
)
original_created = record.created
# Add the index key
setattr(record, ChatMessageListTimestampFilter.INDEX_KEY, 5)
result = log_filter.filter(record)
assert result is True
# Should increment by 5 microseconds (5 * 1e-6)
assert record.created == original_created + 5 * 1e-6
def test_index_key_constant():
"""Test that INDEX_KEY constant is correctly defined."""
assert ChatMessageListTimestampFilter.INDEX_KEY == "chat_message_index"
# region Test get_function_span
def test_start_span_basic():
"""Test starting a span with basic function info."""
mock_tracer = Mock()
with patch("agent_framework.telemetry.tracer", mock_tracer):
mock_span = Mock()
mock_tracer.start_as_current_span.return_value = mock_span
# Create a mock function
mock_function = Mock()
mock_function.name = "test_function"
mock_function.description = "Test function description"
attributes = {
OtelAttr.OPERATION: OtelAttr.TOOL_EXECUTION_OPERATION,
OtelAttr.TOOL_NAME: "test_function",
OtelAttr.TOOL_DESCRIPTION: "Test function description",
OtelAttr.TOOL_TYPE: "function",
}
result = get_function_span(attributes)
assert result == mock_span
mock_tracer.start_as_current_span.assert_called_once()
call_args = mock_tracer.start_as_current_span.call_args
assert call_args[1]["name"] == "execute_tool test_function"
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.OPERATION.value] == OtelAttr.TOOL_EXECUTION_OPERATION
assert attributes[OtelAttr.TOOL_NAME] == "test_function"
assert attributes[OtelAttr.TOOL_DESCRIPTION] == "Test function description"
def test_start_span_with_tool_call_id():
"""Test starting a span with tool_call_id."""
mock_tracer = Mock()
with patch("agent_framework.telemetry.tracer", mock_tracer):
mock_span = CopyingMock()
mock_tracer.start_as_current_span.return_value = mock_span
mock_function = Mock()
mock_function.name = "test_function"
mock_function.description = "Test function"
tool_call_id = "test_call_123"
attributes = {
OtelAttr.OPERATION: OtelAttr.TOOL_EXECUTION_OPERATION,
OtelAttr.TOOL_NAME: "test_function",
OtelAttr.TOOL_DESCRIPTION: "Test function",
OtelAttr.TOOL_TYPE: "function",
OtelAttr.TOOL_CALL_ID: tool_call_id,
}
_ = get_function_span(attributes)
call_args = mock_tracer.start_as_current_span.call_args
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.TOOL_CALL_ID] == "test_call_123"
# region Test use_telemetry decorator
def test_decorator_with_valid_class():
"""Test that decorator works with a valid BaseChatClient-like class."""
# Create a mock class with the required methods
class MockChatClient:
async def get_response(self, messages, **kwargs):
return Mock()
async def get_streaming_response(self, messages, **kwargs):
async def gen():
yield Mock()
return gen()
# Apply the decorator
decorated_class = use_telemetry(MockChatClient)
assert hasattr(decorated_class, OPEN_TELEMETRY_CHAT_CLIENT_MARKER)
def test_decorator_with_missing_methods():
"""Test that decorator handles classes missing required methods gracefully."""
class MockChatClient:
OTEL_PROVIDER_NAME = "test_provider"
# Apply the decorator - should not raise an error
with pytest.raises(ChatClientInitializationError):
use_telemetry(MockChatClient)
def test_decorator_with_partial_methods():
"""Test decorator when only one method is present."""
class MockChatClient:
OTEL_PROVIDER_NAME = "test_provider"
async def get_response(self, messages, **kwargs):
return Mock()
with pytest.raises(ChatClientInitializationError):
use_telemetry(MockChatClient)
# region Test telemetry decorator with mock client
@pytest.fixture
def mock_chat_client():
"""Create a mock chat client for testing."""
class MockChatClient(BaseChatClient):
def service_url(self):
return "https://test.example.com"
async def _inner_get_response(
self, *, messages: MutableSequence[ChatMessage], chat_options: ChatOptions, **kwargs: Any
):
return ChatResponse(
messages=[ChatMessage(role=Role.ASSISTANT, text="Test response")],
usage_details=UsageDetails(input_token_count=10, output_token_count=20),
finish_reason=None,
)
async def _inner_get_streaming_response(
self, *, messages: MutableSequence[ChatMessage], chat_options: ChatOptions, **kwargs: Any
):
yield ChatResponseUpdate(text="Hello", role=Role.ASSISTANT)
yield ChatResponseUpdate(text=" world", role=Role.ASSISTANT)
return MockChatClient
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_instrumentation_enabled(mock_chat_client, otel_settings):
"""Test that when diagnostics are enabled, telemetry is applied."""
client = use_telemetry(mock_chat_client)()
messages = [ChatMessage(role=Role.USER, text="Test message")]
chat_options = ChatOptions()
with (
patch("agent_framework.telemetry._get_span") as mock_response_span,
patch("agent_framework.telemetry._capture_messages") as mock_log_messages,
):
response = await client.get_response(messages=messages, chat_options=chat_options)
assert response is not None
mock_response_span.assert_called_once()
# Check that log messages was called only if sensitive events are enabled
assert mock_log_messages.call_count == (2 if otel_settings.enable_sensitive_data else 0)
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_streaming_response_with_otel(mock_chat_client, otel_settings):
"""Test streaming telemetry through the use_telemetry decorator."""
client = use_telemetry(mock_chat_client)()
messages = [ChatMessage(role=Role.USER, text="Test")]
chat_options = ChatOptions()
with (
patch("agent_framework.telemetry._get_span") as mock_response_span,
patch("agent_framework.telemetry._capture_messages") as mock_log_messages,
patch("agent_framework.telemetry._capture_response") as mock_set_output,
):
# Collect all yielded updates
updates = []
async for update in client.get_streaming_response(messages=messages, chat_options=chat_options):
updates.append(update)
# Verify we got the expected updates, this shouldn't be dependent on otel
assert len(updates) == 2
# Verify telemetry calls were made
mock_response_span.assert_called_once()
if otel_settings.enable_sensitive_data:
mock_log_messages.assert_called()
assert mock_log_messages.call_count == 2 # One for input, one for output
else:
mock_log_messages.assert_not_called()
mock_set_output.assert_called_once()
def test_prepend_user_agent_with_none_value():
"""Test prepend user agent with None value in headers."""
headers = {"User-Agent": None}
result = prepend_agent_framework_to_user_agent(headers)
# Should handle None gracefully
assert "User-Agent" in result
assert AGENT_FRAMEWORK_USER_AGENT in str(result["User-Agent"])
# region Test use_agent_telemetry decorator
def test_agent_decorator_with_valid_class():
"""Test that agent decorator works with a valid ChatAgent-like class."""
# Create a mock class with the required methods
class MockChatClientAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
def __init__(self):
self.id = "test_agent_id"
self.name = "test_agent"
self.display_name = "Test Agent"
self.description = "Test agent description"
async def run(self, messages=None, *, thread=None, **kwargs):
return Mock()
async def run_stream(self, messages=None, *, thread=None, **kwargs):
async def gen():
yield Mock()
return gen()
def get_new_thread(self) -> AgentThread:
return AgentThread()
# Apply the decorator
decorated_class = use_agent_telemetry(MockChatClientAgent)
assert hasattr(decorated_class, OPEN_TELEMETRY_AGENT_MARKER)
def test_agent_decorator_with_missing_methods():
"""Test that agent decorator handles classes missing required methods gracefully."""
class MockAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
# Apply the decorator - should not raise an error
with pytest.raises(AgentInitializationError):
use_agent_telemetry(MockAgent)
def test_agent_decorator_with_partial_methods():
"""Test agent decorator when only one method is present."""
from agent_framework.telemetry import use_agent_telemetry
class MockAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
def __init__(self):
self.id = "test_agent_id"
self.name = "test_agent"
self.display_name = "Test Agent"
async def run(self, messages=None, *, thread=None, **kwargs):
return Mock()
with pytest.raises(AgentInitializationError):
use_agent_telemetry(MockAgent)
# region Test agent telemetry decorator with mock agent
@pytest.fixture
def mock_chat_client_agent():
"""Create a mock chat client agent for testing."""
class MockChatClientAgent:
AGENT_SYSTEM_NAME = "test_agent_system"
def __init__(self):
self.id = "test_agent_id"
self.name = "test_agent"
self.display_name = "Test Agent"
self.description = "Test agent description"
async def run(self, messages=None, *, thread=None, **kwargs):
return AgentRunResponse(
messages=[ChatMessage(role=Role.ASSISTANT, text="Agent response")],
usage_details=UsageDetails(input_token_count=15, output_token_count=25),
response_id="test_response_id",
raw_representation=Mock(finish_reason=Mock(value="stop")),
)
async def run_stream(self, messages=None, *, thread=None, **kwargs):
from agent_framework import AgentRunResponseUpdate
yield AgentRunResponseUpdate(text="Hello", role=Role.ASSISTANT)
yield AgentRunResponseUpdate(text=" from agent", role=Role.ASSISTANT)
return MockChatClientAgent
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_agent_instrumentation_enabled(mock_chat_client_agent: AgentProtocol, otel_settings):
"""Test that when agent diagnostics are enabled, telemetry is applied."""
agent = use_agent_telemetry(mock_chat_client_agent)()
with (
patch("agent_framework.telemetry.use_span") as mock_use_span,
patch("agent_framework.telemetry.logger") as mock_logger,
):
response = await agent.run("Test message")
assert response is not None
mock_use_span.assert_called_once()
# Check that logger.info was called (telemetry logs input/output)
assert mock_logger.info.call_count == (2 if otel_settings.enable_sensitive_data else 0)
@pytest.mark.parametrize("enable_sensitive_data", [True, False], indirect=True)
async def test_agent_streaming_response_with_diagnostics_enabled_via_decorator(
mock_chat_client_agent: AgentProtocol, otel_settings
):
"""Test agent streaming telemetry through the use_agent_telemetry decorator."""
agent = use_agent_telemetry(mock_chat_client_agent)()
with (
patch("agent_framework.telemetry._get_span") as mock_get_span,
patch("agent_framework.telemetry._capture_messages") as mock_capture_messages,
patch("agent_framework.telemetry._capture_response") as mock_capture_response,
):
# Collect all yielded updates
updates = []
async for update in agent.run_stream("Test message"):
updates.append(update)
# Verify we got the expected updates
assert len(updates) == 2
# Verify telemetry calls were made
mock_get_span.assert_called_once()
mock_capture_response.assert_called_once()
if otel_settings.enable_sensitive_data:
mock_capture_messages.assert_called()
else:
mock_capture_messages.assert_not_called()
async def test_agent_run_with_exception_handling(mock_chat_client_agent: AgentProtocol):
"""Test agent run with exception handling."""
async def run_with_error(self, messages=None, *, thread=None, **kwargs):
raise RuntimeError("Agent run error")
mock_chat_client_agent.run = run_with_error
agent = use_agent_telemetry(mock_chat_client_agent)()
from opentelemetry.trace import Span
with (
patch("agent_framework.telemetry._get_span") as mock_get_span,
):
mock_span = MagicMock(spec=Span)
# Ensure the patched context manager returns mock_span when entered
mock_get_span.return_value.__enter__.return_value = mock_span
# Should raise the exception and call error handler
with pytest.raises(RuntimeError, match="Agent run error"):
await agent.run("Test message")
# Verify error was recorded
# Check that both error attributes were set on the span
mock_span.set_attribute.assert_called_with(OtelAttr.ERROR_TYPE, "RuntimeError")
mock_span.record_exception.assert_called_once()
mock_span.set_status.assert_called_once_with(
status=StatusCode.ERROR, description=repr(RuntimeError("Agent run error"))
)
+141 -149
View File
@@ -1,9 +1,10 @@
# Copyright (c) Microsoft. All rights reserved.
from typing import Any
from unittest.mock import Mock, patch
from unittest.mock import Mock
import pytest
from opentelemetry import trace
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from pydantic import BaseModel
from agent_framework import (
@@ -15,9 +16,7 @@ from agent_framework import (
)
from agent_framework._tools import _parse_inputs
from agent_framework.exceptions import ToolException
from agent_framework.telemetry import OtelAttr
from .utils import CopyingMock
from agent_framework.observability import OtelAttr
# region AIFunction and ai_function decorator tests
@@ -85,8 +84,7 @@ async def test_ai_function_decorator_with_async():
assert (await async_test_tool(1, 2)) == 3
@pytest.mark.parametrize("enable_sensitive_data", [True], indirect=True)
async def test_ai_function_invoke_telemetry_enabled(otel_settings):
async def test_ai_function_invoke_telemetry_enabled(span_exporter: InMemorySpanExporter):
"""Test the ai_function invoke method with telemetry enabled."""
@ai_function(
@@ -97,52 +95,83 @@ async def test_ai_function_invoke_telemetry_enabled(otel_settings):
"""A function that adds two numbers for telemetry testing."""
return x + y
# Mock the tracer and span
with (
patch("agent_framework.telemetry.tracer"),
# the span creation uses a form of deepcopy, so need to mock that way
patch("agent_framework._tools.get_function_span", new_callable=CopyingMock) as mock_start_span,
):
mock_span = Mock()
mock_context_manager = Mock()
mock_context_manager.__enter__ = Mock(return_value=mock_span)
mock_context_manager.__exit__ = Mock(return_value=None)
mock_start_span.return_value = mock_context_manager
# Mock the histogram
mock_histogram = Mock()
telemetry_test_tool._invocation_duration_histogram = mock_histogram
span_exporter.clear()
# Call invoke
result = await telemetry_test_tool.invoke(x=1, y=2, tool_call_id="test_call_id")
# Mock the histogram
mock_histogram = Mock()
telemetry_test_tool._invocation_duration_histogram = mock_histogram
# Verify result
assert result == 3
# Call invoke
result = await telemetry_test_tool.invoke(x=1, y=2, tool_call_id="test_call_id")
# Verify telemetry calls
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert OtelAttr.TOOL_EXECUTION_OPERATION.value in span.name
assert "telemetry_test_tool" in span.name
assert span.attributes[OtelAttr.TOOL_NAME] == "telemetry_test_tool"
assert span.attributes[OtelAttr.TOOL_CALL_ID] == "test_call_id"
assert span.attributes[OtelAttr.TOOL_TYPE] == "function"
assert span.attributes[OtelAttr.TOOL_DESCRIPTION] == "A test tool for telemetry"
assert span.attributes[OtelAttr.TOOL_ARGUMENTS] == '{"x": 1, "y": 2}'
assert span.attributes[OtelAttr.TOOL_RESULT] == "3"
# Verify result
assert result == 3
# Verify telemetry calls
mock_start_span.assert_called_once_with(
attributes={
OtelAttr.OPERATION: OtelAttr.TOOL_EXECUTION_OPERATION,
OtelAttr.TOOL_NAME: "telemetry_test_tool",
OtelAttr.TOOL_CALL_ID: "test_call_id",
OtelAttr.TOOL_TYPE: "function",
OtelAttr.TOOL_DESCRIPTION: "A test tool for telemetry",
OtelAttr.TOOL_ARGUMENTS: '{"x": 1, "y": 2}',
}
)
assert mock_span.set_attribute.call_count == 2
# Verify histogram was called with correct attributes
mock_histogram.record.assert_called_once()
call_args = mock_histogram.record.call_args
assert call_args[0][0] > 0 # duration should be positive
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.MEASUREMENT_FUNCTION_TAG_NAME] == "telemetry_test_tool"
assert attributes[OtelAttr.TOOL_CALL_ID] == "test_call_id"
# Verify histogram was called with correct attributes
mock_histogram.record.assert_called_once()
call_args = mock_histogram.record.call_args
assert call_args[0][0] > 0 # duration should be positive
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.MEASUREMENT_FUNCTION_TAG_NAME] == "telemetry_test_tool"
assert attributes[OtelAttr.TOOL_CALL_ID] == "test_call_id"
@pytest.mark.parametrize("enable_sensitive_data", [True], indirect=True)
async def test_ai_function_invoke_telemetry_with_pydantic_args(otel_settings):
@pytest.mark.parametrize("enable_sensitive_data", [False], indirect=True)
async def test_ai_function_invoke_telemetry_sensitive_disabled(span_exporter: InMemorySpanExporter):
"""Test the ai_function invoke method with telemetry enabled."""
@ai_function(
name="telemetry_test_tool",
description="A test tool for telemetry",
)
def telemetry_test_tool(x: int, y: int) -> int:
"""A function that adds two numbers for telemetry testing."""
return x + y
# Mock the histogram
mock_histogram = Mock()
telemetry_test_tool._invocation_duration_histogram = mock_histogram
span_exporter.clear()
# Call invoke
result = await telemetry_test_tool.invoke(x=1, y=2, tool_call_id="test_call_id")
# Verify result
assert result == 3
# Verify telemetry calls
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert OtelAttr.TOOL_EXECUTION_OPERATION.value in span.name
assert "telemetry_test_tool" in span.name
assert span.attributes[OtelAttr.TOOL_NAME] == "telemetry_test_tool"
assert span.attributes[OtelAttr.TOOL_CALL_ID] == "test_call_id"
assert span.attributes[OtelAttr.TOOL_TYPE] == "function"
assert span.attributes[OtelAttr.TOOL_DESCRIPTION] == "A test tool for telemetry"
assert OtelAttr.TOOL_ARGUMENTS not in span.attributes
assert OtelAttr.TOOL_RESULT not in span.attributes
# Verify histogram was called with correct attributes
mock_histogram.record.assert_called_once()
call_args = mock_histogram.record.call_args
assert call_args[0][0] > 0 # duration should be positive
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.MEASUREMENT_FUNCTION_TAG_NAME] == "telemetry_test_tool"
assert attributes[OtelAttr.TOOL_CALL_ID] == "test_call_id"
async def test_ai_function_invoke_telemetry_with_pydantic_args(span_exporter: InMemorySpanExporter):
"""Test the ai_function invoke method with Pydantic model arguments."""
@ai_function(
@@ -156,42 +185,27 @@ async def test_ai_function_invoke_telemetry_with_pydantic_args(otel_settings):
# Create arguments as Pydantic model instance
args_model = pydantic_test_tool.input_model(x=5, y=10)
with (
patch("agent_framework.telemetry.tracer"),
# the span creation uses a form of deepcopy, so need to mock that way
patch("agent_framework._tools.get_function_span", new_callable=CopyingMock) as mock_start_span,
):
mock_span = Mock()
mock_context_manager = Mock()
mock_context_manager.__enter__ = Mock(return_value=mock_span)
mock_context_manager.__exit__ = Mock(return_value=None)
mock_start_span.return_value = mock_context_manager
mock_histogram = Mock()
pydantic_test_tool._invocation_duration_histogram = mock_histogram
span_exporter.clear()
# Call invoke with Pydantic model
result = await pydantic_test_tool.invoke(arguments=args_model, tool_call_id="pydantic_call")
mock_histogram = Mock()
pydantic_test_tool._invocation_duration_histogram = mock_histogram
# Call invoke with Pydantic model
result = await pydantic_test_tool.invoke(arguments=args_model, tool_call_id="pydantic_call")
# Verify result
assert result == 15
# Verify telemetry calls
mock_start_span.assert_called_once_with(
attributes={
OtelAttr.OPERATION: OtelAttr.TOOL_EXECUTION_OPERATION,
OtelAttr.TOOL_NAME: "pydantic_test_tool",
OtelAttr.TOOL_CALL_ID: "pydantic_call",
OtelAttr.TOOL_TYPE: "function",
OtelAttr.TOOL_DESCRIPTION: "A test tool with Pydantic args",
OtelAttr.TOOL_ARGUMENTS: '{"x":5,"y":10}',
}
)
assert mock_span.set_attribute.call_count == 2
# Verify result
assert result == 15
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert OtelAttr.TOOL_EXECUTION_OPERATION.value in span.name
assert "pydantic_test_tool" in span.name
assert span.attributes[OtelAttr.TOOL_NAME] == "pydantic_test_tool"
assert span.attributes[OtelAttr.TOOL_CALL_ID] == "pydantic_call"
assert span.attributes[OtelAttr.TOOL_TYPE] == "function"
assert span.attributes[OtelAttr.TOOL_DESCRIPTION] == "A test tool with Pydantic args"
assert span.attributes[OtelAttr.TOOL_ARGUMENTS] == '{"x":5,"y":10}'
@pytest.mark.parametrize("otel_settings", [(True, True)], indirect=True)
async def test_ai_function_invoke_telemetry_with_exception(otel_settings):
async def test_ai_function_invoke_telemetry_with_exception(span_exporter: InMemorySpanExporter):
"""Test the ai_function invoke method with telemetry when an exception occurs."""
@ai_function(
@@ -202,41 +216,33 @@ async def test_ai_function_invoke_telemetry_with_exception(otel_settings):
"""A function that raises an exception for telemetry testing."""
raise ValueError("Test exception for telemetry")
with (
patch("agent_framework.telemetry.tracer"),
# the span creation uses a form of deepcopy, so need to mock that way
patch("agent_framework._tools.get_function_span", new_callable=CopyingMock) as mock_start_span,
):
mock_span = Mock()
mock_context_manager = Mock()
mock_context_manager.__enter__ = Mock(return_value=mock_span)
mock_context_manager.__exit__ = Mock(return_value=None)
mock_start_span.return_value = mock_context_manager
mock_histogram = Mock()
exception_test_tool._invocation_duration_histogram = mock_histogram
span_exporter.clear()
# Call invoke and expect exception
with pytest.raises(ValueError, match="Test exception for telemetry"):
await exception_test_tool.invoke(x=1, y=2, tool_call_id="exception_call")
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert OtelAttr.TOOL_EXECUTION_OPERATION.value in span.name
assert "exception_test_tool" in span.name
assert span.attributes[OtelAttr.TOOL_NAME] == "exception_test_tool"
assert span.attributes[OtelAttr.TOOL_CALL_ID] == "exception_call"
assert span.attributes[OtelAttr.TOOL_TYPE] == "function"
assert span.attributes[OtelAttr.TOOL_DESCRIPTION] == "A test tool that raises an exception"
assert span.attributes[OtelAttr.TOOL_ARGUMENTS] == '{"x": 1, "y": 2}'
assert span.attributes[OtelAttr.ERROR_TYPE] == ValueError.__name__
assert span.status.status_code == trace.StatusCode.ERROR
mock_histogram = Mock()
exception_test_tool._invocation_duration_histogram = mock_histogram
# Call invoke and expect exception
with pytest.raises(ValueError, match="Test exception for telemetry"):
await exception_test_tool.invoke(x=1, y=2, tool_call_id="exception_call")
# Verify telemetry calls
mock_start_span.assert_called_once()
# Verify span exception recording
mock_span.record_exception.assert_called_once()
mock_span.set_attribute.assert_called()
mock_span.set_status.assert_called_once()
# Verify histogram was called with error attributes
mock_histogram.record.assert_called_once()
call_args = mock_histogram.record.call_args
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.ERROR_TYPE] == ValueError.__name__
# Verify histogram was called with error attributes
mock_histogram.record.assert_called_once()
call_args = mock_histogram.record.call_args
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.ERROR_TYPE] == ValueError.__name__
@pytest.mark.parametrize("enable_sensitive_data", [True], indirect=True)
async def test_ai_function_invoke_telemetry_async_function(otel_settings):
async def test_ai_function_invoke_telemetry_async_function(span_exporter: InMemorySpanExporter):
"""Test the ai_function invoke method with telemetry on async function."""
@ai_function(
@@ -247,44 +253,30 @@ async def test_ai_function_invoke_telemetry_async_function(otel_settings):
"""An async function for telemetry testing."""
return x * y
with (
patch("agent_framework.telemetry.tracer"),
# the span creation uses a form of deepcopy, so need to mock that way
patch("agent_framework._tools.get_function_span", new_callable=CopyingMock) as mock_start_span,
):
mock_span = Mock()
mock_context_manager = Mock()
mock_context_manager.__enter__ = Mock(return_value=mock_span)
mock_context_manager.__exit__ = Mock(return_value=None)
mock_start_span.return_value = mock_context_manager
mock_histogram = Mock()
async_telemetry_test._invocation_duration_histogram = mock_histogram
span_exporter.clear()
# Call invoke
result = await async_telemetry_test.invoke(x=3, y=4, tool_call_id="async_call")
mock_histogram = Mock()
async_telemetry_test._invocation_duration_histogram = mock_histogram
# Verify result
assert result == 12
spans = span_exporter.get_finished_spans()
assert len(spans) == 1
span = spans[0]
assert OtelAttr.TOOL_EXECUTION_OPERATION.value in span.name
assert "async_telemetry_test" in span.name
assert span.attributes[OtelAttr.TOOL_NAME] == "async_telemetry_test"
assert span.attributes[OtelAttr.TOOL_CALL_ID] == "async_call"
assert span.attributes[OtelAttr.TOOL_TYPE] == "function"
assert span.attributes[OtelAttr.TOOL_DESCRIPTION] == "An async test tool for telemetry"
assert span.attributes[OtelAttr.TOOL_ARGUMENTS] == '{"x": 3, "y": 4}'
# Call invoke
result = await async_telemetry_test.invoke(x=3, y=4, tool_call_id="async_call")
# Verify result
assert result == 12
# Verify telemetry calls
mock_start_span.assert_called_once_with(
attributes={
OtelAttr.OPERATION: OtelAttr.TOOL_EXECUTION_OPERATION,
OtelAttr.TOOL_NAME: "async_telemetry_test",
OtelAttr.TOOL_CALL_ID: "async_call",
OtelAttr.TOOL_TYPE: "function",
OtelAttr.TOOL_DESCRIPTION: "An async test tool for telemetry",
OtelAttr.TOOL_ARGUMENTS: '{"x": 3, "y": 4}',
}
)
assert mock_span.set_attribute.call_count == 2
# Verify histogram recording
mock_histogram.record.assert_called_once()
call_args = mock_histogram.record.call_args
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.MEASUREMENT_FUNCTION_TAG_NAME] == "async_telemetry_test"
# Verify histogram recording
mock_histogram.record.assert_called_once()
call_args = mock_histogram.record.call_args
attributes = call_args[1]["attributes"]
assert attributes[OtelAttr.MEASUREMENT_FUNCTION_TAG_NAME] == "async_telemetry_test"
async def test_ai_function_invoke_invalid_pydantic_args():
@@ -3,8 +3,6 @@ from typing import Any
from pytest import fixture
from agent_framework.telemetry import OtelSettings, setup_telemetry
# region Connector Settings fixtures
@fixture
@@ -51,26 +49,3 @@ def openai_unit_test_env(monkeypatch, exclude_list, override_env_param_dict): #
monkeypatch.setenv(key, value) # type: ignore
return env_vars
@fixture
def enable_otel(request: Any) -> bool:
"""Fixture that returns a boolean indicating if Otel is enabled."""
return request.param if hasattr(request, "param") else True
@fixture
def enable_sensitive_data(request: Any) -> bool:
"""Fixture that returns a boolean indicating if sensitive data is enabled."""
return request.param if hasattr(request, "param") else False
@fixture
def otel_settings(enable_otel: bool, enable_sensitive_data: bool) -> OtelSettings:
"""Fixture to set environment variables for OtelSettings."""
from agent_framework.telemetry import OTEL_SETTINGS
setup_telemetry(enable_otel=enable_otel, enable_sensitive_data=enable_sensitive_data)
return OTEL_SETTINGS
@@ -783,7 +783,7 @@ def test_create_streaming_response_content_with_mcp_approval_request() -> None:
@pytest.mark.parametrize("enable_otel", [False], indirect=True)
@pytest.mark.parametrize("enable_sensitive_data", [False], indirect=True)
def test_end_to_end_mcp_approval_flow(otel_settings) -> None:
def test_end_to_end_mcp_approval_flow() -> None:
"""End-to-end mocked test:
model issues an mcp_approval_request, user approves, client sends mcp_approval_response.
"""
@@ -0,0 +1 @@
# Copyright (c) Microsoft. All rights reserved.
@@ -5,9 +5,6 @@ from typing import Any
from unittest.mock import patch
import pytest
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from agent_framework import (
Executor,
@@ -27,54 +24,7 @@ from agent_framework._workflow._edge import (
SwitchCaseEdgeGroupDefault,
)
from agent_framework._workflow._edge_runner import create_edge_runner
from agent_framework._workflow._telemetry import EdgeGroupDeliveryStatus, workflow_tracer
@pytest.fixture
def tracing_enabled():
"""Enable tracing for tests."""
import os
original_value = os.environ.get("AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL_DIAGNOSTICS")
os.environ["AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL_DIAGNOSTICS"] = "true"
# Force reload the settings to pick up the environment variable
from agent_framework._workflow._telemetry import WorkflowDiagnosticSettings
workflow_tracer.settings = WorkflowDiagnosticSettings()
yield
# Restore original value
if original_value is None:
os.environ.pop("AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL_DIAGNOSTICS", None)
else:
os.environ["AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL_DIAGNOSTICS"] = original_value
# Reload settings again
workflow_tracer.settings = WorkflowDiagnosticSettings()
@pytest.fixture
def span_exporter(tracing_enabled):
"""Set up OpenTelemetry test infrastructure."""
# Use the built-in InMemorySpanExporter for better compatibility
exporter = InMemorySpanExporter()
tracer_provider = TracerProvider()
tracer_provider.add_span_processor(SimpleSpanProcessor(exporter))
# Store original tracer
original_tracer = workflow_tracer.tracer
# Set up our test tracer
workflow_tracer.tracer = tracer_provider.get_tracer("agent_framework")
yield exporter
# Clean up
exporter.clear()
workflow_tracer.tracer = original_tracer
from agent_framework.observability import EdgeGroupDeliveryStatus
@dataclass
@@ -1,66 +1,22 @@
# Copyright (c) Microsoft. All rights reserved.
import os
from collections.abc import Generator
from typing import Any, cast
import pytest
from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from agent_framework import WorkflowBuilder
from agent_framework._workflow._executor import Executor, handler
from agent_framework._workflow._runner_context import InProcRunnerContext, Message
from agent_framework._workflow._shared_state import SharedState
from agent_framework._workflow._telemetry import WorkflowTracer, workflow_tracer
from agent_framework._workflow._workflow import Workflow
from agent_framework._workflow._workflow_context import WorkflowContext
@pytest.fixture
def tracing_enabled() -> Generator[None, None, None]:
"""Enable tracing for tests."""
original_value = os.environ.get("AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL")
os.environ["AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL"] = "true"
# Force reload the settings to pick up the environment variable
from agent_framework._workflow._telemetry import WorkflowDiagnosticSettings
workflow_tracer.settings = WorkflowDiagnosticSettings()
yield
# Restore original value
if original_value is None:
os.environ.pop("AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL", None)
else:
os.environ["AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL"] = original_value
# Reload settings again
workflow_tracer.settings = WorkflowDiagnosticSettings()
@pytest.fixture
def span_exporter(tracing_enabled: Any) -> Generator[InMemorySpanExporter, None, None]:
"""Set up OpenTelemetry test infrastructure."""
# Use the built-in InMemorySpanExporter for better compatibility
exporter = InMemorySpanExporter()
tracer_provider = TracerProvider()
tracer_provider.add_span_processor(SimpleSpanProcessor(exporter))
# Store original tracer
original_tracer = workflow_tracer.tracer
# Set up our test tracer
workflow_tracer.tracer = tracer_provider.get_tracer("agent_framework")
yield exporter
# Clean up
exporter.clear()
workflow_tracer.tracer = original_tracer
from agent_framework.observability import (
OtelAttr,
create_processing_span,
create_workflow_span,
)
class MockExecutor(Executor):
@@ -142,34 +98,7 @@ class FanInAggregator(Executor):
return self._processed_messages
async def test_workflow_tracer_configuration() -> None:
"""Test that workflow tracer can be enabled and disabled."""
# Test disabled by default
tracer = WorkflowTracer()
assert not tracer.enabled
# Test enabled with environment variable
original_value = os.environ.get("AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL")
os.environ["AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL"] = "true"
# Force reload the settings to pick up the environment variable
from agent_framework._workflow._telemetry import WorkflowDiagnosticSettings
tracer.settings = WorkflowDiagnosticSettings()
assert tracer.enabled
# Restore original value
if original_value is None:
os.environ.pop("AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL", None)
else:
os.environ["AGENT_FRAMEWORK_WORKFLOW_ENABLE_OTEL"] = original_value
# Reload settings again
tracer.settings = WorkflowDiagnosticSettings()
async def test_span_creation_and_attributes(tracing_enabled: Any, span_exporter: InMemorySpanExporter) -> None:
async def test_span_creation_and_attributes(span_exporter: InMemorySpanExporter) -> None:
"""Test creation and attributes of all span types (workflow, processing, sending)."""
# Create a mock workflow object
mock_workflow = cast(
@@ -186,12 +115,22 @@ async def test_span_creation_and_attributes(tracing_enabled: Any, span_exporter:
)
# Test all span types in nested context
with workflow_tracer.create_workflow_run_span(mock_workflow) as workflow_span:
workflow_tracer.add_workflow_event("workflow.started")
with create_workflow_span(
OtelAttr.WORKFLOW_RUN_SPAN,
{
OtelAttr.WORKFLOW_ID: mock_workflow.id,
},
) as workflow_span:
workflow_span.add_event(OtelAttr.WORKFLOW_STARTED)
sending_attributes = {
OtelAttr.MESSAGE_TYPE: "ResponseMessage",
OtelAttr.MESSAGE_DESTINATION_EXECUTOR_ID: "target-789",
}
with (
workflow_tracer.create_processing_span("executor-456", "TestExecutor", "TestMessage") as processing_span,
workflow_tracer.create_sending_span("ResponseMessage", "target-789") as sending_span,
create_processing_span("executor-456", "TestExecutor", "TestMessage") as processing_span,
create_workflow_span(
OtelAttr.MESSAGE_SEND_SPAN, sending_attributes, kind=trace.SpanKind.PRODUCER
) as sending_span,
):
# Verify all spans are recording
assert workflow_span is not None and workflow_span.is_recording()
@@ -205,7 +144,7 @@ async def test_span_creation_and_attributes(tracing_enabled: Any, span_exporter:
workflow_span = next(s for s in spans if s.name == "workflow.run")
assert workflow_span.kind == trace.SpanKind.INTERNAL
assert workflow_span.attributes is not None
assert workflow_span.attributes.get("workflow.id") == "test-workflow-123"
assert workflow_span.attributes.get(OtelAttr.WORKFLOW_ID) == "test-workflow-123"
assert workflow_span.events is not None
event_names = [event.name for event in workflow_span.events]
assert "workflow.started" in event_names
@@ -226,12 +165,14 @@ async def test_span_creation_and_attributes(tracing_enabled: Any, span_exporter:
assert sending_span.attributes.get("message.destination_executor_id") == "target-789"
async def test_trace_context_handling(tracing_enabled: Any, span_exporter: InMemorySpanExporter) -> None:
async def test_trace_context_handling(span_exporter: InMemorySpanExporter) -> None:
"""Test trace context propagation and handling in messages and executors."""
shared_state = SharedState()
ctx = InProcRunnerContext()
executor = MockExecutor("test-executor")
span_exporter.clear()
# Test trace context propagation in messages
workflow_ctx: WorkflowContext[str] = WorkflowContext(
"test-executor",
@@ -273,7 +214,8 @@ async def test_trace_context_handling(tracing_enabled: Any, span_exporter: InMem
assert processing_span.attributes.get("message.type") == "str"
async def test_trace_context_disabled_when_tracing_disabled() -> None:
@pytest.mark.parametrize("enable_otel", [False], indirect=True)
async def test_trace_context_disabled_when_tracing_disabled(enable_otel, span_exporter: InMemorySpanExporter) -> None:
"""Test that no trace context is added when tracing is disabled."""
# Tracing should be disabled by default
shared_state = SharedState()
@@ -298,7 +240,7 @@ async def test_trace_context_disabled_when_tracing_disabled() -> None:
assert message.source_span_id is None
async def test_end_to_end_workflow_tracing(tracing_enabled: Any, span_exporter: InMemorySpanExporter) -> None:
async def test_end_to_end_workflow_tracing(span_exporter: InMemorySpanExporter) -> None:
"""Test end-to-end tracing including workflow build, execution, and span linking with fan-in edges."""
# Create executors for fan-in scenario
executor1 = MockExecutor("executor1")
@@ -321,7 +263,7 @@ async def test_end_to_end_workflow_tracing(tracing_enabled: Any, span_exporter:
build_span = build_spans[0]
assert build_span.attributes is not None
assert build_span.attributes.get("workflow.id") == workflow.id
assert build_span.attributes.get(OtelAttr.WORKFLOW_ID) == workflow.id
assert build_span.attributes.get("workflow.definition") is not None
definition = build_span.attributes.get("workflow.definition")
assert definition == workflow.model_dump_json(by_alias=True)
@@ -422,7 +364,7 @@ async def test_end_to_end_workflow_tracing(tracing_enabled: Any, span_exporter:
assert len(aggregator_span.links) >= 2, f"Expected at least 2 links, got {len(aggregator_span.links)}"
async def test_workflow_error_handling_in_tracing(tracing_enabled: Any, span_exporter: InMemorySpanExporter) -> None:
async def test_workflow_error_handling_in_tracing(span_exporter: InMemorySpanExporter) -> None:
"""Test that workflow errors are properly recorded in traces."""
class FailingExecutor(Executor):
@@ -457,7 +399,8 @@ async def test_workflow_error_handling_in_tracing(tracing_enabled: Any, span_exp
assert workflow_span.status.status_code.name == "ERROR"
async def test_message_trace_context_serialization() -> None:
@pytest.mark.parametrize("enable_otel", [False], indirect=True)
async def test_message_trace_context_serialization(span_exporter: InMemorySpanExporter) -> None:
"""Test that message trace context is properly serialized/deserialized."""
ctx = InProcRunnerContext()
@@ -491,7 +434,7 @@ async def test_message_trace_context_serialization() -> None:
assert restored_msg.source_span_ids == ["span123"] # Test new format
async def test_workflow_build_error_tracing(tracing_enabled: Any, span_exporter: InMemorySpanExporter) -> None:
async def test_workflow_build_error_tracing(span_exporter: InMemorySpanExporter) -> None:
"""Test that build errors are properly recorded in build spans."""
# Test validation error by not setting start executor