Skip to content

Commit a85b401

Browse files
alexander-alderman-webbalexander-alderman-webb
committed
test(openai-agents): Replace mocks with httpx in handoff tests
1 parent 763b16a commit a85b401

File tree

1 file changed

+206
-114
lines changed

1 file changed

+206
-114
lines changed

‎tests/integrations/openai_agents/test_openai_agents.py‎

Lines changed: 206 additions & 114 deletions
Original file line numberDiff line numberDiff line change
@@ -5,10 +5,11 @@
55
import os
66
import json
77
import logging
8+
import httpx
89

910
import sentry_sdk
1011
from sentry_sdk import start_span
11-
from sentry_sdk.consts import SPANDATA
12+
from sentry_sdk.consts import SPANDATA, OP
1213
from sentry_sdk.integrations.logging import LoggingIntegration
1314
from sentry_sdk.integrations.openai_agents import OpenAIAgentsIntegration
1415
from sentry_sdk.integrations.openai_agents.utils import _set_input_data, safe_serialize
@@ -314,6 +315,25 @@ def test_agent_custom_model():
314315
)
315316

316317

318+
@pytest.fixture
319+
def get_model_response():
320+
def inner(response_content):
321+
model_request = httpx.Request(
322+
"POST",
323+
"/responses",
324+
)
325+
326+
response = httpx.Response(
327+
200,
328+
request=model_request,
329+
content=json.dumps(response_content.model_dump()).encode("utf-8"),
330+
)
331+
332+
return response
333+
334+
return inner
335+
336+
317337
@pytest.mark.asyncio
318338
async def test_agent_invocation_span_no_pii(
319339
sentry_init, capture_events, test_agent, mock_model_response
@@ -917,85 +937,120 @@ def test_agent_invocation_span_sync(
917937

918938

919939
@pytest.mark.asyncio
920-
async def test_handoff_span(sentry_init, capture_events, mock_usage):
940+
async def test_handoff_span(sentry_init, capture_events, get_model_response):
921941
"""
922942
Test that handoff spans are created when agents hand off to other agents.
923943
"""
944+
client = AsyncOpenAI(api_key="test-key")
945+
model = OpenAIResponsesModel(model="gpt-4-mini", openai_client=client)
946+
924947
# Create two simple agents with a handoff relationship
925948
secondary_agent = agents.Agent(
926949
name="secondary_agent",
927950
instructions="You are a secondary agent.",
928-
model="gpt-4o-mini",
951+
model=model,
929952
)
930953

931954
primary_agent = agents.Agent(
932955
name="primary_agent",
933956
instructions="You are a primary agent that hands off to secondary agent.",
934-
model="gpt-4o-mini",
957+
model=model,
935958
handoffs=[secondary_agent],
936959
)
937960

938-
with patch.dict(os.environ, {"OPENAI_API_KEY": "test-key"}):
939-
with patch(
940-
"agents.models.openai_responses.OpenAIResponsesModel.get_response"
941-
) as mock_get_response:
942-
# Mock two responses:
943-
# 1. Primary agent calls handoff tool
944-
# 2. Secondary agent provides final response
945-
handoff_response = ModelResponse(
946-
output=[
947-
ResponseFunctionToolCall(
948-
id="call_handoff_123",
949-
call_id="call_handoff_123",
950-
name="transfer_to_secondary_agent",
951-
type="function_call",
952-
arguments="{}",
953-
)
954-
],
955-
usage=mock_usage,
956-
response_id="resp_handoff_123",
957-
)
958-
959-
final_response = ModelResponse(
960-
output=[
961-
ResponseOutputMessage(
962-
id="msg_final",
963-
type="message",
964-
status="completed",
965-
content=[
966-
ResponseOutputText(
967-
text="I'm the specialist and I can help with that!",
968-
type="output_text",
969-
annotations=[],
970-
)
971-
],
972-
role="assistant",
973-
)
974-
],
975-
usage=mock_usage,
976-
response_id="resp_final_123",
977-
)
961+
handoff_response = get_model_response(
962+
Response(
963+
id="resp_tool_123",
964+
output=[
965+
ResponseFunctionToolCall(
966+
id="call_handoff_123",
967+
call_id="call_handoff_123",
968+
name="transfer_to_secondary_agent",
969+
type="function_call",
970+
arguments="{}",
971+
)
972+
],
973+
parallel_tool_calls=False,
974+
tool_choice="none",
975+
tools=[],
976+
created_at=10000000,
977+
model="gpt-4",
978+
object="response",
979+
usage=ResponseUsage(
980+
input_tokens=10,
981+
input_tokens_details=InputTokensDetails(
982+
cached_tokens=0,
983+
),
984+
output_tokens=20,
985+
output_tokens_details=OutputTokensDetails(
986+
reasoning_tokens=5,
987+
),
988+
total_tokens=30,
989+
),
990+
)
991+
)
978992

979-
mock_get_response.side_effect = [handoff_response, final_response]
993+
final_response = get_model_response(
994+
Response(
995+
id="resp_final_123",
996+
output=[
997+
ResponseOutputMessage(
998+
id="msg_final",
999+
type="message",
1000+
status="completed",
1001+
content=[
1002+
ResponseOutputText(
1003+
text="I'm the specialist and I can help with that!",
1004+
type="output_text",
1005+
annotations=[],
1006+
)
1007+
],
1008+
role="assistant",
1009+
)
1010+
],
1011+
parallel_tool_calls=False,
1012+
tool_choice="none",
1013+
tools=[],
1014+
created_at=10000000,
1015+
model="gpt-4",
1016+
object="response",
1017+
usage=ResponseUsage(
1018+
input_tokens=10,
1019+
input_tokens_details=InputTokensDetails(
1020+
cached_tokens=0,
1021+
),
1022+
output_tokens=20,
1023+
output_tokens_details=OutputTokensDetails(
1024+
reasoning_tokens=5,
1025+
),
1026+
total_tokens=30,
1027+
),
1028+
)
1029+
)
9801030

981-
sentry_init(
982-
integrations=[OpenAIAgentsIntegration()],
983-
traces_sample_rate=1.0,
984-
)
1031+
with patch.object(
1032+
primary_agent.model._client._client,
1033+
"send",
1034+
side_effect=[handoff_response, final_response],
1035+
) as _:
1036+
sentry_init(
1037+
integrations=[OpenAIAgentsIntegration()],
1038+
traces_sample_rate=1.0,
1039+
)
9851040

986-
events = capture_events()
1041+
events = capture_events()
9871042

988-
result = await agents.Runner.run(
989-
primary_agent,
990-
"Please hand off to secondary agent",
991-
run_config=test_run_config,
992-
)
1043+
result = await agents.Runner.run(
1044+
primary_agent,
1045+
"Please hand off to secondary agent",
1046+
run_config=test_run_config,
1047+
)
9931048

994-
assert result is not None
1049+
assert result is not None
9951050

9961051
(transaction,) = events
9971052
spans = transaction["spans"]
998-
handoff_span = spans[2]
1053+
handoff_span = next(span for span in spans if span.get("op") == OP.GEN_AI_HANDOFF)
9991054

10001055
# Verify handoff span was created
10011056
assert handoff_span is not None
@@ -1006,85 +1061,122 @@ async def test_handoff_span(sentry_init, capture_events, mock_usage):
10061061

10071062

10081063
@pytest.mark.asyncio
1009-
async def test_max_turns_before_handoff_span(sentry_init, capture_events, mock_usage):
1064+
async def test_max_turns_before_handoff_span(
1065+
sentry_init, capture_events, mock_usage, get_model_response
1066+
):
10101067
"""
10111068
Example raising agents.exceptions.AgentsException after the agent invocation span is complete.
10121069
"""
1070+
client = AsyncOpenAI(api_key="test-key")
1071+
model = OpenAIResponsesModel(model="gpt-4-mini", openai_client=client)
1072+
10131073
# Create two simple agents with a handoff relationship
10141074
secondary_agent = agents.Agent(
10151075
name="secondary_agent",
10161076
instructions="You are a secondary agent.",
1017-
model="gpt-4o-mini",
1077+
model=model,
10181078
)
10191079

10201080
primary_agent = agents.Agent(
10211081
name="primary_agent",
10221082
instructions="You are a primary agent that hands off to secondary agent.",
1023-
model="gpt-4o-mini",
1083+
model=model,
10241084
handoffs=[secondary_agent],
10251085
)
10261086

1027-
with patch.dict(os.environ, {"OPENAI_API_KEY": "test-key"}):
1028-
with patch(
1029-
"agents.models.openai_responses.OpenAIResponsesModel.get_response"
1030-
) as mock_get_response:
1031-
# Mock two responses:
1032-
# 1. Primary agent calls handoff tool
1033-
# 2. Secondary agent provides final response
1034-
handoff_response = ModelResponse(
1035-
output=[
1036-
ResponseFunctionToolCall(
1037-
id="call_handoff_123",
1038-
call_id="call_handoff_123",
1039-
name="transfer_to_secondary_agent",
1040-
type="function_call",
1041-
arguments="{}",
1042-
)
1043-
],
1044-
usage=mock_usage,
1045-
response_id="resp_handoff_123",
1046-
)
1047-
1048-
final_response = ModelResponse(
1049-
output=[
1050-
ResponseOutputMessage(
1051-
id="msg_final",
1052-
type="message",
1053-
status="completed",
1054-
content=[
1055-
ResponseOutputText(
1056-
text="I'm the specialist and I can help with that!",
1057-
type="output_text",
1058-
annotations=[],
1059-
)
1060-
],
1061-
role="assistant",
1062-
)
1063-
],
1064-
usage=mock_usage,
1065-
response_id="resp_final_123",
1066-
)
1087+
handoff_response = get_model_response(
1088+
Response(
1089+
id="resp_tool_123",
1090+
output=[
1091+
ResponseFunctionToolCall(
1092+
id="call_handoff_123",
1093+
call_id="call_handoff_123",
1094+
name="transfer_to_secondary_agent",
1095+
type="function_call",
1096+
arguments="{}",
1097+
)
1098+
],
1099+
parallel_tool_calls=False,
1100+
tool_choice="none",
1101+
tools=[],
1102+
created_at=10000000,
1103+
model="gpt-4",
1104+
object="response",
1105+
usage=ResponseUsage(
1106+
input_tokens=10,
1107+
input_tokens_details=InputTokensDetails(
1108+
cached_tokens=0,
1109+
),
1110+
output_tokens=20,
1111+
output_tokens_details=OutputTokensDetails(
1112+
reasoning_tokens=5,
1113+
),
1114+
total_tokens=30,
1115+
),
1116+
)
1117+
)
10671118

1068-
mock_get_response.side_effect = [handoff_response, final_response]
1119+
final_response = get_model_response(
1120+
Response(
1121+
id="resp_final_123",
1122+
output=[
1123+
ResponseOutputMessage(
1124+
id="msg_final",
1125+
type="message",
1126+
status="completed",
1127+
content=[
1128+
ResponseOutputText(
1129+
text="I'm the specialist and I can help with that!",
1130+
type="output_text",
1131+
annotations=[],
1132+
)
1133+
],
1134+
role="assistant",
1135+
)
1136+
],
1137+
parallel_tool_calls=False,
1138+
tool_choice="none",
1139+
tools=[],
1140+
created_at=10000000,
1141+
model="gpt-4",
1142+
object="response",
1143+
usage=ResponseUsage(
1144+
input_tokens=10,
1145+
input_tokens_details=InputTokensDetails(
1146+
cached_tokens=0,
1147+
),
1148+
output_tokens=20,
1149+
output_tokens_details=OutputTokensDetails(
1150+
reasoning_tokens=5,
1151+
),
1152+
total_tokens=30,
1153+
),
1154+
)
1155+
)
10691156

1070-
sentry_init(
1071-
integrations=[OpenAIAgentsIntegration()],
1072-
traces_sample_rate=1.0,
1073-
)
1157+
with patch.object(
1158+
primary_agent.model._client._client,
1159+
"send",
1160+
side_effect=[handoff_response, final_response],
1161+
) as _:
1162+
sentry_init(
1163+
integrations=[OpenAIAgentsIntegration()],
1164+
traces_sample_rate=1.0,
1165+
)
10741166

1075-
events = capture_events()
1167+
events = capture_events()
10761168

1077-
with pytest.raises(MaxTurnsExceeded):
1078-
await agents.Runner.run(
1079-
primary_agent,
1080-
"Please hand off to secondary agent",
1081-
run_config=test_run_config,
1082-
max_turns=1,
1083-
)
1169+
with pytest.raises(MaxTurnsExceeded):
1170+
await agents.Runner.run(
1171+
primary_agent,
1172+
"Please hand off to secondary agent",
1173+
run_config=test_run_config,
1174+
max_turns=1,
1175+
)
10841176

10851177
(error, transaction) = events
10861178
spans = transaction["spans"]
1087-
handoff_span = spans[2]
1179+
handoff_span = next(span for span in spans if span.get("op") == OP.GEN_AI_HANDOFF)
10881180

10891181
# Verify handoff span was created
10901182
assert handoff_span is not None

0 commit comments

Comments
 (0)