Skip to content

GitLab

Unverified Commit 7ef1964a authored by Keyang Ru's avatar Keyang Ru Committed by GitHub
Browse files

[router] add basic ci tests for gpt-oss model support (#12651) 


Co-authored-by: default avatarChang Su <chang.s.su@oracle.com>
parent b0476a06
Show whitespace changes
Inline Side-by-side
Showing with 1006 additions and 86 deletions
sgl-router/py_test/e2e_grpc/basic/test_openai_server.py
View file @ 7ef1964a
......@@ -23,6 +23,7 @@ _TEST_DIR = Path(__file__).parent
sys.path.insert(0, str(_TEST_DIR.parent))
from fixtures import popen_launch_workers_and_router
from util import (
DEFAULT_GPT_OSS_MODEL_PATH,
DEFAULT_MODEL_PATH,
DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
DEFAULT_URL_FOR_TEST,
......@@ -293,5 +294,51 @@ The SmartHome Mini is a compact smart home assistant available in black or white
client.models.retrieve("non-existent-model")
class TestOpenAIServerGptOss(TestOpenAIServer):
"""
Test OpenAI API through gRPC router with openai/gpt-oss-20b model.
Extends TestOpenAIServer and only changes the model.
"""
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_GPT_OSS_MODEL_PATH
cls.base_url = DEFAULT_URL_FOR_TEST
cls.api_key = "sk-123456"
cls.cluster = popen_launch_workers_and_router(
cls.model,
cls.base_url,
timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
num_workers=1,
tp_size=2,
policy="round_robin",
api_key=cls.api_key,
)
cls.base_url += "/v1"
cls.tokenizer = get_tokenizer(cls.model)
def test_chat_completion(self):
for parallel_sample_num in [1, 2]:
self.run_chat_completion(None, parallel_sample_num)
def test_chat_completion_stream(self):
for parallel_sample_num in [1, 2]:
self.run_chat_completion_stream(None, parallel_sample_num)
@unittest.skip("Skipping for OSS models")
def test_regex(self):
super().test_regex()
@unittest.skip("Skipping for OSS models")
def test_response_prefill(self):
super().test_response_prefill()
@unittest.skip("Skipping for OSS models")
def test_penalty(self):
super().test_penalty()
if __name__ == "__main__":
unittest.main()
sgl-router/py_test/e2e_grpc/fixtures.py
View file @ 7ef1964a
......@@ -202,8 +202,6 @@ def popen_launch_workers_and_router(
"--grpc-mode", # Enable gRPC for this worker
"--mem-fraction-static",
"0.8",
"--attention-backend",
"fa3",
]
# Add TP size
......
sgl-router/py_test/e2e_grpc/util.py
View file @ 7ef1964a
......@@ -89,6 +89,9 @@ DEFAULT_MISTRAL_FUNCTION_CALLING_MODEL_PATH = _get_model_path(
"mistralai/Mistral-7B-Instruct-v0.3"
)
# GPT-OSS models
DEFAULT_GPT_OSS_MODEL_PATH = _get_model_path("openai/gpt-oss-20b")
# ============================================================================
# Process Management
......
sgl-router/py_test/e2e_response_api/backends/test_grpc_backend.py 0 → 100644
View file @ 7ef1964a
"""
gRPC backend tests for Response API (including Harmony).
Run with:
python3 -m pytest py_test/e2e_response_api/backends/test_grpc_backend.py -v
python3 -m unittest e2e_response_api.backends.test_grpc_backend.TestGrpcBackend
"""
import sys
import unittest
from pathlib import Path
# Add e2e_response_api directory for imports
_TEST_DIR = Path(__file__).parent.parent
sys.path.insert(0, str(_TEST_DIR))
# Import local modules
from mixins.function_call import FunctionCallingBaseTest
from mixins.mcp import MCPTests
from mixins.state_management import StateManagementTests
from router_fixtures import popen_launch_workers_and_router
from util import kill_process_tree
class TestGrpcBackend(StateManagementTests, MCPTests):
"""End to end tests for gRPC backend."""
@classmethod
def setUpClass(cls):
cls.model = "/home/ubuntu/models/meta-llama/Llama-3.1-8B-Instruct"
cls.base_url_port = "http://127.0.0.1:30030"
cls.cluster = popen_launch_workers_and_router(
cls.model,
cls.base_url_port,
timeout=90,
num_workers=1,
tp_size=2,
policy="round_robin",
router_args=["--history-backend", "memory"],
)
cls.base_url = cls.cluster["base_url"]
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.cluster["router"].pid)
for worker in cls.cluster.get("workers", []):
kill_process_tree(worker.pid)
@unittest.skip(
"TODO: transport error, details: [], metadata: MetadataMap { headers: {} }"
)
def test_previous_response_id_chaining(self):
super().test_previous_response_id_chaining()
@unittest.skip("TODO: return 501 Not Implemented")
def test_conversation_with_multiple_turns(self):
super().test_conversation_with_multiple_turns()
@unittest.skip("TODO: decode error message")
def test_mutually_exclusive_parameters(self):
super().test_mutually_exclusive_parameters()
@unittest.skip(
"TODO: Pipeline execution failed: Pipeline stage WorkerSelection failed"
)
def test_mcp_basic_tool_call(self):
super().test_mcp_basic_tool_call()
@unittest.skip("TODO: no event fields")
def test_mcp_basic_tool_call_streaming(self):
return super().test_mcp_basic_tool_call_streaming()
class TestHarmonyBackend(StateManagementTests, MCPTests, FunctionCallingBaseTest):
"""End to end tests for Harmony backend."""
@classmethod
def setUpClass(cls):
cls.model = "/home/ubuntu/models/openai/gpt-oss-20b"
cls.base_url_port = "http://127.0.0.1:30030"
cls.cluster = popen_launch_workers_and_router(
cls.model,
cls.base_url_port,
timeout=90,
num_workers=1,
tp_size=2,
policy="round_robin",
router_args=["--history-backend", "memory"],
)
cls.base_url = cls.cluster["base_url"]
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.cluster["router"].pid)
for worker in cls.cluster.get("workers", []):
kill_process_tree(worker.pid)
def test_previous_response_id_chaining(self):
super().test_previous_response_id_chaining()
@unittest.skip(
"TODO: fix requests.exceptions.JSONDecodeError: Expecting value: line 1 column 1 (char 0)"
)
def test_mutually_exclusive_parameters(self):
super().test_mutually_exclusive_parameters()
def test_mcp_basic_tool_call(self):
"""Test basic MCP tool call (non-streaming)."""
tools = [
{
"type": "mcp",
"server_label": "deepwiki",
"server_url": "https://mcp.deepwiki.com/mcp",
"require_approval": "never",
}
]
resp = self.create_response(
"What transport protocols does the 2025-03-26 version of the MCP spec (modelcontextprotocol/modelcontextprotocol) support?",
tools=tools,
stream=False,
)
# Should successfully make the request
self.assertEqual(resp.status_code, 200)
data = resp.json()
# Basic response structure
self.assertIn("id", data)
self.assertIn("status", data)
self.assertEqual(data["status"], "completed")
self.assertIn("output", data)
self.assertIn("model", data)
# Verify output array is not empty
output = data["output"]
self.assertIsInstance(output, list)
self.assertGreater(len(output), 0)
# Check for MCP-specific output types
output_types = [item.get("type") for item in output]
# Should have mcp_list_tools - tools are listed before calling
self.assertIn(
"mcp_list_tools", output_types, "Response should contain mcp_list_tools"
)
# Should have at least one mcp_call
mcp_calls = [item for item in output if item.get("type") == "mcp_call"]
self.assertGreater(
len(mcp_calls), 0, "Response should contain at least one mcp_call"
)
# Verify mcp_call structure
for mcp_call in mcp_calls:
self.assertIn("id", mcp_call)
self.assertIn("status", mcp_call)
self.assertEqual(mcp_call["status"], "completed")
self.assertIn("server_label", mcp_call)
self.assertEqual(mcp_call["server_label"], "deepwiki")
self.assertIn("name", mcp_call)
self.assertIn("arguments", mcp_call)
self.assertIn("output", mcp_call)
def test_mcp_basic_tool_call_streaming(self):
"""Test basic MCP tool call (streaming)."""
tools = [
{
"type": "mcp",
"server_label": "deepwiki",
"server_url": "https://mcp.deepwiki.com/mcp",
"require_approval": "never",
}
]
resp = self.create_response(
"What transport protocols does the 2025-03-26 version of the MCP spec (modelcontextprotocol/modelcontextprotocol) support?",
tools=tools,
stream=True,
)
# Should successfully make the request
self.assertEqual(resp.status_code, 200)
events = self.parse_sse_events(resp)
self.assertGreater(len(events), 0)
event_types = [e.get("event") for e in events]
# Check for lifecycle events
self.assertIn(
"response.created", event_types, "Should have response.created event"
)
self.assertIn(
"response.completed", event_types, "Should have response.completed event"
)
# Check for MCP list tools events
self.assertIn(
"response.output_item.added",
event_types,
"Should have output_item.added events",
)
self.assertIn(
"response.mcp_list_tools.in_progress",
event_types,
"Should have mcp_list_tools.in_progress event",
)
self.assertIn(
"response.mcp_list_tools.completed",
event_types,
"Should have mcp_list_tools.completed event",
)
# Check for MCP call events
self.assertIn(
"response.mcp_call.in_progress",
event_types,
"Should have mcp_call.in_progress event",
)
self.assertIn(
"response.mcp_call_arguments.delta",
event_types,
"Should have mcp_call_arguments.delta event",
)
self.assertIn(
"response.mcp_call_arguments.done",
event_types,
"Should have mcp_call_arguments.done event",
)
self.assertIn(
"response.mcp_call.completed",
event_types,
"Should have mcp_call.completed event",
)
# Verify final completed event has full response
completed_events = [e for e in events if e.get("event") == "response.completed"]
self.assertEqual(len(completed_events), 1)
final_response = completed_events[0].get("data", {}).get("response", {})
self.assertIn("id", final_response)
self.assertEqual(final_response.get("status"), "completed")
self.assertIn("output", final_response)
# Verify final output contains expected items
final_output = final_response.get("output", [])
final_output_types = [item.get("type") for item in final_output]
self.assertIn("mcp_list_tools", final_output_types)
self.assertIn("mcp_call", final_output_types)
# Verify mcp_call items in final output
mcp_calls = [item for item in final_output if item.get("type") == "mcp_call"]
self.assertGreater(len(mcp_calls), 0)
for mcp_call in mcp_calls:
self.assertEqual(mcp_call.get("status"), "completed")
self.assertEqual(mcp_call.get("server_label"), "deepwiki")
self.assertIn("name", mcp_call)
self.assertIn("arguments", mcp_call)
self.assertIn("output", mcp_call)
@unittest.skip("TODO: 501 Not Implemented")
def test_conversation_with_multiple_turns(self):
super().test_conversation_with_multiple_turns()
if __name__ == "__main__":
unittest.main()
sgl-router/py_test/e2e_response_api/test_response_api.py → sgl-router/py_test/e2e_response_api/backends/test_http_backend.py
View file @ 7ef1964a
"""
OpenAI backend tests for Response API.
HTTP backend tests for Response API (OpenAI and XAI).
Run with:
export OPENAI_API_KEY=your_key
python3 -m pytest py_test/e2e_response_api/test_openai_backend.py -v
python3 -m unittest e2e_response_api.test_openai_backend.TestOpenAIStateManagement
export XAI_API_KEY=your_key
python3 -m pytest py_test/e2e_response_api/backends/test_http_backend.py -v
python3 -m unittest e2e_response_api.backends.test_http_backend.TestOpenaiBackend
"""
import os
......@@ -12,23 +13,25 @@ import sys
import unittest
from pathlib import Path
# Add current directory for imports
_TEST_DIR = Path(__file__).parent
# Add e2e_response_api directory for imports
_TEST_DIR = Path(__file__).parent.parent
sys.path.insert(0, str(_TEST_DIR))
# Import local modules
from base import ConversationCRUDBaseTest, ResponseCRUDBaseTest
from mcp import MCPTests
from router_fixtures import (
popen_launch_openai_xai_router,
popen_launch_workers_and_router,
)
from state_management import StateManagementTests
from mixins.basic_crud import ConversationCRUDBaseTest, ResponseCRUDBaseTest
from mixins.function_call import FunctionCallingBaseTest
from mixins.mcp import MCPTests
from mixins.state_management import StateManagementTests
from router_fixtures import popen_launch_openai_xai_router
from util import kill_process_tree
class TestOpenaiBackend(
ResponseCRUDBaseTest, ConversationCRUDBaseTest, StateManagementTests, MCPTests
ResponseCRUDBaseTest,
ConversationCRUDBaseTest,
StateManagementTests,
MCPTests,
FunctionCallingBaseTest,
):
"""End to end tests for OpenAI backend."""
......@@ -75,79 +78,5 @@ class TestXaiBackend(StateManagementTests):
kill_process_tree(cls.cluster["router"].pid)
class TestOracleStore(ResponseCRUDBaseTest, ConversationCRUDBaseTest):
"""End to end tests for Oracle database storage backend."""
api_key = os.environ.get("OPENAI_API_KEY")
@classmethod
def setUpClass(cls):
cls.model = "gpt-5-nano"
cls.base_url_port = "http://127.0.0.1:30040"
cls.cluster = popen_launch_openai_xai_router(
backend="openai",
base_url=cls.base_url_port,
history_backend="oracle",
)
cls.base_url = cls.cluster["base_url"]
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.cluster["router"].pid)
class TestGrpcBackend(StateManagementTests, MCPTests):
"""End to end tests for gRPC backend."""
@classmethod
def setUpClass(cls):
cls.model = "/home/ubuntu/models/meta-llama/Llama-3.1-8B-Instruct"
cls.base_url_port = "http://127.0.0.1:30030"
cls.cluster = popen_launch_workers_and_router(
cls.model,
cls.base_url_port,
timeout=90,
num_workers=1,
tp_size=2,
policy="round_robin",
router_args=["--history-backend", "memory"],
)
cls.base_url = cls.cluster["base_url"]
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.cluster["router"].pid)
for worker in cls.cluster.get("workers", []):
kill_process_tree(worker.pid)
@unittest.skip(
"TODO: transport error, details: [], metadata: MetadataMap { headers: {} }"
)
def test_previous_response_id_chaining(self):
super().test_previous_response_id_chaining()
@unittest.skip("TODO: return 501 Not Implemented")
def test_conversation_with_multiple_turns(self):
super().test_conversation_with_multiple_turns()
@unittest.skip("TODO: decode error message")
def test_mutually_exclusive_parameters(self):
super().test_mutually_exclusive_parameters()
@unittest.skip(
"TODO: Pipeline execution failed: Pipeline stage WorkerSelection failed"
)
def test_mcp_basic_tool_call(self):
super().test_mcp_basic_tool_call()
@unittest.skip("TODO: no event fields")
def test_mcp_basic_tool_call_streaming(self):
return super().test_mcp_basic_tool_call_streaming()
if __name__ == "__main__":
unittest.main()
sgl-router/py_test/e2e_response_api/conftest.py
View file @ 7ef1964a
......@@ -18,6 +18,7 @@ def pytest_collection_modifyitems(config, items):
- ConversationCRUDBaseTest
- MCPTests
- StateManagementTests
- FunctionCallingBaseTest
"""
base_class_names = {
"StateManagementBaseTest",
......@@ -25,6 +26,7 @@ def pytest_collection_modifyitems(config, items):
"ConversationCRUDBaseTest",
"MCPTests",
"StateManagementTests",
"FunctionCallingBaseTest",
}
# Filter out tests from base classes
......
sgl-router/py_test/e2e_response_api/mixins/function_call.py 0 → 100644
View file @ 7ef1964a
"""
Base test class for function calling tests.
This module provides test cases for function calling functionality
across different backends.
"""
import json
import sys
from pathlib import Path
# Add current directory for local imports
_TEST_DIR = Path(__file__).parent
sys.path.insert(0, str(_TEST_DIR))
from util import CustomTestCase
class ResponseAPIBaseTest(CustomTestCase):
"""Base class for Response API tests with common utilities."""
# To be set by subclasses
base_url: str = None
api_key: str = None
model: str = None
def make_request(
self,
endpoint: str,
method: str = "POST",
json_data: dict = None,
params: dict = None,
):
"""
Make HTTP request to router.
This is a minimal implementation - subclasses should import from basic_crud.
"""
import requests
url = f"{self.base_url}{endpoint}"
headers = {"Content-Type": "application/json"}
if self.api_key:
headers["Authorization"] = f"Bearer {self.api_key}"
if method == "POST":
resp = requests.post(url, json=json_data, headers=headers, params=params)
elif method == "GET":
resp = requests.get(url, headers=headers, params=params)
elif method == "DELETE":
resp = requests.delete(url, headers=headers, params=params)
else:
raise ValueError(f"Unsupported method: {method}")
return resp
class FunctionCallingBaseTest(ResponseAPIBaseTest):
"""Base class for function calling tests."""
def test_basic_function_call(self):
"""
Test basic function calling workflow.
This test follows the pattern from function_call_test.py:
1. Define a function tool (get_horoscope)
2. Send user message asking for horoscope
3. Model should return function_call
4. Execute function locally and provide output
5. Model should generate final response using the function output
"""
# 1. Define a list of callable tools for the model
tools = [
{
"type": "function",
"name": "get_horoscope",
"description": "Get today's horoscope for an astrological sign.",
"parameters": {
"type": "object",
"properties": {
"sign": {
"type": "string",
"description": "An astrological sign like Taurus or Aquarius",
},
},
"required": ["sign"],
},
},
]
# Create a running input list we will add to over time
input_list = [
{"role": "user", "content": "What is my horoscope? I am an Aquarius."}
]
# 2. Prompt the model with tools defined
resp = self.make_request(
"/v1/responses",
"POST",
{
"model": self.model,
"tools": tools,
"input": input_list,
},
)
# Should successfully make the request
self.assertEqual(resp.status_code, 200)
data = resp.json()
# Basic response structure
self.assertIn("id", data)
self.assertIn("status", data)
self.assertEqual(data["status"], "completed")
self.assertIn("output", data)
# Verify output array is not empty
output = data["output"]
self.assertIsInstance(output, list)
self.assertGreater(len(output), 0)
# Check for function_call in output
function_calls = [
item for item in output if item.get("type") == "function_call"
]
self.assertGreater(
len(function_calls), 0, "Response should contain at least one function_call"
)
# Verify function_call structure
function_call = function_calls[0]
self.assertIn("call_id", function_call)
self.assertIn("name", function_call)
self.assertEqual(function_call["name"], "get_horoscope")
self.assertIn("arguments", function_call)
# Parse arguments
args = json.loads(function_call["arguments"])
self.assertIn("sign", args)
self.assertEqual(args["sign"].lower(), "aquarius")
# 3. Save function call outputs for subsequent requests
input_list += output
# 4. Execute the function logic for get_horoscope
horoscope = f"{args['sign']}: Next Tuesday you will befriend a baby otter."
# 5. Provide function call results to the model
input_list.append(
{
"type": "function_call_output",
"call_id": function_call["call_id"],
"output": json.dumps({"horoscope": horoscope}),
}
)
# 6. Make second request with function output
resp2 = self.make_request(
"/v1/responses",
"POST",
{
"model": self.model,
"instructions": "Respond only with a horoscope generated by a tool.",
"tools": tools,
"input": input_list,
},
)
self.assertEqual(resp2.status_code, 200)
data2 = resp2.json()
self.assertEqual(data2["status"], "completed")
# The model should be able to give a response using the function output
output2 = data2["output"]
self.assertGreater(len(output2), 0)
# Find message output
messages = [item for item in output2 if item.get("type") == "message"]
self.assertGreater(
len(messages), 0, "Response should contain at least one message"
)
# Verify message contains the horoscope
message = messages[0]
self.assertIn("content", message)
content_parts = message["content"]
self.assertGreater(len(content_parts), 0)
# Get text from content
text_parts = [
part.get("text", "")
for part in content_parts
if part.get("type") == "output_text"
]
full_text = " ".join(text_parts).lower()
# Should mention the horoscope or baby otter
self.assertTrue(
"baby otter" in full_text or "aquarius" in full_text,
"Response should reference the horoscope content",
)
sgl-router/py_test/e2e_response_api/mcp.py → sgl-router/py_test/e2e_response_api/mixins/mcp.py
View file @ 7ef1964a
......@@ -5,7 +5,7 @@ Tests MCP tool calling in both streaming and non-streaming modes.
These tests should work across all backends that support MCP (OpenAI, XAI).
"""
from base import ResponseAPIBaseTest
from basic_crud import ResponseAPIBaseTest
class MCPTests(ResponseAPIBaseTest):
......
sgl-router/py_test/e2e_response_api/state_management.py → sgl-router/py_test/e2e_response_api/mixins/state_management.py
View file @ 7ef1964a
......@@ -7,7 +7,7 @@ These tests should work across all backends (OpenAI, XAI, gRPC).
import unittest
from base import ResponseAPIBaseTest
from basic_crud import ResponseAPIBaseTest
class StateManagementTests(ResponseAPIBaseTest):
......
sgl-router/py_test/e2e_response_api/persistence/test_oracle_store.py 0 → 100644
View file @ 7ef1964a
"""
Oracle database storage backend tests for Response API.
Run with:
export OPENAI_API_KEY=your_key
python3 -m pytest py_test/e2e_response_api/persistence/test_oracle_store.py -v
python3 -m unittest e2e_response_api.persistence.test_oracle_store.TestOracleStore
"""
import os
import sys
import unittest
from pathlib import Path
# Add e2e_response_api directory for imports
_TEST_DIR = Path(__file__).parent.parent
sys.path.insert(0, str(_TEST_DIR))
# Import local modules
from mixins.basic_crud import ConversationCRUDBaseTest, ResponseCRUDBaseTest
from router_fixtures import popen_launch_openai_xai_router
from util import kill_process_tree
class TestOracleStore(ResponseCRUDBaseTest, ConversationCRUDBaseTest):
"""End to end tests for Oracle database storage backend."""
api_key = os.environ.get("OPENAI_API_KEY")
@classmethod
def setUpClass(cls):
cls.model = "gpt-5-nano"
cls.base_url_port = "http://127.0.0.1:30040"
cls.cluster = popen_launch_openai_xai_router(
backend="openai",
base_url=cls.base_url_port,
history_backend="oracle",
)
cls.base_url = cls.cluster["base_url"]
@classmethod
def tearDownClass(cls):
kill_process_tree(cls.cluster["router"].pid)
if __name__ == "__main__":
unittest.main()
sgl-router/py_test/e2e_response_api/router_fixtures.py
View file @ 7ef1964a
......@@ -425,8 +425,6 @@ def popen_launch_workers_and_router(
"--grpc-mode", # Enable gRPC for this worker
"--mem-fraction-static",
"0.8",
"--attention-backend",
"fa3",
]
# Add TP size
......
sgl-router/src/protocols/responses.rs
View file @ 7ef1964a
......@@ -128,6 +128,7 @@ pub enum ResponseInputOutputItem {
id: String,
summary: Vec<String>,
#[serde(skip_serializing_if = "Vec::is_empty")]
#[serde(default)]
content: Vec<ResponseReasoningContent>,
#[serde(skip_serializing_if = "Option::is_none")]
status: Option<String>,
......@@ -168,6 +169,7 @@ pub enum ResponseContentPart {
#[serde(rename = "output_text")]
OutputText {
text: String,
#[serde(default)]
#[serde(skip_serializing_if = "Vec::is_empty")]
annotations: Vec<String>,
#[serde(skip_serializing_if = "Option::is_none")]
......
sgl-router/src/routers/grpc/common/responses/mod.rs
View file @ 7ef1964a
......@@ -6,4 +6,4 @@ pub mod utils;
pub use handlers::{cancel_response_impl, get_response_impl};
pub use streaming::{build_sse_response, OutputItemType, ResponseStreamEventEmitter};
pub use utils::ensure_mcp_connection;
pub use utils::{ensure_mcp_connection, persist_response_if_needed};
sgl-router/src/routers/grpc/common/responses/streaming.rs
View file @ 7ef1964a
......@@ -11,7 +11,14 @@ use uuid::Uuid;
use crate::{
mcp,
protocols::{chat::ChatCompletionStreamResponse, responses::ResponsesRequest},
protocols::{
chat::ChatCompletionStreamResponse,
common::{Usage, UsageInfo},
responses::{
ResponseOutputItem, ResponseStatus, ResponsesRequest, ResponsesResponse, ResponsesUsage,
},
},
routers::grpc::harmony::responses::ToolResult,
};
pub enum OutputItemType {
......@@ -69,6 +76,7 @@ pub struct ResponseStreamEventEmitter {
has_emitted_content_part_added: bool,
// MCP call tracking
mcp_call_accumulated_args: HashMap<String, String>,
pub(crate) mcp_server_label: Option<String>, // Server label for MCP tools
// Output item tracking
output_items: Vec<OutputItemState>,
next_output_index: usize,
......@@ -93,6 +101,7 @@ impl ResponseStreamEventEmitter {
has_emitted_output_item_added: false,
has_emitted_content_part_added: false,
mcp_call_accumulated_args: HashMap::new(),
mcp_server_label: None,
output_items: Vec::new(),
next_output_index: 0,
current_message_output_index: None,
......@@ -106,6 +115,41 @@ impl ResponseStreamEventEmitter {
self.original_request = Some(request);
}
/// Set the MCP server label for MCP tool calls
pub fn set_mcp_server_label(&mut self, server_label: String) {
self.mcp_server_label = Some(server_label);
}
/// Update mcp_call output items with tool execution results
///
/// After MCP tools are executed, this updates the stored output items
/// to include the output field from the tool results.
pub(crate) fn update_mcp_call_outputs(&mut self, tool_results: &[ToolResult]) {
for tool_result in tool_results {
// Find the output item with matching call_id
for item_state in self.output_items.iter_mut() {
if let Some(ref mut item_data) = item_state.item_data {
// Check if this is an mcp_call item with matching call_id
if item_data.get("type").and_then(|t| t.as_str()) == Some("mcp_call")
&& item_data.get("call_id").and_then(|c| c.as_str())
== Some(&tool_result.call_id)
{
// Add output field
let output_str = serde_json::to_string(&tool_result.output)
.unwrap_or_else(|_| "{}".to_string());
item_data["output"] = json!(output_str);
// Update status based on success
if tool_result.is_error {
item_data["status"] = json!("failed");
}
break;
}
}
}
}
}
fn next_sequence(&mut self) -> u64 {
let seq = self.sequence_number;
self.sequence_number += 1;
......@@ -542,6 +586,78 @@ impl ResponseStreamEventEmitter {
}
}
/// Finalize and return the complete ResponsesResponse
///
/// This constructs the final ResponsesResponse from all accumulated output items
/// for persistence. Should be called after streaming is complete.
pub fn finalize(&self, usage: Option<Usage>) -> ResponsesResponse {
// Build output array from tracked items
let output: Vec<ResponseOutputItem> = self
.output_items
.iter()
.filter_map(|item| {
item.item_data
.as_ref()
.and_then(|data| serde_json::from_value(data.clone()).ok())
})
.collect();
// Convert Usage to ResponsesUsage
let responses_usage = usage.map(|u| {
let usage_info = UsageInfo {
prompt_tokens: u.prompt_tokens,
completion_tokens: u.completion_tokens,
total_tokens: u.total_tokens,
reasoning_tokens: u
.completion_tokens_details
.as_ref()
.and_then(|d| d.reasoning_tokens),
prompt_tokens_details: None,
};
ResponsesUsage::Classic(usage_info)
});
// Get original request fields or use defaults
let req = self.original_request.as_ref();
// Convert tool_choice to String
let tool_choice = req
.and_then(|r| r.tool_choice.as_ref())
.map(|tc| serde_json::to_string(tc).unwrap_or_else(|_| "auto".to_string()))
.unwrap_or_else(|| "auto".to_string());
ResponsesResponse {
id: self.response_id.clone(),
object: "response".to_string(),
created_at: self.created_at as i64,
status: ResponseStatus::Completed,
error: None,
incomplete_details: None,
instructions: req.and_then(|r| r.instructions.clone()),
max_output_tokens: req.and_then(|r| r.max_output_tokens),
model: self.model.clone(),
output,
parallel_tool_calls: req.and_then(|r| r.parallel_tool_calls).unwrap_or(true),
previous_response_id: req.and_then(|r| r.previous_response_id.clone()),
reasoning: None, // TODO: Extract from output items if needed
store: req.and_then(|r| r.store).unwrap_or(true),
temperature: req.and_then(|r| r.temperature),
text: None,
tool_choice,
tools: req
.map(|r| r.tools.clone().unwrap_or_default())
.unwrap_or_default(),
top_p: req.and_then(|r| r.top_p),
truncation: None, // Convert from Truncation to String if needed
usage: responses_usage,
metadata: req
.map(|r| r.metadata.clone().unwrap_or_default())
.unwrap_or_default(),
user: req.and_then(|r| r.user.clone()),
safety_identifier: None,
}
}
/// Emit reasoning item wrapper events (added + done)
///
/// Reasoning items in OpenAI format are simple placeholders emitted between tool iterations.
......
sgl-router/src/routers/grpc/common/responses/utils.rs
View file @ 7ef1964a
......@@ -6,16 +6,21 @@ use axum::{
http::StatusCode,
response::{IntoResponse, Response},
};
use serde_json::json;
use serde_json::{json, to_value};
use tracing::{debug, warn};
use crate::{
core::WorkerRegistry,
data_connector::{ConversationItemStorage, ConversationStorage, ResponseStorage},
mcp::McpManager,
protocols::{
common::Tool,
responses::{ResponseTool, ResponseToolType},
responses::{ResponseTool, ResponseToolType, ResponsesRequest, ResponsesResponse},
},
routers::{
grpc::error,
openai::{conversations::persist_conversation_items, mcp::ensure_request_mcp_client},
},
routers::{grpc::error, openai::mcp::ensure_request_mcp_client},
};
/// Ensure MCP connection succeeds if MCP tools are declared
......@@ -123,3 +128,35 @@ pub fn extract_tools_from_response_tools(
})
.collect()
}
/// Persist response to storage if store=true
///
/// Common helper function to avoid duplication across sync and streaming paths
/// in both harmony and regular responses implementations.
pub async fn persist_response_if_needed(
conversation_storage: Arc<dyn ConversationStorage>,
conversation_item_storage: Arc<dyn ConversationItemStorage>,
response_storage: Arc<dyn ResponseStorage>,
response: &ResponsesResponse,
original_request: &ResponsesRequest,
) {
if !original_request.store.unwrap_or(true) {
return;
}
if let Ok(response_json) = to_value(response) {
if let Err(e) = persist_conversation_items(
conversation_storage,
conversation_item_storage,
response_storage,
&response_json,
original_request,
)
.await
{
warn!("Failed to persist response: {}", e);
} else {
debug!("Persisted response: {}", response.id);
}
}
}
sgl-router/src/routers/grpc/harmony/responses.rs
View file @ 7ef1964a
......@@ -49,7 +49,7 @@ use tracing::{debug, warn};
use uuid::Uuid;
use crate::{
data_connector::{ResponseId, ResponseStorage},
data_connector::{ConversationItemStorage, ConversationStorage, ResponseId, ResponseStorage},
mcp::{self, McpManager},
protocols::{
common::{Function, ToolCall, ToolChoice, ToolChoiceValue, Usage},
......@@ -62,7 +62,7 @@ use crate::{
},
routers::grpc::{
common::responses::{
build_sse_response, ensure_mcp_connection,
build_sse_response, ensure_mcp_connection, persist_response_if_needed,
streaming::{OutputItemType, ResponseStreamEventEmitter},
},
context::SharedComponents,
......@@ -157,6 +157,12 @@ pub struct HarmonyResponsesContext {
/// Response storage for loading conversation history
pub response_storage: Arc<dyn ResponseStorage>,
/// Conversation storage for persisting conversations
pub conversation_storage: Arc<dyn ConversationStorage>,
/// Conversation item storage for persisting conversation items
pub conversation_item_storage: Arc<dyn ConversationItemStorage>,
/// Optional streaming sender (for future streaming support)
pub stream_tx: Option<mpsc::UnboundedSender<Result<String, String>>>,
}
......@@ -168,12 +174,16 @@ impl HarmonyResponsesContext {
components: Arc<SharedComponents>,
mcp_manager: Arc<McpManager>,
response_storage: Arc<dyn ResponseStorage>,
conversation_storage: Arc<dyn ConversationStorage>,
conversation_item_storage: Arc<dyn ConversationItemStorage>,
) -> Self {
Self {
pipeline,
components,
mcp_manager,
response_storage,
conversation_storage,
conversation_item_storage,
stream_tx: None,
}
}
......@@ -184,6 +194,8 @@ impl HarmonyResponsesContext {
components: Arc<SharedComponents>,
mcp_manager: Arc<McpManager>,
response_storage: Arc<dyn ResponseStorage>,
conversation_storage: Arc<dyn ConversationStorage>,
conversation_item_storage: Arc<dyn ConversationItemStorage>,
stream_tx: mpsc::UnboundedSender<Result<String, String>>,
) -> Self {
Self {
......@@ -191,6 +203,8 @@ impl HarmonyResponsesContext {
components,
mcp_manager,
response_storage,
conversation_storage,
conversation_item_storage,
stream_tx: Some(stream_tx),
}
}
......@@ -237,6 +251,9 @@ pub async fn serve_harmony_responses(
ctx: &HarmonyResponsesContext,
request: ResponsesRequest,
) -> Result<ResponsesResponse, Response> {
// Clone request for persistence
let original_request = request.clone();
// Load previous conversation history if previous_response_id is set
let current_request = load_previous_messages(ctx, request).await?;
......@@ -244,12 +261,24 @@ pub async fn serve_harmony_responses(
let has_mcp_tools =
ensure_mcp_connection(&ctx.mcp_manager, current_request.tools.as_deref()).await?;
if has_mcp_tools {
execute_with_mcp_loop(ctx, current_request).await
let response = if has_mcp_tools {
execute_with_mcp_loop(ctx, current_request).await?
} else {
// No MCP tools - execute pipeline once (may have function tools or no tools)
execute_without_mcp_loop(ctx, current_request).await
}
execute_without_mcp_loop(ctx, current_request).await?
};
// Persist response to storage if store=true
persist_response_if_needed(
ctx.conversation_storage.clone(),
ctx.conversation_item_storage.clone(),
ctx.response_storage.clone(),
&response,
&original_request,
)
.await;
Ok(response)
}
/// Execute Harmony Responses with MCP tool loop
......@@ -260,7 +289,10 @@ async fn execute_with_mcp_loop(
mut current_request: ResponsesRequest,
) -> Result<ResponsesResponse, Response> {
let mut iteration_count = 0;
let mut mcp_tracking = McpCallTracking::new("sglang-mcp".to_string());
// Extract server_label from request tools
let server_label = extract_mcp_server_label(current_request.tools.as_deref());
let mut mcp_tracking = McpCallTracking::new(server_label.clone());
// Extract user's max_tool_calls limit (if set)
let max_tool_calls = current_request.max_tool_calls.map(|n| n as usize);
......@@ -472,7 +504,7 @@ pub async fn serve_harmony_responses_stream(
request: ResponsesRequest,
) -> Response {
// Load previous conversation history if previous_response_id is set
let current_request = match load_previous_messages(ctx, request).await {
let current_request = match load_previous_messages(ctx, request.clone()).await {
Ok(req) => req,
Err(err_response) => return err_response,
};
......@@ -517,9 +549,10 @@ pub async fn serve_harmony_responses_stream(
}
if has_mcp_tools {
execute_mcp_tool_loop_streaming(ctx, current_request, &mut emitter, &tx).await;
execute_mcp_tool_loop_streaming(ctx, current_request, &request, &mut emitter, &tx)
.await;
} else {
execute_without_mcp_streaming(ctx, &current_request, &mut emitter, &tx).await;
execute_without_mcp_streaming(ctx, &current_request, &request, &mut emitter, &tx).await;
}
});
......@@ -534,14 +567,22 @@ pub async fn serve_harmony_responses_stream(
/// - Emits mcp_list_tools events
/// - Loops through tool execution iterations
/// - Emits final response.completed event
/// - Persists response internally
async fn execute_mcp_tool_loop_streaming(
ctx: &HarmonyResponsesContext,
mut current_request: ResponsesRequest,
original_request: &ResponsesRequest,
emitter: &mut ResponseStreamEventEmitter,
tx: &mpsc::UnboundedSender<Result<Bytes, std::io::Error>>,
) {
// Extract server_label from request tools
let server_label = extract_mcp_server_label(current_request.tools.as_deref());
// Set server label in emitter for MCP call items
emitter.set_mcp_server_label(server_label.clone());
// Initialize MCP call tracking
let mut mcp_tracking = McpCallTracking::new("sglang-mcp".to_string());
let mut mcp_tracking = McpCallTracking::new(server_label.clone());
// Extract user's max_tool_calls limit (if set)
let max_tool_calls = current_request.max_tool_calls.map(|n| n as usize);
......@@ -576,11 +617,26 @@ async fn execute_mcp_tool_loop_streaming(
})
.collect();
// Build final item with completed status and tools
let item_done = json!({
"id": item_id,
"type": "mcp_list_tools",
"server_label": server_label,
"status": "completed",
"tools": tool_items
});
// Store the completed item data and mark as completed FIRST
// This ensures it appears in final response even if event sending fails
emitter.emit_output_item_done(output_index, &item_done);
emitter.complete_output_item(output_index);
// Now emit all the events (failures won't affect the stored data)
// Emit output_item.added
let item = json!({
"id": item_id,
"type": "mcp_list_tools",
"server_label": "sglang-mcp",
"server_label": server_label,
"status": "in_progress",
"tools": []
});
......@@ -602,20 +658,11 @@ async fn execute_mcp_tool_loop_streaming(
}
// Emit output_item.done
let item_done = json!({
"id": item_id,
"type": "mcp_list_tools",
"server_label": "sglang-mcp",
"status": "completed",
"tools": tool_items
});
let event = emitter.emit_output_item_done(output_index, &item_done);
if emitter.send_event(&event, tx).is_err() {
return;
}
emitter.complete_output_item(output_index);
debug!(
tool_count = mcp_tools.len(),
"Emitted mcp_list_tools on first iteration"
......@@ -736,6 +783,9 @@ async fn execute_mcp_tool_loop_streaming(
}
};
// Update mcp_call output items with execution results
emitter.update_mcp_call_outputs(&tool_results);
// Build next request with appended history
current_request = match build_next_request_with_tools(
current_request,
......@@ -765,6 +815,19 @@ async fn execute_mcp_tool_loop_streaming(
"Harmony Responses streaming completed - no more tool calls"
);
// Finalize response from emitter's accumulated data
let final_response = emitter.finalize(Some(usage.clone()));
// Persist response to storage if store=true
persist_response_if_needed(
ctx.conversation_storage.clone(),
ctx.conversation_item_storage.clone(),
ctx.response_storage.clone(),
&final_response,
original_request,
)
.await;
// Emit response.completed with usage
let usage_json = json!({
"input_tokens": usage.prompt_tokens,
......@@ -786,6 +849,7 @@ async fn execute_mcp_tool_loop_streaming(
async fn execute_without_mcp_streaming(
ctx: &HarmonyResponsesContext,
current_request: &ResponsesRequest,
original_request: &ResponsesRequest,
emitter: &mut ResponseStreamEventEmitter,
tx: &mpsc::UnboundedSender<Result<Bytes, std::io::Error>>,
) {
......@@ -830,6 +894,19 @@ async fn execute_without_mcp_streaming(
ResponsesIterationResult::Completed { usage, .. } => usage,
};
// Finalize response from emitter's accumulated data
let final_response = emitter.finalize(Some(usage.clone()));
// Persist response to storage if store=true
persist_response_if_needed(
ctx.conversation_storage.clone(),
ctx.conversation_item_storage.clone(),
ctx.response_storage.clone(),
&final_response,
original_request,
)
.await;
// Emit response.completed with usage
let usage_json = json!({
"input_tokens": usage.prompt_tokens,
......@@ -1191,19 +1268,19 @@ fn build_next_request_with_tools(
/// Tool execution result
///
/// Contains the result of executing a single MCP tool.
struct ToolResult {
pub(crate) struct ToolResult {
/// Tool call ID (for matching with request)
call_id: String,
pub(crate) call_id: String,
/// Tool name
#[allow(dead_code)] // Kept for documentation and future use
tool_name: String,
pub(crate) tool_name: String,
/// Tool output (JSON value)
output: Value,
pub(crate) output: Value,
/// Whether this is an error result
is_error: bool,
pub(crate) is_error: bool,
}
/// Convert MCP tools to Responses API tool format
......@@ -1297,6 +1374,24 @@ fn inject_mcp_metadata(
response.output.extend(mcp_call_items);
}
/// Extract MCP server label from request tools
///
/// Searches for the first MCP tool in the tools array and returns its server_label.
/// Falls back to "sglang-mcp" if no MCP tool with server_label is found.
fn extract_mcp_server_label(tools: Option<&[ResponseTool]>) -> String {
tools
.and_then(|tools| {
tools.iter().find_map(|tool| {
if matches!(tool.r#type, ResponseToolType::Mcp) {
tool.server_label.clone()
} else {
None
}
})
})
.unwrap_or_else(|| "sglang-mcp".to_string())
}
/// Load previous conversation messages from storage
///
/// If the request has `previous_response_id`, loads the response chain from storage
......
sgl-router/src/routers/grpc/harmony/streaming.rs
View file @ 7ef1964a
......@@ -733,7 +733,7 @@ impl HarmonyStreamingProcessor {
// Emit output_item.added wrapper event
let call_id = tc_delta.id.as_ref().unwrap();
let item = json!({
let mut item = json!({
"id": item_id,
"type": mode.type_str(),
"name": tool_name,
......@@ -741,6 +741,14 @@ impl HarmonyStreamingProcessor {
"arguments": "",
"status": "in_progress"
});
// Add server_label for MCP calls
if mode.emits_status_events() {
if let Some(ref server_label) = emitter.mcp_server_label {
item["server_label"] = json!(server_label);
}
}
let event = emitter.emit_output_item_added(output_index, &item);
emitter.send_event_best_effort(&event, tx);
......@@ -836,7 +844,7 @@ impl HarmonyStreamingProcessor {
}
// Emit output_item.done wrapper event
let item = json!({
let mut item = json!({
"id": item_id,
"type": mode.type_str(),
"name": tool_name,
......@@ -844,11 +852,21 @@ impl HarmonyStreamingProcessor {
"arguments": args_str,
"status": "completed"
});
// Add server_label for MCP calls
if mode.emits_status_events() {
// MCP mode - include server_label
if let Some(ref server_label) = emitter.mcp_server_label {
item["server_label"] = json!(server_label);
}
}
let event = emitter.emit_output_item_done(*output_index, &item);
emitter.send_event_best_effort(&event, tx);
// Mark output item as completed
// Mark output item as completed before sending
emitter.complete_output_item(*output_index);
emitter.send_event_best_effort(&event, tx);
}
}
}
......@@ -878,9 +896,11 @@ impl HarmonyStreamingProcessor {
}]
});
let event = emitter.emit_output_item_done(output_index, &item);
emitter.send_event_best_effort(&event, tx);
// Mark as completed before sending (so it's included in final output even if send fails)
emitter.complete_output_item(output_index);
emitter.send_event_best_effort(&event, tx);
}
}
Some(proto::generate_response::Response::Error(err)) => {
......@@ -939,7 +959,7 @@ impl HarmonyStreamingProcessor {
}
// Emit output_item.done wrapper event
let item = json!({
let mut item = json!({
"id": item_id,
"type": mode.type_str(),
"name": tool_name,
......@@ -947,11 +967,20 @@ impl HarmonyStreamingProcessor {
"arguments": args_str,
"status": "completed"
});
// Add server_label for MCP calls
if mode.emits_status_events() {
if let Some(ref server_label) = emitter.mcp_server_label {
item["server_label"] = json!(server_label);
}
}
let event = emitter.emit_output_item_done(*output_index, &item);
emitter.send_event_best_effort(&event, tx);
// Mark output item as completed
// Mark output item as completed before sending
emitter.complete_output_item(*output_index);
emitter.send_event_best_effort(&event, tx);
}
}
}
......
sgl-router/src/routers/grpc/regular/responses/handlers.rs
View file @ 7ef1964a
......@@ -65,15 +65,13 @@ use crate::{
ResponsesUsage,
},
},
routers::{
grpc::{
routers::grpc::{
common::responses::{
build_sse_response, ensure_mcp_connection, streaming::ResponseStreamEventEmitter,
build_sse_response, ensure_mcp_connection, persist_response_if_needed,
streaming::ResponseStreamEventEmitter,
},
error,
},
openai::conversations::persist_conversation_items,
},
};
/// Main handler for POST /v1/responses
......@@ -221,21 +219,14 @@ async fn route_responses_internal(
};
// 5. Persist response to storage if store=true
if request.store.unwrap_or(true) {
if let Ok(response_json) = serde_json::to_value(&responses_response) {
if let Err(e) = persist_conversation_items(
persist_response_if_needed(
ctx.conversation_storage.clone(),
ctx.conversation_item_storage.clone(),
ctx.response_storage.clone(),
&response_json,
&responses_response,
&request,
)
.await
{
warn!("Failed to persist response: {}", e);
}
}
}
.await;
Ok(responses_response)
}
......@@ -454,25 +445,15 @@ async fn process_and_transform_sse_stream(
event_emitter.send_event(&completed_event, &tx)?;
// Finalize and persist accumulated response
if original_request.store.unwrap_or(true) {
let final_response = accumulator.finalize();
if let Ok(response_json) = serde_json::to_value(&final_response) {
if let Err(e) = persist_conversation_items(
conversation_storage.clone(),
conversation_item_storage.clone(),
response_storage.clone(),
&response_json,
persist_response_if_needed(
conversation_storage,
conversation_item_storage,
response_storage,
&final_response,
&original_request,
)
.await
{
warn!("Failed to persist streaming response: {}", e);
} else {
debug!("Persisted streaming response: {}", final_response.id);
}
}
}
.await;
Ok(())
}
......
sgl-router/src/routers/grpc/router.rs
View file @ 7ef1964a
......@@ -216,6 +216,10 @@ impl GrpcRouter {
self.shared_components.clone(),
self.harmony_responses_context.mcp_manager.clone(),
self.harmony_responses_context.response_storage.clone(),
self.harmony_responses_context.conversation_storage.clone(),
self.harmony_responses_context
.conversation_item_storage
.clone(),
);
if body.stream.unwrap_or(false) {
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment