Skip to content

[https://nvbugs/5485593][fix] improve accuracy/test_disaggregated_serving.py #7366

New issue

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

Sign up for GitHub

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

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Sep 3, 2025
+32 −20
Merged

[https://nvbugs/5485593][fix] improve accuracy/test_disaggregated_serving.py #7366

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e0ab265
detect subprocess status and exit early if subprocess failed
reasonsolo Aug 29, 2025
b0c0686
Merge branch 'main' into bug5485593
reasonsolo Sep 3, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
52 changes: 32 additions & 20 deletions tests/integration/defs/accuracy/test_disaggregated_serving.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@
# Please take a look at the existing test_llm_api_pytorch.py file for reference.
import concurrent
import contextlib
import itertools
import json
import os
import tempfile
Expand Down Expand Up @@ -47,6 +48,9 @@ def result(self):

DuckLLM = namedtuple('DuckLLM', ['args', 'tokenizer', 'generate_async'])

DEFAULT_TEST_TIMEOUT = 1800
DEFAULT_SERVER_WAITING_TIMEOUT = 1200


class MyThreadPoolExecutor(ThreadPoolExecutor):

Expand All @@ -67,13 +71,15 @@ def __exit__(self, exc_type, exc_val, exc_tb):


@contextlib.contextmanager
def launch_disaggregated_llm(disaggregated_server_config: Dict[str, Any],
ctx_server_config: Dict[str, Any],
gen_server_config: Dict[str, Any],
model_name: str,
tensor_parallel_size: int = 1,
ctx_model: str = None,
gen_model: str = None):
def launch_disaggregated_llm(
disaggregated_server_config: Dict[str, Any],
ctx_server_config: Dict[str, Any],
gen_server_config: Dict[str, Any],
model_name: str,
tensor_parallel_size: int = 1,
ctx_model: str = None,
gen_model: str = None,
server_waiting_timeout: int = DEFAULT_SERVER_WAITING_TIMEOUT):
temp_dir = tempfile.TemporaryDirectory()
disaggregated_serving_config_path = os.path.join(
temp_dir.name, "disaggregated_serving_config.yaml")
Expand Down Expand Up @@ -197,16 +203,22 @@ def multi_popen(server_configs):
)
raise

server_cmd = [
trtllm_serve_path, "disaggregated", "-c",
disaggregated_serving_config_path, "--server_start_timeout",
str(server_waiting_timeout)
]
with (MyThreadPoolExecutor(max_workers=16) as
thread_pool, temp_dir, multi_popen(ctx_servers + gen_servers),
popen([
trtllm_serve_path, "disaggregated", "-c",
disaggregated_serving_config_path, "--server_start_timeout",
"3600"
])):
thread_pool, temp_dir, multi_popen(ctx_servers + gen_servers) as
worker_processes, popen(server_cmd) as server_process):
start_time = time.time()
while time.time() - start_time < 3600:
time.sleep(1)
while time.time() - start_time < server_waiting_timeout:
time.sleep(5)
for process in itertools.chain(worker_processes, [server_process]):
if process.poll() is not None:
raise Exception(
f"process {process.pid} exited with code {process.returncode}"
)
try:
print("Checking health endpoint")
response = requests.get("http://localhost:8000/health")
Expand Down Expand Up @@ -339,7 +351,7 @@ def run_parallel_test(model_name: str,
task.evaluate(llm)


@pytest.mark.timeout(3600)
@pytest.mark.timeout(DEFAULT_TEST_TIMEOUT)
class TestLlama3_1_8BInstruct(LlmapiAccuracyTestHarness):
MODEL_NAME = "meta-llama/Llama-3.1-8B-Instruct"
MODEL_PATH = f"{llm_models_root()}/llama-3.1-model/Llama-3.1-8B-Instruct"
Expand Down Expand Up @@ -639,7 +651,7 @@ def test_auto_dtype(self, overlap_scheduler):
task.evaluate(llm)


@pytest.mark.timeout(3600)
@pytest.mark.timeout(DEFAULT_TEST_TIMEOUT)
class TestDeepSeekV3Lite(LlmapiAccuracyTestHarness):
MODEL_NAME = "deepseek-ai/DeepSeek-V3-Lite"
MODEL_PATH = f"{llm_models_root()}/DeepSeek-V3-Lite/bf16"
Expand Down Expand Up @@ -724,7 +736,7 @@ def test_auto_dtype(self, overlap_scheduler, mtp_nextn):
task.evaluate(llm)


@pytest.mark.timeout(3600)
@pytest.mark.timeout(DEFAULT_TEST_TIMEOUT)
class TestGemma3_1BInstruct(LlmapiAccuracyTestHarness):
MODEL_NAME = "google/gemma-3-1b-it"
MODEL_PATH = f"{llm_models_root()}/gemma/gemma-3-1b-it/"
Expand Down Expand Up @@ -780,7 +792,7 @@ def test_auto_dtype(self, overlap_scheduler):
task.evaluate(llm)


@pytest.mark.timeout(3600)
@pytest.mark.timeout(DEFAULT_TEST_TIMEOUT)
class TestQwen3_8B(LlmapiAccuracyTestHarness):
MODEL_NAME = "Qwen3/Qwen3-8B"
MODEL_PATH = f"{llm_models_root()}/Qwen3/Qwen3-8B-FP8"
Expand Down Expand Up @@ -897,7 +909,7 @@ def test_chunked_prefill(self):


@skip_pre_blackwell
@pytest.mark.timeout(3600)
@pytest.mark.timeout(DEFAULT_TEST_TIMEOUT)
class TestQwen3_30B_A3B(LlmapiAccuracyTestHarness):
FP4_MODEL = f"{llm_models_root()}/Qwen3/saved_models_Qwen3-30B-A3B_nvfp4_hf"
FP8_MODEL = f"{llm_models_root()}/Qwen3/saved_models_Qwen3-30B-A3B_fp8_hf"
Expand Down