minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[CI][RFC] Replace black-jupyter with ruff-format (#37210)

Browse Source 
Co-authored-by: Alison Shao <a.shao@wustl.edu>
This commit is contained in:
Alex Nails
2026-09-02 19:46:08 -07:00
committed by GitHub
co-authored by Alison Shao
parent 2641e427be
commit 28262c20df
1411 changed files with 7766 additions and 8176 deletions
Download Patch File Download Diff File Expand all files Collapse all files
sgl-model-gateway/bindings/python/src/sglang_router/mini_lb.py
-2
View File
@@ -131,7 +131,6 @@ class MiniLoadBalancer:
total=self.timeout
) # Add timeout for request reliability
) as session:
tasks = [
session.post(f"{prefill_server}/{endpoint}", json=prefill_req),
session.post(f"{decode_server}/{endpoint}", json=decode_req),
@@ -141,7 +140,6 @@ class MiniLoadBalancer:
prefill_response, decode_response = await asyncio.gather(*tasks)
if "return_logprob" in modified_request:
prefill_json = await prefill_response.json()
ret_json = await decode_response.json()
sgl-model-gateway/bindings/python/src/sglang_router/router_args.py
-1
View File
@@ -1069,7 +1069,6 @@ class RouterArgs:
prefill_urls = []
for prefill_args in prefill_list:
url = prefill_args[0]
# Handle optional bootstrap port
sgl-model-gateway/bindings/python/tests/test_startup_sequence.py
+1 -2
View File
@@ -788,7 +788,7 @@ def test_find_available_ports_and_wait_health(monkeypatch):
monkeypatch.setattr(
ls.time,
"perf_counter",
lambda: (base.__setitem__("t", base["t"] + 0.1) or base["t"]),
lambda: base.__setitem__("t", base["t"] + 0.1) or base["t"],
)
assert ls.wait_for_server_health("127.0.0.1", 12345, timeout=1)
@@ -922,7 +922,6 @@ def test_launch_server_process_declares_on_a_resolved_record(monkeypatch):
)
with patch("sglang_router.launch_router.logger") as mock_logger:
with pytest.raises(
ValueError, match="PD disaggregation mode requires --prefill"
):
sgl-model-gateway/e2e_test/chat_completions/test_enable_thinking.py
+6 -6
View File
@@ -120,9 +120,9 @@ class TestEnableThinking:
if "content" in delta and delta["content"]:
has_content = True
assert (
has_reasoning
), "The reasoning content is not included in the stream response"
assert has_reasoning, (
"The reasoning content is not included in the stream response"
)
assert has_content, "The stream response does not contain normal content"
def test_stream_chat_completion_without_reasoning(self, setup_backend):
@@ -162,7 +162,7 @@ class TestEnableThinking:
if "content" in delta and delta["content"]:
has_content = True
assert (
not has_reasoning
), "The reasoning content should not be included in the stream response"
assert not has_reasoning, (
"The reasoning content should not be included in the stream response"
)
assert has_content, "The stream response does not contain normal content"
sgl-model-gateway/e2e_test/chat_completions/test_function_calling.py
+69 -65
View File
@@ -153,9 +153,9 @@ class TestOpenAIServerFunctionCalling:
tool_calls = response.choices[0].message.tool_calls
assert (
isinstance(tool_calls, list) and len(tool_calls) > 0
), "tool_calls should be a non-empty list"
assert isinstance(tool_calls, list) and len(tool_calls) > 0, (
"tool_calls should be a non-empty list"
)
function_name = tool_calls[0].function.name
assert function_name == "add", "Function name should be 'add'"
@@ -221,20 +221,20 @@ class TestOpenAIServerFunctionCalling:
if choice.delta.tool_calls:
tool_call = choice.delta.tool_calls[0]
if tool_call.function.name:
assert (
tool_call.function.name == "get_current_weather"
), "Function name should be 'get_current_weather'"
assert tool_call.function.name == "get_current_weather", (
"Function name should be 'get_current_weather'"
)
found_function_name = True
break
assert (
found_function_name
), "Target function name 'get_current_weather' was not found in the streaming chunks"
assert found_function_name, (
"Target function name 'get_current_weather' was not found in the streaming chunks"
)
finish_reason = chunks[-1].choices[0].finish_reason
assert (
finish_reason == "tool_calls"
), "Final response of function calling should have finish_reason 'tool_calls'"
assert finish_reason == "tool_calls", (
"Final response of function calling should have finish_reason 'tool_calls'"
)
def test_function_calling_streaming_args_parsing(self, setup_backend):
"""Test: Whether the function call arguments returned in streaming mode can be correctly concatenated into valid JSON.
@@ -299,14 +299,14 @@ class TestOpenAIServerFunctionCalling:
assert function_name == "add", "Function name should be 'add'"
joined_args = "".join(argument_fragments)
assert (
len(joined_args) > 0
), "No parameter fragments were returned in the function call"
assert len(joined_args) > 0, (
"No parameter fragments were returned in the function call"
)
finish_reason = chunks[-1].choices[0].finish_reason
assert (
finish_reason == "tool_calls"
), "Final response of function calling should have finish_reason 'tool_calls'"
assert finish_reason == "tool_calls", (
"Final response of function calling should have finish_reason 'tool_calls'"
)
# Check whether the concatenated JSON is valid
try:
@@ -445,21 +445,21 @@ class TestOpenAIServerFunctionCalling:
arguments = tool_calls[0].function.arguments
args_obj = json.loads(arguments)
assert (
function_name == "get_weather"
), f"Function name should be 'get_weather', got: {function_name}"
assert (
"city" in args_obj
), f"Function arguments should have 'city', got: {args_obj}"
assert function_name == "get_weather", (
f"Function name should be 'get_weather', got: {function_name}"
)
assert "city" in args_obj, (
f"Function arguments should have 'city', got: {args_obj}"
)
# Make the test more robust by checking type and accepting valid responses
city_value = args_obj["city"]
assert isinstance(
city_value, str
), f"Parameter city should be a string, got: {type(city_value)}"
assert (
"Paris" in city_value or "France" in city_value
), f"Parameter city should contain either 'Paris' or 'France', got: {city_value}"
assert isinstance(city_value, str), (
f"Parameter city should be a string, got: {type(city_value)}"
)
assert "Paris" in city_value or "France" in city_value, (
f"Parameter city should contain either 'Paris' or 'France', got: {city_value}"
)
def test_function_call_specific(self, setup_backend):
"""Test: Whether tool_choice: ToolChoice works as expected.
@@ -592,9 +592,9 @@ class TestOpenAIServerFunctionCalling:
finish_reason_chunks[index].append(choice.finish_reason)
# Verify we got finish_reason chunks for both indices
assert (
len(finish_reason_chunks) == 2
), f"Expected finish_reason chunks for 2 indices, got {len(finish_reason_chunks)}"
assert len(finish_reason_chunks) == 2, (
f"Expected finish_reason chunks for 2 indices, got {len(finish_reason_chunks)}"
)
# Verify both index 0 and 1 have finish_reason
assert 0 in finish_reason_chunks, "Missing finish_reason chunk for index 0"
@@ -602,9 +602,9 @@ class TestOpenAIServerFunctionCalling:
# Verify the finish_reason is "tool_calls" since we forced tool calls
for index, reasons in finish_reason_chunks.items():
assert (
reasons[-1] == "tool_calls"
), f"Expected finish_reason 'tool_calls' for index {index}, got {reasons[-1]}"
assert reasons[-1] == "tool_calls", (
f"Expected finish_reason 'tool_calls' for index {index}, got {reasons[-1]}"
)
def test_function_calling_streaming_no_tool_call(self, setup_backend):
"""Test: Whether the finish_reason is stop in streaming mode when no tool call is given.
@@ -663,14 +663,14 @@ class TestOpenAIServerFunctionCalling:
found_tool_call = True
break
assert (
not found_tool_call
), "Shouldn't have any tool_call in the streaming chunks"
assert not found_tool_call, (
"Shouldn't have any tool_call in the streaming chunks"
)
finish_reason = chunks[-1].choices[0].finish_reason
assert (
finish_reason == "stop"
), "Final response of no function calling should have finish_reason 'stop'"
assert finish_reason == "stop", (
"Final response of no function calling should have finish_reason 'stop'"
)
def test_streaming_multiple_choices_without_tools(self, setup_backend):
"""Test: Verify that each choice gets its own finish_reason chunk without tool calls.
@@ -705,9 +705,9 @@ class TestOpenAIServerFunctionCalling:
finish_reason_chunks[index].append(choice.finish_reason)
# Verify we got finish_reason chunks for both indices
assert (
len(finish_reason_chunks) == 2
), f"Expected finish_reason chunks for 2 indices, got {len(finish_reason_chunks)}"
assert len(finish_reason_chunks) == 2, (
f"Expected finish_reason chunks for 2 indices, got {len(finish_reason_chunks)}"
)
# Verify both index 0 and 1 have finish_reason
assert 0 in finish_reason_chunks, "Missing finish_reason chunk for index 0"
@@ -718,7 +718,9 @@ class TestOpenAIServerFunctionCalling:
assert reasons[-1] in [
"stop",
"length",
], f"Expected finish_reason 'stop' or 'length' for index {index}, got {reasons[-1]}"
], (
f"Expected finish_reason 'stop' or 'length' for index {index}, got {reasons[-1]}"
)
# =============================================================================
@@ -749,9 +751,9 @@ class TestOpenAIPythonicFunctionCalling:
assert isinstance(tool_calls, list), "No tool_calls found"
assert len(tool_calls) >= 1
names = [tc.function.name for tc in tool_calls]
assert (
"get_weather" in names or "get_tourist_attractions" in names
), f"Function name '{names}' should contain either 'get_weather' or 'get_tourist_attractions'"
assert "get_weather" in names or "get_tourist_attractions" in names, (
f"Function name '{names}' should contain either 'get_weather' or 'get_tourist_attractions'"
)
def test_pythonic_tool_call_streaming(self, setup_backend):
"""Test: Streaming pythonic tool call format; assert tool_call index is present."""
@@ -782,7 +784,9 @@ class TestOpenAIPythonicFunctionCalling:
assert found_index, "No index field found in any streamed tool_call"
assert (
"get_weather" in found_names or "get_tourist_attractions" in found_names
), f"Function name '{found_names}' should contain either 'get_weather' or 'get_tourist_attractions'"
), (
f"Function name '{found_names}' should contain either 'get_weather' or 'get_tourist_attractions'"
)
# =============================================================================
@@ -1132,13 +1136,13 @@ class _TestToolChoiceBase:
tool_call["id"] = tool_call_delta.id
if tool_call_delta.function:
if tool_call_delta.function.name:
tool_call["function"][
"name"
] = tool_call_delta.function.name
tool_call["function"]["name"] = (
tool_call_delta.function.name
)
if tool_call_delta.function.arguments:
tool_call["function"][
"arguments"
] += tool_call_delta.function.arguments
tool_call["function"]["arguments"] += (
tool_call_delta.function.arguments
)
assert len(tool_calls_by_index) > 0
@@ -1271,9 +1275,9 @@ class _TestToolChoiceBase:
assert len(tool_calls) == 2, f"Expected 2 tool calls, got {len(tool_calls)}"
called_functions = {call.function.name for call in tool_calls}
assert (
called_functions == expected_functions
), f"Expected functions {expected_functions}, got {called_functions}"
assert called_functions == expected_functions, (
f"Expected functions {expected_functions}, got {called_functions}"
)
def test_multi_tool_scenario_required(self, setup_backend):
"""Test multi-tool scenario with tool_choice='required'."""
@@ -1307,9 +1311,9 @@ class _TestToolChoiceBase:
if self._is_flaky_test("test_multi_tool_scenario_required"):
# For flaky tests, just ensure basic functionality works
assert (
len(tool_calls) > 0
), f"Expected at least 1 tool call, got {len(tool_calls)}"
assert len(tool_calls) > 0, (
f"Expected at least 1 tool call, got {len(tool_calls)}"
)
for call in tool_calls:
assert call.function.name in available_names
else:
@@ -1317,9 +1321,9 @@ class _TestToolChoiceBase:
assert len(tool_calls) == 2, f"Expected 2 tool calls, got {len(tool_calls)}"
called_functions = {call.function.name for call in tool_calls}
assert (
called_functions == expected_functions
), f"Expected functions {expected_functions}, got {called_functions}"
assert called_functions == expected_functions, (
f"Expected functions {expected_functions}, got {called_functions}"
)
def test_error_handling_invalid_tool_choice(self, setup_backend):
"""Test error handling for invalid tool_choice."""
sgl-model-gateway/e2e_test/chat_completions/test_openai_server.py
+15 -15
View File
@@ -172,9 +172,9 @@ The SmartHome Mini is a compact smart home assistant available in black or white
ret_num_top_logprobs = len(
response.choices[0].logprobs.content[0].top_logprobs
)
assert (
ret_num_top_logprobs == logprobs
), f"{ret_num_top_logprobs} vs {logprobs}"
assert ret_num_top_logprobs == logprobs, (
f"{ret_num_top_logprobs} vs {logprobs}"
)
assert len(response.choices) == parallel_sample_num
assert response.choices[0].message.role == "assistant"
@@ -223,9 +223,9 @@ The SmartHome Mini is a compact smart home assistant available in black or white
data = response.choices[0].delta
if is_firsts.get(index, True):
assert (
data.role == "assistant"
), "data.role was not 'assistant' for first chunk"
assert data.role == "assistant", (
"data.role was not 'assistant' for first chunk"
)
is_firsts[index] = False
continue
@@ -240,9 +240,9 @@ The SmartHome Mini is a compact smart home assistant available in black or white
ret_num_top_logprobs = len(
response.choices[0].logprobs.content[0].top_logprobs
)
assert (
ret_num_top_logprobs == logprobs
), f"{ret_num_top_logprobs} vs {logprobs}"
assert ret_num_top_logprobs == logprobs, (
f"{ret_num_top_logprobs} vs {logprobs}"
)
assert (
isinstance(data.content, str)
@@ -254,14 +254,14 @@ The SmartHome Mini is a compact smart home assistant available in black or white
assert response.created
for index in range(parallel_sample_num):
assert not is_firsts.get(
index, True
), f"index {index} is not found in the response"
assert not is_firsts.get(index, True), (
f"index {index} is not found in the response"
)
for index in range(parallel_sample_num):
assert (
index in finish_reason_counts
), f"No finish_reason found for index {index}"
assert index in finish_reason_counts, (
f"No finish_reason found for index {index}"
)
assert finish_reason_counts[index] == 1, (
f"Expected 1 finish_reason chunk for index {index}, "
f"got {finish_reason_counts[index]}"
sgl-model-gateway/e2e_test/chat_completions/test_validation.py
+6 -6
View File
@@ -91,9 +91,9 @@ class TestIgnoreEOS:
# The ignore_eos response should either:
# 1. Have more tokens than the default response (if default stopped at EOS before max_tokens)
# 2. Have exactly max_tokens (if it reached the max_tokens limit)
assert (
ignore_eos_tokens > default_tokens or ignore_eos_tokens >= max_tokens
), f"ignore_eos did not generate more tokens: {ignore_eos_tokens} vs {default_tokens}"
assert ignore_eos_tokens > default_tokens or ignore_eos_tokens >= max_tokens, (
f"ignore_eos did not generate more tokens: {ignore_eos_tokens} vs {default_tokens}"
)
assert response_ignore_eos.choices[0].finish_reason == "length", (
f"Expected finish_reason='length' for ignore_eos=True, "
@@ -158,9 +158,9 @@ class TestLargeMaxNewTokens:
# Verify all requests completed successfully
assert len(responses) == num_requests
for i, response in enumerate(responses):
assert response.choices[
0
].message.content, f"Request {i} returned empty content"
assert response.choices[0].message.content, (
f"Request {i} returned empty content"
)
assert response.choices[0].finish_reason in ("stop", "length"), (
f"Request {i} had unexpected finish_reason: "
f"{response.choices[0].finish_reason}"
sgl-model-gateway/e2e_test/embeddings/test_correctness.py
+6 -6
View File
@@ -239,9 +239,9 @@ class TestEmbeddingCorrectness:
# Verify all similarities are close to 1.0
for j, sim in enumerate(similarities):
assert (
abs(sim - 1.0) < tolerance
), f"Set {i+1}, text {j+1}: similarity {sim:.4f} not close to 1.0"
assert abs(sim - 1.0) < tolerance, (
f"Set {i + 1}, text {j + 1}: similarity {sim:.4f} not close to 1.0"
)
logger.info("Semantic similarity test set %d passed", i + 1)
@@ -273,8 +273,8 @@ class TestEmbeddingCorrectness:
logger.info("Gateway relevance scores: %s", scores_gateway)
logger.info("HF relevance scores: %s", scores_hf)
assert np.allclose(
scores_gateway, scores_hf, atol=tolerance
), f"Scores differ beyond tolerance:\nGateway: {scores_gateway}\nHF: {scores_hf}"
assert np.allclose(scores_gateway, scores_hf, atol=tolerance), (
f"Scores differ beyond tolerance:\nGateway: {scores_gateway}\nHF: {scores_hf}"
)
logger.info("Relevance scores comparison passed")
sgl-model-gateway/e2e_test/infra/gpu_monitor.py
+6 -6
View File
@@ -300,17 +300,17 @@ class GPUMonitor:
mean_threshold = thresholds.get("gpu_util_mean_min")
if mean_threshold is not None:
mean_value = overall.get("mean", 0.0)
assert (
mean_value >= mean_threshold
), f"GPU utilization mean below threshold: {mean_value:.2f}% < {mean_threshold}%"
assert mean_value >= mean_threshold, (
f"GPU utilization mean below threshold: {mean_value:.2f}% < {mean_threshold}%"
)
p50_threshold = thresholds.get("gpu_util_p50_min")
if p50_threshold is not None:
p50_value = overall.get("p50")
if p50_value is not None:
assert (
p50_value >= p50_threshold
), f"GPU utilization p50 below threshold: {p50_value:.2f}% < {p50_threshold}%"
assert p50_value >= p50_threshold, (
f"GPU utilization p50 below threshold: {p50_value:.2f}% < {p50_threshold}%"
)
def should_monitor(thresholds: dict[str, Any] | None) -> bool:
sgl-model-gateway/e2e_test/k8s_integration/test_cross_namespace.py
+6 -6
View File
@@ -255,12 +255,12 @@ class TestClusterWideDiscovery:
# A regression that quietly hardcoded a namespace filter would
# still produce total=2 if labels happened to match elsewhere,
# but only one of these IPs would surface.
assert any(
ip_a in u for u in urls
), f"worker_a IP {ip_a} (ns {NAMESPACE}) not in {urls}"
assert any(
ip_b in u for u in urls
), f"worker_b IP {ip_b} (ns {EXTRA_NAMESPACE}) not in {urls}"
assert any(ip_a in u for u in urls), (
f"worker_a IP {ip_a} (ns {NAMESPACE}) not in {urls}"
)
assert any(ip_b in u for u in urls), (
f"worker_b IP {ip_b} (ns {EXTRA_NAMESPACE}) not in {urls}"
)
finally:
_safe_delete_pod(worker_a, NAMESPACE)
_safe_delete_pod(worker_b, EXTRA_NAMESPACE)
sgl-model-gateway/e2e_test/k8s_integration/test_lifecycle.py
+22 -18
View File
@@ -256,9 +256,9 @@ class TestGatewayRestart:
"-l",
"app=smg-gateway-restart",
)
assert res.get(
"items"
), "No pods found for selector app=smg-gateway-restart"
assert res.get("items"), (
"No pods found for selector app=smg-gateway-restart"
)
old_pod = res["items"][0]["metadata"]["name"]
_kubectl(
"delete",
@@ -304,9 +304,9 @@ class TestGatewayRestart:
logger.info("Workers after restart: %s", urls_after)
# No duplicates: each pod should appear exactly once.
assert len(urls_after) == len(
set(urls_after)
), f"Duplicate worker registrations after gateway restart: {urls_after}"
assert len(urls_after) == len(set(urls_after)), (
f"Duplicate worker registrations after gateway restart: {urls_after}"
)
# Set equality: the same workers come back, neither dropped
# nor duplicated.
assert set(urls_after) == set(urls_before), (
@@ -364,9 +364,9 @@ class TestPodIpChange:
ip_before = _get_pod_ip(pod_name)
urls_before = {w["url"] for w in _get_workers(gateway_url)["workers"]}
assert any(
ip_before in url for url in urls_before
), f"Expected initial worker URL containing {ip_before}, got {urls_before}"
assert any(ip_before in url for url in urls_before), (
f"Expected initial worker URL containing {ip_before}, got {urls_before}"
)
logger.info("Pod IP before: %s, urls: %s", ip_before, urls_before)
# Force-delete and wait until the registry no longer references
@@ -384,9 +384,11 @@ class TestPodIpChange:
)
_wait_for_pod_gone(pod_name)
_poll_until(
lambda: not any(
ip_before in w["url"]
for w in _get_workers(gateway_url).get("workers", [])
lambda: (
not any(
ip_before in w["url"]
for w in _get_workers(gateway_url).get("workers", [])
)
),
f"stale worker for IP {ip_before} removed",
timeout=RECONCILIATION_WAIT_SECS,
@@ -431,9 +433,9 @@ class TestPodIpChange:
f"Expected exactly one worker URL containing current IP "
f"{ip_after}, got {matching_after} (all urls: {urls_after})"
)
assert not any(
ip_before in u for u in urls_after
), f"Stale URL with old IP {ip_before} still in registry: {urls_after}"
assert not any(ip_before in u for u in urls_after), (
f"Stale URL with old IP {ip_before} still in registry: {urls_after}"
)
finally:
_safe_force_delete(pod_name)
@@ -496,9 +498,11 @@ class TestGracefulDrain:
# without affecting whether *this* pod's IP got removed. The
# meaningful timing guarantee (`elapsed < grace_secs`) is below.
_poll_until(
lambda: not any(
pod_ip in w["url"]
for w in _get_workers(gateway_url).get("workers", [])
lambda: (
not any(
pod_ip in w["url"]
for w in _get_workers(gateway_url).get("workers", [])
)
),
f"worker for ip {pod_ip} deregistered after graceful delete",
timeout=RECONCILIATION_WAIT_SECS,
sgl-model-gateway/e2e_test/k8s_integration/test_multi_model.py
+3 -3
View File
@@ -196,9 +196,9 @@ class TestMultiModelSelectorIsolation:
# No URL should appear in both views — that would mean a
# selector mismatch leaked a worker into the wrong gateway.
cross_talk = set(llama_urls) & set(qwen_urls)
assert (
not cross_talk
), f"Workers leaked across model selectors: {cross_talk}"
assert not cross_talk, (
f"Workers leaked across model selectors: {cross_talk}"
)
finally:
for name in llama_workers + qwen_workers:
_safe_force_delete(name)
sgl-model-gateway/e2e_test/k8s_integration/test_pd_type_change.py
+9 -9
View File
@@ -181,9 +181,9 @@ class TestPDRolloutTypeChange:
by_type = _get_workers_by_type(pd_gateway)
logger.info("Workers by type: %s", json.dumps(by_type, indent=2))
assert (
"prefill" in by_type
), f"Expected prefill, got: {list(by_type.keys())}"
assert "prefill" in by_type, (
f"Expected prefill, got: {list(by_type.keys())}"
)
finally:
_safe_delete_pod(pod_name)
@@ -257,12 +257,12 @@ class TestPDRolloutTypeChange:
by_type = _get_workers_by_type(pd_gateway)
logger.info("After rollout: %s", json.dumps(by_type, indent=2))
assert (
"decode" in by_type
), f"Expected decode worker after rollout, got: {list(by_type.keys())}"
assert (
"prefill" not in by_type
), "Stale prefill worker persists after rollout"
assert "decode" in by_type, (
f"Expected decode worker after rollout, got: {list(by_type.keys())}"
)
assert "prefill" not in by_type, (
"Stale prefill worker persists after rollout"
)
finally:
_safe_delete_pod(pod_name)
sgl-model-gateway/e2e_test/k8s_integration/test_reconciliation.py
+9 -9
View File
@@ -377,9 +377,9 @@ class TestReconciliationMetrics:
{"source": "kubernetes", "result": "success"},
)
logger.info("Registration success metric: %s", reg_value)
assert (
reg_value is not None and reg_value >= 1
), f"Expected at least 1 registration, got {reg_value}"
assert reg_value is not None and reg_value >= 1, (
f"Expected at least 1 registration, got {reg_value}"
)
gauge_value = _parse_metric_value(
metrics_text,
@@ -387,9 +387,9 @@ class TestReconciliationMetrics:
{"source": "kubernetes"},
)
logger.info("Workers discovered gauge: %s", gauge_value)
assert (
gauge_value is not None and gauge_value >= 1
), f"Expected workers_discovered >= 1, got {gauge_value}"
assert gauge_value is not None and gauge_value >= 1, (
f"Expected workers_discovered >= 1, got {gauge_value}"
)
finally:
_safe_delete_worker_pod(pod_name)
@@ -497,9 +497,9 @@ class TestReconciliationConsistency:
logger.info("Worker count samples over time: %s", samples)
assert all(
s == stable_count for s in samples
), f"Worker count fluctuated: {samples} (expected stable at {stable_count})"
assert all(s == stable_count for s in samples), (
f"Worker count fluctuated: {samples} (expected stable at {stable_count})"
)
finally:
for name in pod_names:
sgl-model-gateway/e2e_test/responses/test_streaming_events.py
+24 -24
View File
@@ -58,15 +58,15 @@ class TestStreamingEventsLocal:
first_item_event = output_item_added_events[0]
assert first_item_event.item is not None
assert first_item_event.output_index is not None
assert (
first_item_event.output_index == 0
), "First output item must have output_index: 0 (zero-based indexing)"
assert first_item_event.output_index == 0, (
"First output item must have output_index: 0 (zero-based indexing)"
)
# Verify subsequent items increment correctly
for i, event in enumerate(output_item_added_events):
assert (
event.output_index == i
), f"Output item {i} should have output_index: {i}"
assert event.output_index == i, (
f"Output item {i} should have output_index: {i}"
)
# Verify output_item.done event exists
output_item_done_events = [
@@ -101,9 +101,9 @@ class TestStreamingEventsLocal:
output_item_added_events = [
event for event in events if event.type == "response.output_item.added"
]
assert len(output_item_added_events) == len(
output_array
), "Number of output_item.added events should match output array length"
assert len(output_item_added_events) == len(output_array), (
"Number of output_item.added events should match output array length"
)
# =============================================================================
@@ -147,15 +147,15 @@ class TestStreamingEventsHarmony:
first_item_event = output_item_added_events[0]
assert first_item_event.item is not None
assert first_item_event.output_index is not None
assert (
first_item_event.output_index == 0
), "First output item must have output_index: 0 (zero-based indexing)"
assert first_item_event.output_index == 0, (
"First output item must have output_index: 0 (zero-based indexing)"
)
# Verify subsequent items increment correctly
for i, event in enumerate(output_item_added_events):
assert (
event.output_index == i
), f"Output item {i} should have output_index: {i}"
assert event.output_index == i, (
f"Output item {i} should have output_index: {i}"
)
# Verify output_item.done event exists
output_item_done_events = [
@@ -190,9 +190,9 @@ class TestStreamingEventsHarmony:
output_item_added_events = [
event for event in events if event.type == "response.output_item.added"
]
assert len(output_item_added_events) == len(
output_array
), "Number of output_item.added events should match output array length"
assert len(output_item_added_events) == len(output_array), (
"Number of output_item.added events should match output array length"
)
def test_reasoning_content(self, setup_backend):
"""Test that reasoning content has correct zero-based output_index.
@@ -228,16 +228,16 @@ class TestStreamingEventsHarmony:
# If reasoning is present, verify it has output_index: 0
if reasoning_items:
reasoning_item = reasoning_items[0]
assert (
reasoning_item.output_index == 0
), "Reasoning item should have output_index: 0"
assert reasoning_item.output_index == 0, (
"Reasoning item should have output_index: 0"
)
# If message is present after reasoning, verify it has output_index: 1
if reasoning_items and message_items:
message_item = message_items[0]
assert (
message_item.output_index == 1
), "Message item after reasoning should have output_index: 1"
assert message_item.output_index == 1, (
"Message item after reasoning should have output_index: 1"
)
# Find response.completed event
completed_events = [
sgl-model-gateway/e2e_test/responses/test_tools_call.py
+48 -48
View File
@@ -168,9 +168,9 @@ class TestToolCallingCloud:
# Check for function_call in output
function_calls = [item for item in output if item.type == "function_call"]
assert (
len(function_calls) > 0
), "Response should contain at least one function_call"
assert len(function_calls) > 0, (
"Response should contain at least one function_call"
)
# Verify function_call structure
function_call = function_calls[0]
@@ -285,30 +285,30 @@ class TestToolCallingCloud:
event_types = [event.type for event in events]
assert "response.created" in event_types, "Should have response.created event"
assert (
"response.completed" in event_types
), "Should have response.completed event"
assert (
"response.output_item.added" in event_types
), "Should have output_item.added events"
assert (
"response.mcp_list_tools.in_progress" in event_types
), "Should have mcp_list_tools.in_progress event"
assert (
"response.mcp_list_tools.completed" in event_types
), "Should have mcp_list_tools.completed event"
assert (
"response.mcp_call.in_progress" in event_types
), "Should have mcp_call.in_progress event"
assert (
"response.mcp_call_arguments.delta" in event_types
), "Should have mcp_call_arguments.delta event"
assert (
"response.mcp_call_arguments.done" in event_types
), "Should have mcp_call_arguments.done event"
assert (
"response.mcp_call.completed" in event_types
), "Should have mcp_call.completed event"
assert "response.completed" in event_types, (
"Should have response.completed event"
)
assert "response.output_item.added" in event_types, (
"Should have output_item.added events"
)
assert "response.mcp_list_tools.in_progress" in event_types, (
"Should have mcp_list_tools.in_progress event"
)
assert "response.mcp_list_tools.completed" in event_types, (
"Should have mcp_list_tools.completed event"
)
assert "response.mcp_call.in_progress" in event_types, (
"Should have mcp_call.in_progress event"
)
assert "response.mcp_call_arguments.delta" in event_types, (
"Should have mcp_call_arguments.delta event"
)
assert "response.mcp_call_arguments.done" in event_types, (
"Should have mcp_call_arguments.done event"
)
assert "response.mcp_call.completed" in event_types, (
"Should have mcp_call.completed event"
)
completed_events = [e for e in events if e.type == "response.completed"]
assert len(completed_events) == 1
@@ -336,18 +336,18 @@ class TestToolCallingCloud:
assert mcp_call.output is not None
# Strict validation for cloud backends - check for text output events
assert (
"response.content_part.added" in event_types
), "Should have content_part.added event"
assert (
"response.output_text.delta" in event_types
), "Should have output_text.delta events"
assert (
"response.output_text.done" in event_types
), "Should have output_text.done event"
assert (
"response.content_part.done" in event_types
), "Should have content_part.done event"
assert "response.content_part.added" in event_types, (
"Should have content_part.added event"
)
assert "response.output_text.delta" in event_types, (
"Should have output_text.delta events"
)
assert "response.output_text.done" in event_types, (
"Should have output_text.done event"
)
assert "response.content_part.done" in event_types, (
"Should have content_part.done event"
)
assert "message" in final_output_types
@@ -400,9 +400,9 @@ class TestToolChoiceHarmony:
assert len(output) > 0
function_calls = [item for item in output if item.type == "function_call"]
assert (
len(function_calls) > 0
), "Model should choose to call function with tool_choice='auto'"
assert len(function_calls) > 0, (
"Model should choose to call function with tool_choice='auto'"
)
def test_tool_choice_required(self, setup_backend):
"""Test tool_choice="required" forces the model to call at least one tool."""
@@ -423,9 +423,9 @@ class TestToolChoiceHarmony:
output = resp.output
function_calls = [item for item in output if item.type == "function_call"]
assert (
len(function_calls) > 0
), "tool_choice='required' must force at least one function call"
assert len(function_calls) > 0, (
"tool_choice='required' must force at least one function call"
)
def test_tool_choice_specific_function(self, setup_backend):
"""Test tool_choice with specific function name forces that function to be called."""
@@ -447,9 +447,9 @@ class TestToolChoiceHarmony:
output = resp.output
function_calls = [item for item in output if item.type == "function_call"]
assert len(function_calls) > 0, "Must call the specified function"
assert (
function_calls[0].name == "search_web"
), "Must call the function specified in tool_choice"
assert function_calls[0].name == "search_web", (
"Must call the function specified in tool_choice"
)
def test_tool_choice_streaming(self, setup_backend):
"""Test tool_choice parameter works correctly with streaming."""
sgl-model-gateway/e2e_test/router/test_mmlu.py
+6 -6
View File
@@ -47,9 +47,9 @@ class TestMMLU:
)
metrics = run_eval(args)
assert (
metrics["score"] >= 0.65
), f"MMLU score {metrics['score']:.2f} below threshold 0.65"
assert metrics["score"] >= 0.65, (
f"MMLU score {metrics['score']:.2f} below threshold 0.65"
)
logger.info("MMLU score: %.2f (threshold: 0.65)", metrics["score"])
def test_mmlu_extended(self, setup_backend):
@@ -70,7 +70,7 @@ class TestMMLU:
)
metrics = run_eval(args)
assert (
metrics["score"] >= 0.65
), f"MMLU score {metrics['score']:.2f} below threshold 0.65"
assert metrics["score"] >= 0.65, (
f"MMLU score {metrics['score']:.2f} below threshold 0.65"
)
logger.info("MMLU extended score: %.2f (threshold: 0.65)", metrics["score"])
sgl-model-gateway/e2e_test/router/test_pd_mmlu.py
+3 -3
View File
@@ -55,7 +55,7 @@ class TestPDMMLU:
)
metrics = run_eval(args)
assert (
metrics["score"] >= 0.65
), f"PD MMLU score {metrics['score']:.2f} below threshold 0.65"
assert metrics["score"] >= 0.65, (
f"PD MMLU score {metrics['score']:.2f} below threshold 0.65"
)
logger.info("PD MMLU score: %.2f (threshold: 0.65)", metrics["score"])
sgl-model-gateway/e2e_test/router/test_worker_api.py
+3 -3
View File
@@ -201,9 +201,9 @@ class TestDisableHealthCheck:
worker.metadata.get("disable_health_check"),
)
# Worker should be healthy immediately
assert (
worker.status == "healthy"
), "Worker should be healthy when health checks disabled"
assert worker.status == "healthy", (
"Worker should be healthy when health checks disabled"
)
finally:
gateway.shutdown()
http_instance.release()