minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

Clean up environ.py: remove dead env vars, unify deprecation handling, move examples to a unit test (#35060)

Browse Source
This commit is contained in:
Lianmin Zheng
2026-08-17 06:53:34 -07:00
committed by GitHub
parent d97b796c16
commit af743371cc
12 changed files with 294 additions and 269 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/registered/npu/accuracy/minimax_m2_5/test_npu_minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms_gpqa.py
+2 -1
View File
@@ -26,7 +26,6 @@ MINIMAX_M2_5_W8A8_4P_IN64K_OUT1K_PREFIX90_ENVS = {
"ASCEND_USE_FIA": "1",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_NPU_FUSED_MOE_MODE": "2",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "140000",
"DEEP_NORMAL_MODE_USE_INT8_QUANT": "1",
"DEEPEP_HCCL_BUFFSIZE": "1024",
@@ -65,6 +64,8 @@ MINIMAX_M2_5_W8A8_4P_IN64K_OUT1K_PREFIX90_OTHER_ARGS = [
26,
"--moe-a2a-backend",
"ascend_fuseep",
"--fuseep-mode",
2,
"--deepep-mode",
"auto",
"--quantization",
test/registered/npu/basic_function/speculative_inference/test_npu_speculative_moe_a2a_backend.py
+2 -1
View File
@@ -31,7 +31,6 @@ class TestAscendDistTimeout(CustomTestCase):
os.environ["HCCL_BUFFSIZE"] = "2048"
os.environ["SGLANG_ENABLE_OVERLAP_PLAN_STREAM"] = "1"
os.environ["SGLANG_ENABLE_SPEC_V2"] = "1"
os.environ["SGLANG_NPU_FUSED_MOE_MODE"] = "1"
os.environ["TRANSFORMERS_VERBOSITY"] = "error"
cls.env = os.environ.copy()
cls.common_args = [
@@ -58,6 +57,8 @@ class TestAscendDistTimeout(CustomTestCase):
2,
"--moe-a2a-backend",
"ascend_fuseep",
"--fuseep-mode",
1,
"--deepep-mode",
"auto",
"--speculative-draft-model-quantization",
test/registered/npu/performance/minimax_m2_5/test_npu_minimax_m2_5_w8a8_4p_in64k_out1k_prefix90_50ms.py
+2 -1
View File
@@ -20,7 +20,6 @@ MINIMAX_M2_5_W8A8_4P_IN64K_OUT1K_PREFIX90_ENVS = {
"ASCEND_USE_FIA": "1",
"SGLANG_SET_CPU_AFFINITY": "1",
"SGLANG_ENABLE_OVERLAP_PLAN_STREAM": "1",
"SGLANG_NPU_FUSED_MOE_MODE": "2",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "140000",
"DEEP_NORMAL_MODE_USE_INT8_QUANT": "1",
"DEEPEP_HCCL_BUFFSIZE": "1024",
@@ -62,6 +61,8 @@ MINIMAX_M2_5_W8A8_4P_IN64K_OUT1K_PREFIX90_OTHER_ARGS = [
26,
"--moe-a2a-backend",
"ascend_fuseep",
"--fuseep-mode",
2,
"--deepep-mode",
"auto",
"--quantization",
test/registered/npu/performance/qwen3_235b_a22b/test_npu_qwen3_235b_w8a8_8p_in3k5_out1k5_50ms.py
+2 -1
View File
@@ -26,7 +26,6 @@ QWEN3_235B_ENVS = {
"SGLANG_NPU_PROFILING": "0",
"SGLANG_NPU_PROFILING_BS": "27",
"SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK": "188416",
"SGLANG_NPU_FUSED_MOE_MODE": "2",
}
QWEN3_235B_OTHER_ARGS = [
@@ -60,6 +59,8 @@ QWEN3_235B_OTHER_ARGS = [
"--disable-radix-cache",
"--moe-a2a-backend",
"ascend_fuseep",
"--fuseep-mode",
2,
"--speculative-algorithm",
"EAGLE3",
"--speculative-draft-model-path",
test/registered/unit/test_environ.py
+152
View File
@@ -0,0 +1,152 @@
"""Unit tests for sglang.srt.environ: EnvField semantics and the deprecated-env registry."""
import os
import re
import subprocess
import sys
import unittest
import warnings
from contextlib import ExitStack
from sglang.srt.environ import _DEPRECATED_ENVS, _DeprecatedEnv, envs
from sglang.test.ci.ci_register import register_cpu_ci
register_cpu_ci(est_time=15, suite="base-a-test-cpu")
class TestEnvField(unittest.TestCase):
def setUp(self):
envs.SGLANG_TEST_RETRACT.clear()
self.addCleanup(envs.SGLANG_TEST_RETRACT.clear)
def test_set_get_clear_is_set(self):
self.assertFalse(envs.SGLANG_TEST_RETRACT.is_set())
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), False)
envs.SGLANG_TEST_RETRACT.set(True)
self.assertTrue(envs.SGLANG_TEST_RETRACT.is_set())
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), True)
envs.SGLANG_TEST_RETRACT.clear()
self.assertFalse(envs.SGLANG_TEST_RETRACT.is_set())
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), False)
def test_set_to_none_is_distinct_from_clear(self):
envs.SGLANG_TEST_RETRACT.set(None)
self.assertTrue(envs.SGLANG_TEST_RETRACT.is_set())
self.assertIsNone(envs.SGLANG_TEST_RETRACT.get())
envs.SGLANG_TEST_RETRACT.clear()
self.assertFalse(envs.SGLANG_TEST_RETRACT.is_set())
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), False)
def test_override_restores_previous_state(self):
envs.SGLANG_TEST_RETRACT.set(True)
with envs.SGLANG_TEST_RETRACT.override(None):
self.assertTrue(envs.SGLANG_TEST_RETRACT.is_set())
self.assertIsNone(envs.SGLANG_TEST_RETRACT.get())
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), True)
envs.SGLANG_TEST_RETRACT.set(None)
with envs.SGLANG_TEST_RETRACT.override(True):
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), True)
self.assertTrue(envs.SGLANG_TEST_RETRACT.is_set())
self.assertIsNone(envs.SGLANG_TEST_RETRACT.get())
def test_override_with_exit_stack(self):
envs.SGLANG_TEST_RETRACT.set(None)
exit_stack = ExitStack()
exit_stack.enter_context(envs.SGLANG_TEST_RETRACT.override(False))
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), False)
exit_stack.close()
self.assertIsNone(envs.SGLANG_TEST_RETRACT.get())
def test_override_is_inherited_by_subprocess(self):
command = [
sys.executable,
"-c",
"import os; print(os.getenv('SGLANG_TEST_RETRACT'))",
]
with envs.SGLANG_TEST_RETRACT.override(True):
output = subprocess.check_output(command).decode().strip()
self.assertEqual(output, "True")
output = subprocess.check_output(command).decode().strip()
self.assertEqual(output, "None")
def test_implicit_bool_raises(self):
message = re.escape(
"Please use `envs.YOUR_FLAG.get()` instead of `envs.YOUR_FLAG`"
)
with self.assertRaisesRegex(RuntimeError, message):
if envs.SGLANG_TEST_RETRACT:
pass
with self.assertRaisesRegex(RuntimeError, message):
if (1 != 1) or envs.SGLANG_TEST_RETRACT:
pass
with self.assertRaisesRegex(RuntimeError, message):
if envs.SGLANG_TEST_RETRACT or (1 == 1):
pass
def test_invalid_value_warns_and_returns_default(self):
os.environ["SGLANG_TEST_RETRACT"] = "not-a-bool"
with warnings.catch_warnings(record=True) as caught:
warnings.simplefilter("always")
self.assertIs(envs.SGLANG_TEST_RETRACT.get(), False)
self.assertIn("Invalid value", str(caught[0].message))
class TestDeprecatedEnvRegistry(unittest.TestCase):
def _apply(self, old_name, deprecation):
with warnings.catch_warnings(record=True) as caught:
warnings.simplefilter("always")
deprecation.apply(old_name)
return caught
def test_removed_env_warns_without_forwarding(self):
old_name = "SGLANG_TEST_REMOVED_ENV"
os.environ[old_name] = "1"
self.addCleanup(os.environ.pop, old_name, None)
caught = self._apply(old_name, _DeprecatedEnv())
self.assertIn(f"{old_name} is deprecated", str(caught[0].message))
def test_renamed_env_forwards_value(self):
old_name, new_name = "SGLANG_TEST_OLD_ENV", "SGLANG_TEST_NEW_ENV"
os.environ[old_name] = "abc"
self.addCleanup(os.environ.pop, old_name, None)
self.addCleanup(os.environ.pop, new_name, None)
caught = self._apply(old_name, _DeprecatedEnv(replacement=new_name))
self.assertIn(new_name, str(caught[0].message))
self.assertEqual(os.environ[new_name], "abc")
def test_unset_env_is_a_no_op(self):
caught = self._apply("SGLANG_TEST_UNSET_ENV", _DeprecatedEnv())
self.assertEqual(len(caught), 0)
def test_disable_tp_imbalance_check_polarity_is_inverted(self):
old_name = "SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK"
new_name = "SGLANG_ENABLE_TP_MEMORY_INBALANCE_CHECK"
os.environ[old_name] = "1"
self.addCleanup(os.environ.pop, old_name, None)
self.addCleanup(os.environ.pop, new_name, None)
self._apply(old_name, _DEPRECATED_ENVS[old_name])
self.assertIs(envs.SGLANG_ENABLE_TP_MEMORY_INBALANCE_CHECK.get(), False)
def test_ms_to_s_transform(self):
old_name = "SGLANG_QUEUED_TIMEOUT_MS"
os.environ[old_name] = "1500"
self.addCleanup(os.environ.pop, old_name, None)
self.addCleanup(os.environ.pop, "SGLANG_REQ_WAITING_TIMEOUT", None)
self._apply(old_name, _DEPRECATED_ENVS[old_name])
self.assertEqual(envs.SGLANG_REQ_WAITING_TIMEOUT.get(), 1.5)
if __name__ == "__main__":
unittest.main()
test/registered/unit/test_model_overrides.py
-23
View File
@@ -1773,29 +1773,6 @@ class TestGoldenModelOverrides(_IsolatedPublish):
)
self.assertEqual(_moe_runner_backend_quant_constraints(_view()), {})
def test_cutlass_moe_env_override_pass(self):
from sglang.srt.arg_groups.overrides import (
ResolvedView,
_cutlass_moe_env_override,
)
with patch("sglang.srt.environ.envs.SGLANG_CUTLASS_MOE") as e:
e.get.return_value = True
self.assertEqual(
_cutlass_moe_env_override(
ResolvedView(SimpleNamespace(quantization="fp8"))
),
{"moe_runner_backend": "cutlass"},
)
with self.assertRaises(AssertionError):
_cutlass_moe_env_override(
ResolvedView(SimpleNamespace(quantization=None))
)
e.get.return_value = False
self.assertEqual(
_cutlass_moe_env_override(ResolvedView(SimpleNamespace())), {}
)
def test_gguf_quantization_pass(self):
from sglang.srt.arg_groups.overrides import ResolvedView, _gguf_quantization