Enable async-assert invariant probes by default in CI (#27461)

python/sglang/test/server_fixtures/eagle_fixture.py

@@ -4,7 +4,6 @@ import time
 
 import requests
 
-from sglang.srt.environ import envs
 from sglang.srt.utils.common import kill_process_tree
 from sglang.test.test_utils import (
     DEFAULT_DRAFT_MODEL_EAGLE,
@@ -37,21 +36,20 @@ class EagleServerBase(CustomTestCase):
     @classmethod
     def setUpClass(cls):
         cls.base_url = DEFAULT_URL_FOR_TEST
-        with envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True):
-            cls.process = popen_launch_server(
-                cls.target_model,
-                cls.base_url,
-                timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-                other_args=[
-                    f"--speculative-algorithm={cls.spec_algo}",
-                    f"--speculative-draft-model-path={cls.draft_model}",
-                    f"--speculative-num-steps={cls.spec_steps}",
-                    f"--speculative-eagle-topk={cls.spec_topk}",
-                    f"--speculative-num-draft-tokens={cls.spec_tokens}",
-                    f"--mem-fraction-static={cls.mem_fraction_static}",
-                ]
-                + cls.extra_args,
-            )
+        cls.process = popen_launch_server(
+            cls.target_model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=[
+                f"--speculative-algorithm={cls.spec_algo}",
+                f"--speculative-draft-model-path={cls.draft_model}",
+                f"--speculative-num-steps={cls.spec_steps}",
+                f"--speculative-eagle-topk={cls.spec_topk}",
+                f"--speculative-num-draft-tokens={cls.spec_tokens}",
+                f"--mem-fraction-static={cls.mem_fraction_static}",
+            ]
+            + cls.extra_args,
+        )
 
     @classmethod
     def tearDownClass(cls):

python/sglang/test/server_fixtures/spec_eagle_fixture.py

@@ -122,7 +122,6 @@ class SpecEagleServerBase(CustomTestCase):
         cls.target_model = cls.model
         cls._tokenizer = None
         with contextlib.ExitStack() as stack:
-            stack.enter_context(envs.SGLANG_ENABLE_ASYNC_ASSERT.override(True))
             stack.enter_context(
                 envs.SGLANG_ALLOW_OVERWRITE_LONGER_CONTEXT_LEN.override(True)
             )