[CI] Graceful teardown for kv_canary and EAGLE spec fixtures (#32829)

test/registered/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_mamba.py

@@ -1,12 +1,10 @@
 import os
 import shutil
-import subprocess
 import tempfile
 import unittest
 
 from test_unified_radix_cache_kl_nightly import AccuracyTwoPassMixin
 
-from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kits.unified_radix_cache_kit import UnifiedRadixTreeTestMixin
 from sglang.test.kl_multiturn_utils import (
@@ -19,6 +17,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=800, stage="extra-b", runner_config="4-gpu-h100")
@@ -69,7 +68,7 @@ class TestUnifiedMambaRadixCache(UnifiedRadixTreeTestMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
 
 # ─── Mamba + HiCache L2 ──────────────────────────────────────────────────────
@@ -128,12 +127,7 @@ class TestUnifiedMambaHiCache(UnifiedRadixTreeTestMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        cls.process.terminate()
-        try:
-            cls.process.wait(timeout=60)
-        except subprocess.TimeoutExpired:
-            pass
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
 
 # ─── Mamba + HiCache L3 (file backend) ───────────────────────────────────────
@@ -194,12 +188,7 @@ class TestUnifiedMambaHiCacheL3(AccuracyTwoPassMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        cls.process.terminate()
-        try:
-            cls.process.wait(timeout=60)
-        except subprocess.TimeoutExpired:
-            pass
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
         if os.path.isdir(cls.hicache_dir):
             shutil.rmtree(cls.hicache_dir, ignore_errors=True)
 

test/registered/vlm/test_vision_openai_server_a.py

@@ -17,6 +17,7 @@ from sglang.test.vlm_utils import (
     OmniOpenAITestMixin,
     TestOpenAIMLLMServerBase,
     VideoOpenAITestMixin,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=780, stage="base-b", runner_config="1-gpu-large")
@@ -74,7 +75,7 @@ class TestQwen2VLContextLengthServer(CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
     def test_single_image_chat_completion(self):
         client = openai.Client(api_key=self.api_key, base_url=self.base_url)

test/registered/xpu/test_gemma_4_e2b.py

@@ -29,8 +29,8 @@ from sglang.srt.utils.common import is_xpu
 from sglang.test.test_utils import CustomTestCase
 from sglang.test.vlm_utils import (
     DEFAULT_URL_FOR_TEST,
-    kill_process_tree,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 MODEL = "google/gemma-4-E2B-it"
@@ -98,7 +98,7 @@ class TestGemma4E2BXPU(CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
         _empty_xpu_cache()
 
     def test_simple_qa(self):