[CI] Graceful teardown for the radix_cache server fixtures (#36605)

test/registered/radix_cache/test_radix_attention.py

@@ -13,8 +13,8 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     is_in_ci,
-    kill_process_tree,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 # RadixAttention server integration tests
@@ -44,7 +44,7 @@ class TestRadixCacheFCFS(CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
     def test_radix_attention(self):
         run_radix_attention_test(self.base_url)

test/registered/radix_cache/test_radix_cache_hit.py

@@ -1,6 +1,5 @@
 import unittest
 
-from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import (
     register_amd_ci,
     register_cpu_ci,
@@ -13,6 +12,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=55, stage="base-b", runner_config="1-gpu-small")
@@ -35,7 +35,7 @@ class TestRadixCacheHit(CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
     def test_multiturn_cache_hit(self):
         run_multiturn_cache_hit_test(

test/registered/radix_cache/unified_radix_tree/test_unified_radix_cache_hicache_pp_kl.py

@@ -5,7 +5,6 @@ import unittest
 from types import SimpleNamespace
 from urllib.parse import urlparse
 
-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 (
     AccuracyTwoPassMixin,
@@ -17,6 +16,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=900, stage="base-c", runner_config="4-gpu-h100")
@@ -105,7 +105,7 @@ class TestUnifiedQwen3HiCachePP(UnifiedRadixTreeTestMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
 
 class TestUnifiedQwen3HiCachePPL3(AccuracyTwoPassMixin, CustomTestCase):
@@ -157,7 +157,7 @@ class TestUnifiedQwen3HiCachePPL3(AccuracyTwoPassMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        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/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_cp.py

@@ -1,6 +1,5 @@
 import unittest
 
-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 get_input_ids
@@ -9,6 +8,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=950, stage="extra-b", runner_config="4-gpu-h100")
@@ -69,7 +69,7 @@ class TestUnifiedQwen3HiCacheCP(UnifiedRadixTreeTestMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
 
 if __name__ == "__main__":

test/registered/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_dcp.py

@@ -10,7 +10,6 @@ Blackwell-only: the MLA DCP decode path needs ``tokenspeed_mla`` (SM100/12x).
 import subprocess
 import unittest
 
-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 (
@@ -23,6 +22,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=1500, stage="extra-b", runner_config="4-gpu-b200")
@@ -104,7 +104,7 @@ class TestUnifiedKimiLinearDcpHiCache(UnifiedRadixTreeTestMixin, CustomTestCase)
             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 __name__ == "__main__":

test/registered/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_dsv4.py

@@ -6,7 +6,6 @@ import unittest
 
 import requests
 
-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 (
     AccuracyTwoPassMixin,
@@ -18,6 +17,7 @@ from sglang.test.test_utils import (
     CustomTestCase,
     is_in_ci,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 DSV4_FLASH_MODEL = "sgl-project/DeepSeek-V4-Flash-FP8"
@@ -109,7 +109,7 @@ class TestUnifiedDeepSeekV4FlashHiCache(UnifiedRadixTreeTestMixin, CustomTestCas
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
 
 class TestUnifiedDeepSeekV4FlashHiCachePageFirstDirect(
@@ -180,7 +180,7 @@ class TestUnifiedDeepSeekV4FlashHiCacheL3(AccuracyTwoPassMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        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)
 
@@ -257,7 +257,7 @@ class TestUnifiedDeepSeekV4FlashEagleHiCacheL3(AccuracyTwoPassMixin, CustomTestC
 
     @classmethod
     def tearDownClass(cls):
-        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/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_full.py

@@ -1,6 +1,5 @@
 import unittest
 
-from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_amd_ci, register_cuda_ci
 from sglang.test.kits.unified_radix_cache_kit import UnifiedRadixTreeTestMixin
 from sglang.test.kl_multiturn_utils import get_input_ids
@@ -9,6 +8,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=250, stage="base-b", runner_config="2-gpu-large")
@@ -44,7 +44,7 @@ class TestUnifiedFullRadixCache(UnifiedRadixTreeTestMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
 
 if __name__ == "__main__":

test/registered/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_glm52.py

@@ -9,13 +9,13 @@ import shutil
 import tempfile
 import unittest
 
-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 AccuracyTwoPassMixin
 from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 GLM5_MODEL = "zai-org/GLM-5.2-FP8"
@@ -76,7 +76,7 @@ class TestGLM5UnifiedRadixCacheL3Accuracy(AccuracyTwoPassMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        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/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_hybrid_bitexact.py

@@ -42,7 +42,6 @@ import os
 import random
 import unittest
 
-from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kl_multiturn_utils import (
     make_mamba_decode_assert,
@@ -67,6 +66,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=1150, stage="base-b", runner_config="1-gpu-large")
@@ -158,7 +158,7 @@ class TestUnifiedHybridBitExact(CustomTestCase):
     @classmethod
     def tearDownClass(cls):
         if getattr(cls, "process", None) is not None:
-            kill_process_tree(cls.process.pid)
+            terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
     def _run(self, helper):
         helper(
@@ -270,7 +270,7 @@ class TestUnifiedHybridHiCacheBitExact(CustomTestCase):
     @classmethod
     def tearDownClass(cls):
         if getattr(cls, "process", None) is not None:
-            kill_process_tree(cls.process.pid)
+            terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
     def test_multiturn_decode_cache_hit_branching(self):
         groups, branches = 3, 3
@@ -350,7 +350,7 @@ class TestUnifiedHybridMTPBitExact(CustomTestCase):
     @classmethod
     def tearDownClass(cls):
         if getattr(cls, "process", None) is not None:
-            kill_process_tree(cls.process.pid)
+            terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
     def _run(self, helper):
         helper(

test/registered/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_mimo.py

@@ -3,7 +3,6 @@
 import random
 import unittest
 
-from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.kl_test_utils import (
     _extract_output_logprobs,
@@ -16,6 +15,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 MIMO_MODEL = "XiaomiMiMo/MiMo-V2.5"
@@ -97,7 +97,7 @@ class TestUnifiedMiMoHiCacheLoadBackKL(CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
     @classmethod
     def _prompt(cls, seed: int) -> list[int]:

test/registered/radix_cache/unified_radix_tree/test_unified_radix_cache_kl_swa.py

@@ -1,6 +1,5 @@
 import unittest
 
-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 get_input_ids
@@ -9,6 +8,7 @@ from sglang.test.test_utils import (
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
     popen_launch_server,
+    terminate_and_kill_process_tree,
 )
 
 register_cuda_ci(est_time=250, stage="base-b", runner_config="2-gpu-large")
@@ -43,7 +43,7 @@ class TestUnifiedSWARadixCache(UnifiedRadixTreeTestMixin, CustomTestCase):
 
     @classmethod
     def tearDownClass(cls):
-        kill_process_tree(cls.process.pid)
+        terminate_and_kill_process_tree(cls.process, wait_timeout=60)
 
 
 if __name__ == "__main__":