feat: support Kimi Linear PD disaggregation with DCP (#32837)

Co-authored-by: Yangmin Li <yangminl@nvidia.com>

test/registered/unit/server_args/test_server_args.py

@@ -8,6 +8,7 @@ from types import SimpleNamespace
 from unittest.mock import MagicMock, patch
 
 import sglang.srt.server_args as server_args_module
+from sglang.srt.arg_groups import pd_disaggregation_hook
 from sglang.srt.arg_groups.speculative_hook import handle_speculative_decoding
 from sglang.srt.entrypoints.sidecar import (
     SGLANG_GRPC_ENDPOINT_ENV,
@@ -219,6 +220,56 @@ class TestLoadBalanceMethod(unittest.TestCase):
         server_args = self._load_balance_args(disaggregation_mode="decode")
         self.assertEqual(server_args.load_balance_method, "round_robin")
 
+    def test_pd_prefill_dcp_warns_about_performance(self):
+        server_args = ServerArgs(
+            model_path="dummy",
+            disaggregation_mode="prefill",
+            dcp_size=4,
+        )
+        with self.assertLogs(pd_disaggregation_hook.logger, level="WARNING") as logs:
+            server_args._handle_pd_disaggregation()
+        self.assertIn("without improving prefill performance", "\n".join(logs.output))
+
+    def test_pd_decode_dcp_forces_chunk_cache(self):
+        server_args = self._load_balance_args(
+            disaggregation_mode="decode",
+            disaggregation_transfer_backend="mooncake",
+            dcp_size=4,
+        )
+        self.assertTrue(server_args.disable_radix_cache)
+
+    def test_pd_decode_dcp_rejects_unsupported_transfer_backend(self):
+        server_args = ServerArgs(
+            model_path="dummy",
+            disaggregation_mode="decode",
+            disaggregation_transfer_backend="fake",
+            dcp_size=4,
+        )
+        with self.assertRaisesRegex(ValueError, "mooncake or nixl"):
+            server_args._handle_pd_disaggregation()
+
+    def test_pd_decode_dcp_rejects_radix_cache(self):
+        server_args = ServerArgs(
+            model_path="dummy",
+            disaggregation_mode="decode",
+            disaggregation_transfer_backend="nixl",
+            disaggregation_decode_enable_radix_cache=True,
+            dcp_size=4,
+        )
+        with self.assertRaisesRegex(ValueError, "currently requires chunk cache"):
+            server_args._handle_pd_disaggregation()
+
+    def test_pd_decode_dcp_rejects_hierarchical_cache(self):
+        server_args = ServerArgs(
+            model_path="dummy",
+            disaggregation_mode="decode",
+            disaggregation_transfer_backend="nixl",
+            enable_hierarchical_cache=True,
+            dcp_size=4,
+        )
+        with self.assertRaisesRegex(ValueError, "--enable-hierarchical-cache"):
+            server_args._handle_pd_disaggregation()
+
     def test_pd_decode_radix_cache_rejects_hisparse(self):
         server_args = ServerArgs(
             model_path="dummy",