[1/2][lmcache connector] clean up lmcache multi-process adapter (#31838)

Signed-off-by: ApostaC <yihua98@uchicago.edu>

vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/multi_process_adapter.py

@@ -95,6 +95,10 @@ class LMCacheMPSchedulerAdapter:
             kv_rank: The kv rank used for LMCache keys
             vllm_block_size: The block size used in vLLM
         """
+        logger.warning(
+            "Importing LMCacheMPSchedulerAdapter is deprecated. "
+            "Please update your LMCache to the latest version."
+        )
         self.mq_client = MessageQueueClient(server_url, context)
 
         # Request futures
@@ -147,7 +151,7 @@ class LMCacheMPSchedulerAdapter:
         """
         return self.blocks_in_chunk
 
-    def _cleanup_lookup_result(self, request_id: str) -> None:
+    def cleanup_lookup_result(self, request_id: str) -> None:
         """
         Clean up lookup future for a finished request to prevent memory leak.
         Args:
@@ -176,6 +180,10 @@ class LMCacheMPWorkerAdapter:
         kv_rank: int,
         vllm_block_size: int,
     ):
+        logger.warning(
+            "Importing LMCacheMPWorkerAdapter is deprecated. "
+            "Please update your LMCache to the latest version."
+        )
         self.mq_client = MessageQueueClient(server_url, context)
 
         # Instance id for GPU worker

vllm/distributed/kv_transfer/kv_connector/v1/lmcache_mp_connector.py

@@ -17,15 +17,23 @@ from vllm.distributed.kv_transfer.kv_connector.v1.base import (
     KVConnectorMetadata,
     KVConnectorRole,
 )
+from vllm.v1.core.sched.output import SchedulerOutput
+from vllm.v1.outputs import KVConnectorOutput
+from vllm.v1.request import RequestStatus
+from vllm.v1.utils import ConstantList
+
+try:
+    from lmcache.integration.vllm.vllm_multi_process_adapter import (
+        LMCacheMPSchedulerAdapter,
+        LMCacheMPWorkerAdapter,
+        LoadStoreOp,
+    )
+except ImportError:
     from vllm.distributed.kv_transfer.kv_connector.v1.lmcache_integration import (
         LMCacheMPSchedulerAdapter,
         LMCacheMPWorkerAdapter,
         LoadStoreOp,
     )
-from vllm.v1.core.sched.output import SchedulerOutput
-from vllm.v1.outputs import KVConnectorOutput
-from vllm.v1.request import RequestStatus
-from vllm.v1.utils import ConstantList
 
 if TYPE_CHECKING:
     from vllm.config import VllmConfig
@@ -702,7 +710,7 @@ class LMCacheMPConnector(KVConnectorBase_V1):
                 else LMCacheMPRequestState.READY
             )
             # Clean up lookup future in scheduler adapter
-            self.scheduler_adapter._cleanup_lookup_result(request.request_id)
+            self.scheduler_adapter.cleanup_lookup_result(request.request_id)
 
     def build_connector_meta(
         self, scheduler_output: SchedulerOutput