add engine id

vllm/attention/layer.py

@@ -398,7 +398,6 @@ def maybe_save_kv_layer_to_connector(
     assert isinstance(attn_metadata, dict)
     connector.save_kv_layer(layer_name, kv_cache_layer,
                             attn_metadata[layer_name])
-    
     get_lmcache_connector().save_kv_layer(layer_name, kv_cache_layer,
                             attn_metadata[layer_name])
 

vllm/distributed/kv_transfer/kv_transfer_state.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 from typing import TYPE_CHECKING, Optional
-
+import copy
 from vllm import envs
 from vllm.config import get_current_vllm_config
 from vllm.distributed.kv_transfer.kv_connector.base import KVConnectorBaseType
@@ -64,6 +64,15 @@ def ensure_kv_transfer_initialized(vllm_config: "VllmConfig") -> None:
     global _KV_CONNECTOR_AGENT
     global _KV_LMCACHE_CONNECTOR_AGENT
 
+    if _KV_LMCACHE_CONNECTOR_AGENT is None:
+        lmcache_config = copy.deepcopy(vllm_config)
+        from vllm.config import KVTransferConfig
+        lmcache_config.kv_transfer_config = KVTransferConfig(kv_connector="LMCacheConnectorV1", kv_role="kv_both")
+        lmcache_config.kv_transfer_config.engine_id = "ed9e943a-e455-4ed6-b88c-09ae6263f0c9"
+        lmcache_connector: LMCacheConnectorV1 = LMCacheConnectorV1(lmcache_config, role=KVConnectorRole.WORKER)
+
+        _KV_LMCACHE_CONNECTOR_AGENT = lmcache_connector
+
     if vllm_config.kv_transfer_config is None:
         return
 
@@ -72,13 +81,6 @@ def ensure_kv_transfer_initialized(vllm_config: "VllmConfig") -> None:
         if envs.VLLM_USE_V1:
             _KV_CONNECTOR_AGENT = KVConnectorFactory.create_connector_v1(
                 config=vllm_config, role=KVConnectorRole.WORKER)
-            
-            lmcache_config = vllm_config
-            lmcache_config.kv_transfer_config.kv_role = "kv_both"
-            lmcache_connector: LMCacheConnectorV1 = LMCacheConnectorV1(
-            lmcache_config, role=KVConnectorRole.WORKER)
-
-            _KV_LMCACHE_CONNECTOR_AGENT = lmcache_connector
         else:
             _KV_CONNECTOR_AGENT = KVConnectorFactory.create_connector_v0(
                 rank=get_world_group().rank,

vllm/v1/core/sched/scheduler.py

@@ -5,11 +5,12 @@ from __future__ import annotations
 
 import itertools
 import time
+import copy
 from collections import defaultdict
 from collections.abc import Iterable
 from typing import Any, Optional, Union
 
-from vllm.config import VllmConfig
+from vllm.config import KVTransferConfig,VllmConfig
 from vllm.distributed.kv_events import EventPublisherFactory, KVEventBatch
 from vllm.distributed.kv_transfer.kv_connector.factory import (
     KVConnectorFactory)
@@ -87,8 +88,9 @@ class Scheduler(SchedulerInterface):
             self.connector = KVConnectorFactory.create_connector_v1(
                 config=self.vllm_config, role=KVConnectorRole.SCHEDULER)
 
-        lmcache_config = self.vllm_config
-        lmcache_config.kv_transfer_config.kv_role = "kv_both"
+        lmcache_config = copy.deepcopy(self.vllm_config)
+        lmcache_config.kv_transfer_config = KVTransferConfig(kv_connector="LMCacheConnectorV1", kv_role="kv_both")
+        lmcache_config.kv_transfer_config.engine_id = "ed9e943a-e455-4ed6-b88c-09ae6263f0c9"
         self.lmcache_connector: LMCacheConnectorV1 = LMCacheConnectorV1(
             lmcache_config, role=KVConnectorRole.SCHEDULER)
 
@@ -400,7 +402,6 @@ class Scheduler(SchedulerInterface):
                             self.lmcache_connector.get_num_new_matched_tokens(
                                 request, num_new_local_computed_tokens))
 
-
                     # Total computed tokens (local + external).
                     num_computed_tokens = (num_new_local_computed_tokens +
                                            num_external_computed_tokens)
@@ -698,7 +699,7 @@ class Scheduler(SchedulerInterface):
                     num_external_computed_tokens, load_kv_async = (
                         self.connector.get_num_new_matched_tokens(
                             request, num_new_local_computed_tokens))
-                    
+
                 if self.lmcache_connector is not None:
                     num_external_computed_tokens, load_kv_async = (
                         self.lmcache_connector.get_num_new_matched_tokens(

vllm/v1/engine/core.py

@@ -864,7 +864,7 @@ class DPEngineCoreProc(EngineCoreProc):
             vllm_config.kv_transfer_config.engine_id = (
                 f"{vllm_config.kv_transfer_config.engine_id}_dp{local_dp_rank}"
             )
-            logger.debug("Setting kv_transfer_config.engine_id to %s",
+            logger.info("Setting kv_transfer_config.engine_id to %s",
                          vllm_config.kv_transfer_config.engine_id)
 
         from vllm.platforms import current_platform