xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-18 05:35:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Core] Fix abrupt request abort (#18485)

Browse Source 
Signed-off-by: nicklucche <nlucches@redhat.com>
Signed-off-by: Nick Hill <nhill@redhat.com>

Co-authored-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nicolò Lucchesi 2025-06-07 01:27:59 +02:00 committed by GitHub
parent ca27f0f9c1
commit b6a3a9f76d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 9 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/core/kv_cache_coordinator.py
View File

@ -164,7 +164,7 @@ class KVCacheCoordinator(ABC):
Get the blocks for the request. Get the blocks for the request.
""" """
return [ return [
manager.req_to_blocks[request_id] manager.req_to_blocks.get(request_id) or []
for manager in self.single_type_managers for manager in self.single_type_managers
] ]

14
vllm/v1/core/sched/scheduler.py
View File

@ -76,6 +76,9 @@ class Scheduler(SchedulerInterface):
# KV Connector pushes/pull of remote KVs for P/D and offloading. # KV Connector pushes/pull of remote KVs for P/D and offloading.
self.connector = None self.connector = None
if self.vllm_config.kv_transfer_config is not None: if self.vllm_config.kv_transfer_config is not None:
assert len(self.kv_cache_config.kv_cache_groups) == 1, (
"Multiple KV cache groups are not currently supported "
"with KV connectors")
self.connector = KVConnectorFactory.create_connector_v1( self.connector = KVConnectorFactory.create_connector_v1(
config=self.vllm_config, role=KVConnectorRole.SCHEDULER) config=self.vllm_config, role=KVConnectorRole.SCHEDULER)
@ -985,9 +988,8 @@ class Scheduler(SchedulerInterface):
""" """
if self.connector is None: if self.connector is None:
return False, None return False, None
assert len(self.kv_cache_config.kv_cache_groups
) == 1, "KV connector only supports one KV cache group now" (block_ids, ) = self.kv_cache_manager.get_block_ids(request.request_id)
block_ids = self.kv_cache_manager.get_block_ids(request.request_id)[0]
return self.connector.request_finished(request, block_ids) return self.connector.request_finished(request, block_ids)
def _update_waiting_for_remote_kv(self, request: Request) -> bool: def _update_waiting_for_remote_kv(self, request: Request) -> bool:
@ -1002,12 +1004,12 @@ class Scheduler(SchedulerInterface):
and the request state will be moved back to WAITING from and the request state will be moved back to WAITING from
WAITING_FOR_REMOTE_KV. WAITING_FOR_REMOTE_KV.
""" """
assert self.connector is not None
if request.request_id not in self.finished_recving_kv_req_ids: if request.request_id not in self.finished_recving_kv_req_ids:
return False return False
assert len(self.kv_cache_config.kv_cache_groups
) == 1, "KV connector only supports one KV cache group now"
# Now that the blocks are ready, actually cache them. # Now that the blocks are ready, actually cache them.
block_ids = self.kv_cache_manager.get_block_ids(request.request_id)[0] (block_ids, ) = self.kv_cache_manager.get_block_ids(request.request_id)
num_computed_tokens = len(block_ids) * self.block_size num_computed_tokens = len(block_ids) * self.block_size
# Handle the case where num request tokens less then one block. # Handle the case where num request tokens less then one block.
num_computed_tokens = min(num_computed_tokens, request.num_tokens) num_computed_tokens = min(num_computed_tokens, request.num_tokens)