xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-13 12:04:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Fix handling of resumed reqs in SharedStorageConnector (#27719)

Browse Source 
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill 2025-10-29 13:16:52 -07:00 committed by GitHub
parent fcb1d570bb
commit d4aa144343
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 22 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/distributed/kv_transfer/kv_connector/v1/shared_storage_connector.py
View File

@ -336,16 +336,14 @@ class SharedStorageConnector(KVConnectorBase_V1):
cached_reqs = scheduler_output.scheduled_cached_reqs
for i, req_id in enumerate(cached_reqs.req_ids):
resumed_from_preemption = cached_reqs.resumed_from_preemption[i]
if not resumed_from_preemption or req_id not in self._requests_need_load:
continue
num_computed_tokens = cached_reqs.num_computed_tokens[i]
num_new_tokens = scheduler_output.num_scheduled_tokens[req_id]
new_block_ids = cached_reqs.new_block_ids[i]
resumed_from_preemption = cached_reqs.resumed_from_preemption[i]
# NOTE(rob): here we rely on the resumed requests being
# the first N requests in the list scheduled_cache_reqs.
if not resumed_from_preemption:
break
if req_id in self._requests_need_load:
# NOTE(rob): cached_req_data does not have the full
# list of token ids (only new tokens). So we look it
# up in the actual request object.