chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

check_offload_progress more frequently (#11656)

Browse Source
This commit is contained in:
pansicheng
2025-10-28 00:37:38 +08:00
committed by GitHub
parent a8b91f6b2d
commit 0cae873fcd
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/disaggregation/decode.py
View File

@@ -1008,6 +1008,9 @@ class SchedulerDisaggregationDecodeMixin:
return new_batch
def process_decode_queue(self: Scheduler):
if self.server_args.disaggregation_decode_enable_offload_kvcache:
self.decode_offload_manager.check_offload_progress()
# try to resume retracted requests if there are enough space for another `num_reserved_decode_tokens` decode steps
resumed_reqs = self.disagg_decode_prealloc_queue.resume_retracted_reqs()
self.waiting_queue.extend(resumed_reqs)
@@ -1030,6 +1033,3 @@ class SchedulerDisaggregationDecodeMixin:
self.disagg_decode_transfer_queue.pop_transferred()
) # the requests which kv has arrived
self.waiting_queue.extend(alloc_reqs)
if self.server_args.disaggregation_decode_enable_offload_kvcache:
self.decode_offload_manager.check_offload_progress()