minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Auto Sync] Update scheduler_runtime_checker_mixin.py (20251219) (#15437)

Browse Source 
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Hanming Lu <69857889+hanming-lu@users.noreply.github.com>
This commit is contained in:
Lianmin Zheng
2025-12-19 01:46:03 -08:00
committed by GitHub
co-authored by github-actions[bot] Hanming Lu
parent 6559e43f30
commit 92e6b3c30e
1 changed files with 31 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/managers/scheduler_runtime_checker_mixin.py
+18 -4
View File
@@ -122,15 +122,24 @@ class SchedulerRuntimeCheckerMixin:
full_num_used != self.tree_cache.full_protected_size() full_num_used != self.tree_cache.full_protected_size()
or mamba_num_used != self.tree_cache.mamba_protected_size() or mamba_num_used != self.tree_cache.mamba_protected_size()
) )
if memory_leak:
free_full_pages = set( free_full_pages = set(
self.token_to_kv_pool_allocator.free_pages.tolist() self.token_to_kv_pool_allocator.free_pages.tolist()
+ self.token_to_kv_pool_allocator.release_pages.tolist() + self.token_to_kv_pool_allocator.release_pages.tolist()
) )
cached_full_pages = set(self.tree_cache.all_values_flatten().tolist()) cached_full_pages = set(self.tree_cache.all_values_flatten().tolist())
expected_full_pages = set(range(1, self.token_to_kv_pool_allocator.size + 1)) expected_full_pages = set(
leaked_full_pages = expected_full_pages - free_full_pages - cached_full_pages range(1, self.token_to_kv_pool_allocator.size + 1)
free_mamba_pages = set(self.req_to_token_pool.mamba_pool.free_slots.tolist()) )
cached_mamba_pages = set(self.tree_cache.all_mamba_values_flatten().tolist()) leaked_full_pages = (
expected_full_pages - free_full_pages - cached_full_pages
)
free_mamba_pages = set(
self.req_to_token_pool.mamba_pool.free_slots.tolist()
)
cached_mamba_pages = set(
self.tree_cache.all_mamba_values_flatten().tolist()
)
expected_mamba_pages = set(range(self.req_to_token_pool.mamba_pool.size)) expected_mamba_pages = set(range(self.req_to_token_pool.mamba_pool.size))
leaked_mamba_pages = ( leaked_mamba_pages = (
expected_mamba_pages - free_mamba_pages - cached_mamba_pages expected_mamba_pages - free_mamba_pages - cached_mamba_pages
@@ -139,6 +148,11 @@ class SchedulerRuntimeCheckerMixin:
f"{full_available_size=}, {full_evictable_size=}, {self.token_to_kv_pool_allocator.size=}, {self.tree_cache.full_protected_size()=}\n" f"{full_available_size=}, {full_evictable_size=}, {self.token_to_kv_pool_allocator.size=}, {self.tree_cache.full_protected_size()=}\n"
f"{mamba_available_size=}, {mamba_evictable_size=}, {self.req_to_token_pool.mamba_pool.size=}, {self.tree_cache.mamba_protected_size()=}, leaked_full_pages={leaked_full_pages if len(leaked_full_pages) > 0 else None}, leaked_mamba_pages={leaked_mamba_pages if len(leaked_mamba_pages) > 0 else None}\n" f"{mamba_available_size=}, {mamba_evictable_size=}, {self.req_to_token_pool.mamba_pool.size=}, {self.tree_cache.mamba_protected_size()=}, leaked_full_pages={leaked_full_pages if len(leaked_full_pages) > 0 else None}, leaked_mamba_pages={leaked_mamba_pages if len(leaked_mamba_pages) > 0 else None}\n"
) )
else:
token_msg = (
f"{full_available_size=}, {full_evictable_size=}, {self.token_to_kv_pool_allocator.size=}, {self.tree_cache.full_protected_size()=}\n"
f"{mamba_available_size=}, {mamba_evictable_size=}, {self.req_to_token_pool.mamba_pool.size=}, {self.tree_cache.mamba_protected_size()=}\n"
)
return memory_leak, token_msg return memory_leak, token_msg
def _check_radix_cache_memory(self: Scheduler): def _check_radix_cache_memory(self: Scheduler):