apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[BugFix] Add safety checks in recycle_gpu_blocks to prevent block allocation errors (#6531)

Browse Source 
* [BugFix] Add safety checks in recycle_gpu_blocks to prevent block allocation errors

- Check prefix tree status before recycling GPU blocks
- Validate gpu_block_ids is a list
- Add overflow check to prevent free block count exceeding total blocks

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* [BugFix] Fix AttributeError in recycle_gpu_blocks when prefix_tree_status_signal not initialized

- Add hasattr check before accessing prefix_tree_status_signal
- The signal is only initialized in launch_cache_messager, not in __init__
- Fixes CI test failure in test_prefix_cache_manager.py

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* [BugFix] Reset prefix cache when model weights are updating

- Call self.reset() before setting status to NORMAL in UPDATING state
- Ensure cache consistency when model weights change
- Consistent with CLEARING state handling

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
kevin
2026-03-02 13:12:29 +08:00
committed by GitHub
parent 481d0e385f
commit ecfd088a03
1 changed files with 19 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/cache_manager/prefix_cache_manager.py
+19
View File
@@ -498,6 +498,24 @@ class PrefixCacheManager:
"""
recycle gpu blocks.
"""
if (
hasattr(self, "prefix_tree_status_signal")
and self.prefix_tree_status_signal.value[0] != PrefixTreeStatus.NORMAL
):
# Prefix Tree Clearing, skip recycle gpu blocks
logger.warning("Prefix tree is not normal, skip recycle gpu blocks")
return
if not isinstance(gpu_block_ids, list):
gpu_block_ids = [gpu_block_ids]
if len(self.gpu_free_block_list) + len(gpu_block_ids) > self.num_gpu_blocks:
# The block allocation and recycling are abnormal, and the test results are not convincing
logger.error(
f"The number of free gpu blocks {len(self.gpu_free_block_list)} plus the number of recycled "
f"gpu blocks {len(gpu_block_ids)} exceeds the total number of gpu blocks {self.num_gpu_blocks} \n"
f"this indicates a block allocation and deallocation error, recycled blocks will be discarded {gpu_block_ids}"
)
return
logger.info(
f"req_id:{req_id} recycle_gpu_blocks: {gpu_block_ids}, len(self.gpu_free_block_list) {len(self.gpu_free_block_list)}"
)
@@ -2143,6 +2161,7 @@ class PrefixCacheManager:
prefix_tree_status_signal.value[0] = PrefixTreeStatus.CLEARED
logger.info("Prefix cache tree is cleared.")
if prefix_tree_status_signal.value[0] == PrefixTreeStatus.UPDATING:
self.reset()
prefix_tree_status_signal.value[0] = PrefixTreeStatus.NORMAL
logger.info("Prefix cache tree is updated.")
time.sleep(0.01)