apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-12-24 13:28:13 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[fix] fix requests & block metrics (#4404)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details
Deploy GitHub Pages / deploy (push) Has been cancelled
Details

Browse Source 
* [fix] fix requests & block metrics

* [chore] rename variables
This commit is contained in:
李泳桦
2025-10-15 11:49:24 +08:00
committed by GitHub
parent abb62624b8
commit ffe7af8a97
5 changed files with 20 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
fastdeploy/cache_manager/prefix_cache_manager.py
View File

@@ -115,6 +115,7 @@ class PrefixCacheManager:
main_process_metrics.max_gpu_block_num.set(self.num_gpu_blocks)
main_process_metrics.available_gpu_block_num.set(self.num_gpu_blocks)
main_process_metrics.free_gpu_block_num.set(self.num_gpu_blocks)
main_process_metrics.available_gpu_resource.set(1.0)
@property
@@ -358,6 +359,7 @@ class PrefixCacheManager:
main_process_metrics.max_gpu_block_num.set(self.num_gpu_blocks)
main_process_metrics.available_gpu_block_num.set(self.num_gpu_blocks)
main_process_metrics.free_gpu_block_num.set(self.num_gpu_blocks)
main_process_metrics.available_gpu_resource.set(1.0)
def can_allocate_gpu_blocks(self, num_blocks: int):

4
fastdeploy/engine/resource_manager.py
View File

@@ -311,8 +311,8 @@ class ResourceManager:
break
# record batch size here
task_used_block_num = sum([len(task.block_tables) if task else 0 for task in self.tasks_list])
main_process_metrics.available_gpu_block_num.set(self.total_block_number() - task_used_block_num)
num_blocks_used_by_tasks = sum([len(task.block_tables) if task else 0 for task in self.tasks_list])
main_process_metrics.available_gpu_block_num.set(self.total_block_number() - num_blocks_used_by_tasks)
main_process_metrics.batch_size.set(self.max_num_seqs - self.available_batch())
main_process_metrics.gpu_cache_usage_perc.set(self.get_gpu_cache_usage_perc())
llm_logger.info(

16
fastdeploy/engine/sched/resource_manager_v1.py
View File

@@ -630,8 +630,6 @@ class ResourceManagerV1(ResourceManager):
request, self.config.cache_config.block_size, request.num_computed_tokens
)
request.status = RequestStatus.RUNNING
main_process_metrics.num_requests_waiting.dec(1)
main_process_metrics.num_requests_running.inc(1)
else:
if self.config.cache_config.enable_prefix_caching:
self._free_blocks(request)
@@ -640,11 +638,17 @@ class ResourceManagerV1(ResourceManager):
llm_logger.error("Unknown request status type")
if scheduled_reqs:
task_used_block_num = sum([len(task.block_tables) if task else 0 for task in self.tasks_list])
main_process_metrics.available_gpu_block_num.set(self.total_block_number() - task_used_block_num)
main_process_metrics.batch_size.set(self.max_num_seqs - self.available_batch())
main_process_metrics.gpu_cache_usage_perc.set(self.get_gpu_cache_usage_perc())
llm_logger.debug(f"schedued_reqs: {scheduled_reqs}")
# Update metrics
num_tasks = sum([1 if task else 0 for task in self.tasks_list])
num_blocks_used_by_tasks = sum([len(task.block_tables) if task else 0 for task in self.tasks_list])
main_process_metrics.available_gpu_block_num.set(self.total_block_number() - num_blocks_used_by_tasks)
main_process_metrics.batch_size.set(self.max_num_seqs - self.available_batch())
main_process_metrics.gpu_cache_usage_perc.set(self.get_gpu_cache_usage_perc())
main_process_metrics.num_requests_running.set(len(self.running))
main_process_metrics.num_requests_waiting.set(num_tasks - len(self.running))
return scheduled_reqs
def get_available_position(self) -> int:

2
fastdeploy/metrics/metrics.py
View File

@@ -316,7 +316,7 @@ class MetricsManager:
"available_gpu_block_num": {
"type": Gauge,
"name": "fastdeploy:available_gpu_block_num",
"description": "Number of available gpu blocks in cache, including prefix caching blocks that are not officially released",
"description": "Number of available gpu blocks in cache, including blocks in LRU list",
"kwargs": {},
},
"free_gpu_block_num": {

7
fastdeploy/output/token_processor.py
View File

@@ -421,9 +421,12 @@ class TokenProcessor:
if task_id in self.resource_manager.req_dict:
del self.resource_manager.req_dict[task_id]
task_used_block_num = sum([len(task.block_tables) if task else 0 for task in self.resource_manager.tasks_list])
# Update block metrics
num_blocks_used_by_tasks = sum(
[len(task.block_tables) if task else 0 for task in self.resource_manager.tasks_list]
)
main_process_metrics.available_gpu_block_num.set(
self.resource_manager.total_block_number() - task_used_block_num
self.resource_manager.total_block_number() - num_blocks_used_by_tasks
)
main_process_metrics.batch_size.set(
self.resource_manager.max_num_seqs - self.resource_manager.available_batch()