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

[Log] Add trace log and add loggingInstrumentor tool (#4692)

Browse Source 
* add trace logger and trace print

* trigger ci

* fix unittest

* translate notes and add copyright

---------

Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com>
Co-authored-by: YuBaoku <49938469+EmmonsCurse@users.noreply.github.com>
This commit is contained in:
qwes5s5
2025-11-17 11:08:57 +08:00
committed by GitHub
parent 5444af6ff6
commit 36216e62f0
21 changed files with 941 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
fastdeploy/engine/common_engine.py
View File

@@ -49,6 +49,8 @@ from fastdeploy.plugins.token_processor import load_token_processor_plugins
from fastdeploy.router.utils import check_service_health
from fastdeploy.splitwise.internal_adapter_utils import InternalAdapter
from fastdeploy.splitwise.splitwise_connector import SplitwiseConnector
from fastdeploy.trace.constants import LoggingEventName
from fastdeploy.trace.trace_logger import print as trace_print
from fastdeploy.utils import (
EngineError,
check_download_links,
@@ -364,7 +366,7 @@ class EngineService:
for item in tasks:
item.schedule_start_time = time.time()
trace_print(LoggingEventName.RESOURCE_ALLOCATE_START, item.request_id, getattr(item, "user", ""))
available_batch = np.sum(self.resource_manager.stop_flags)
if len(tasks) > available_batch:
self.llm_logger.error(f"Inserting batch:{len(tasks)} exceeds the available batch:{available_batch}.")
@@ -398,6 +400,9 @@ class EngineService:
self.llm_logger.info(f"Tasks are sent to engine, req_ids={req_ids}")
for task in tasks:
task.inference_start_time = time.time()
trace_print(LoggingEventName.RESOURCE_ALLOCATE_END, task.request_id, getattr(task, "user", ""))
trace_print(LoggingEventName.REQUEST_SCHEDULE_END, task.request_id, getattr(task, "user", ""))
trace_print(LoggingEventName.INFERENCE_START, task.request_id, getattr(task, "user", ""))
if not is_prefill:
if not self.cfg.model_config.enable_mm:
self.update_requests_chunk_size(tasks)
@@ -636,7 +641,8 @@ class EngineService:
max_num_batched_tokens=self.cfg.scheduler_config.max_num_batched_tokens,
batch=num_prefill_batch,
)
for task in tasks:
trace_print(LoggingEventName.REQUEST_QUEUE_END, task.request_id, getattr(task, "user", ""))
if len(tasks) == 0:
time.sleep(0.001)
continue
@@ -689,6 +695,8 @@ class EngineService:
max_num_batched_tokens=max_num_batched_tokens,
batch=num_prefill_batch,
)
for task in tasks:
trace_print(LoggingEventName.REQUEST_QUEUE_END, task.request_id, getattr(task, "user", ""))
if self.cfg.scheduler_config.splitwise_role == "decode":
# Decode will instert the request sent by prefill to engine,
@@ -761,6 +769,10 @@ class EngineService:
time.sleep(0.001)
# Fetch requests and add them to the scheduling queue
if tasks:
for task in tasks:
trace_print(
LoggingEventName.RESOURCE_ALLOCATE_START, task.request_id, getattr(task, "user", "")
)
if self.cfg.scheduler_config.splitwise_role == "prefill":
self.resource_manager.add_request_in_p(tasks)
else:
@@ -816,6 +828,10 @@ class EngineService:
]
)
self.resource_manager.get_real_bsz()
for task in tasks:
trace_print(LoggingEventName.RESOURCE_ALLOCATE_END, task.request_id, getattr(task, "user", ""))
trace_print(LoggingEventName.REQUEST_SCHEDULE_END, task.request_id, getattr(task, "user", ""))
trace_print(LoggingEventName.INFERENCE_START, task.request_id, getattr(task, "user", ""))
self.engine_worker_queue.put_tasks((tasks, self.resource_manager.real_bsz))
else:
time.sleep(0.005)
@@ -877,6 +893,10 @@ class EngineService:
request.llm_engine_recv_req_timestamp = time.time()
start_span("ENQUEUE_ZMQ", data, trace.SpanKind.PRODUCER)
main_process_metrics.requests_number.inc()
self.llm_logger.debug(f"Receive request: {request}")
trace_print(LoggingEventName.PREPROCESSING_END, data["request_id"], data.get("user", ""))
trace_print(LoggingEventName.REQUEST_SCHEDULE_START, data["request_id"], data.get("user", ""))
trace_print(LoggingEventName.REQUEST_QUEUE_START, data["request_id"], data.get("user", ""))
self.llm_logger.debug(f"Receive request from api server: {request}")
except Exception as e:
self.llm_logger.error(f"Receive request error: {e}, {traceback.format_exc()!s}")