Skip to content

[Refactor] Organize token processor metrics/traces code #7762

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
liyonghua0910 wants to merge 2 commits into
base: develop
Choose a base branch
from
+180 −157
Draft

[Refactor] Organize token processor metrics/traces code #7762

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
cc79d79
[Refactor] Organize token processor metrics/traces code
liyonghua0910 May 9, 2026
fe437d2
[feat] add env var to disable observability
liyonghua0910 May 11, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions fastdeploy/engine/request.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1086,8 +1086,8 @@ def to_dict(self):
"""
return {k: v for k, v in asdict(self).items()}

def record_recv_first_token(self):
cur_time = time.time()
def record_recv_first_token(self, cur_time: float = None):
cur_time = time.time() if cur_time is None else cur_time
self.record_recv_token(cur_time)
self.engine_recv_first_token_time = cur_time

Expand All @@ -1099,8 +1099,8 @@ def record_recv_token(self, cur_time: float = None):
if self.inference_start_time:
self.model_forward_time = cur_time - self.inference_start_time

def record_decode_recv_second_token(self):
cur_time = time.time()
def record_decode_recv_second_token(self, cur_time: float = None):
cur_time = time.time() if cur_time is None else cur_time
self.record_recv_token(cur_time)
self.decode_recv_second_token_time = cur_time

Expand Down
14 changes: 8 additions & 6 deletions fastdeploy/entrypoints/openai/serving_chat.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -325,10 +325,12 @@ async def chat_completion_stream_generator(
raise ValueError("{}".format(res["error_msg"]))

if inference_start_time[idx] == 0:
arrival_time = res["metrics"]["first_token_time"]
inference_start_time[idx] = res["metrics"]["inference_start_time"]
arrival_time = res["metrics"]["first_token_time"] or 0
inference_start_time[idx] = res["metrics"]["inference_start_time"] or 0
else:
arrival_time = res["metrics"]["engine_recv_latest_token_time"] - inference_start_time[idx]
arrival_time = (res["metrics"]["engine_recv_latest_token_time"] or 0) - inference_start_time[
idx
]
if first_iteration:
num_prompt_tokens = len(prompt_token_ids)
num_cached_tokens = res.get("num_cached_tokens", 0)
Expand Down Expand Up @@ -468,7 +470,7 @@ async def chat_completion_stream_generator(
trace_carrier = res.get("trace_carrier")
if trace_carrier:
tracing.trace_set_proc_propagate_context(request_id, trace_carrier)
start_time = res["metrics"]["engine_recv_latest_token_time"]
start_time = res["metrics"]["engine_recv_latest_token_time"] or 0
tracing.trace_report_span(
tracing.TraceSpanName.POSTPROCESSING,
request_id,
Expand All @@ -480,7 +482,7 @@ async def chat_completion_stream_generator(
del res["trace_carrier"]
num_choices -= 1
main_process_metrics.e2e_request_latency.observe(
time.time() - res["metrics"]["request_start_time"]
time.time() - (res["metrics"]["request_start_time"] or 0)
)
if previous_num_tokens[idx] != max_tokens:
choice.finish_reason = "stop"
Expand Down Expand Up @@ -715,7 +717,7 @@ async def chat_completion_full_generator(
trace_carrier = data.get("trace_carrier")
if trace_carrier:
tracing.trace_set_proc_propagate_context(request_id, trace_carrier)
start_time = data["metrics"]["engine_recv_latest_token_time"]
start_time = data["metrics"]["engine_recv_latest_token_time"] or 0
tracing.trace_report_span(
tracing.TraceSpanName.POSTPROCESSING,
request_id,
Expand Down
12 changes: 7 additions & 5 deletions fastdeploy/entrypoints/openai/serving_completion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -369,7 +369,7 @@ async def completion_full_generator(
trace_carrier = data.get("trace_carrier")
if trace_carrier:
tracing.trace_set_proc_propagate_context(request_id, trace_carrier)
start_time = data["metrics"]["engine_recv_latest_token_time"]
start_time = data["metrics"]["engine_recv_latest_token_time"] or 0
tracing.trace_report_span(
tracing.TraceSpanName.POSTPROCESSING,
request_id,
Expand Down Expand Up @@ -561,10 +561,12 @@ async def completion_stream_generator(

await self._call_process_response_dict(res, request, stream=True)
if inference_start_time[idx] == 0:
arrival_time = res["metrics"]["first_token_time"]
inference_start_time[idx] = res["metrics"]["inference_start_time"]
arrival_time = res["metrics"]["first_token_time"] or 0
inference_start_time[idx] = res["metrics"]["inference_start_time"] or 0
else:
arrival_time = res["metrics"]["engine_recv_latest_token_time"] - inference_start_time[idx]
arrival_time = (res["metrics"]["engine_recv_latest_token_time"] or 0) - inference_start_time[
idx
]

await self._process_echo_logic(request, idx, res["outputs"])
output = res["outputs"]
Expand Down Expand Up @@ -655,7 +657,7 @@ async def completion_stream_generator(
trace_carrier = res.get("trace_carrier")
if trace_carrier:
tracing.trace_set_proc_propagate_context(request_id, trace_carrier)
start_time = res["metrics"]["engine_recv_latest_token_time"]
start_time = res["metrics"]["engine_recv_latest_token_time"] or 0
tracing.trace_report_span(
tracing.TraceSpanName.POSTPROCESSING,
request_id,
Expand Down
1 change: 1 addition & 0 deletions fastdeploy/envs.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,6 +207,7 @@ def _validate_split_kv_size(value: int) -> int:
"FD_OTLP_EXPORTER_SCHEDULE_DELAY_MILLIS": lambda: int(os.getenv("FD_OTLP_EXPORTER_SCHEDULE_DELAY_MILLIS", "500")),
"FD_OTLP_EXPORTER_MAX_EXPORT_BATCH_SIZE": lambda: int(os.getenv("FD_OTLP_EXPORTER_MAX_EXPORT_BATCH_SIZE", "64")),
"FD_TOKEN_PROCESSOR_HEALTH_TIMEOUT": lambda: float(os.getenv("FD_TOKEN_PROCESSOR_HEALTH_TIMEOUT", "120")),
"FD_ENABLE_OBSERVABILITY": lambda: os.getenv("FD_ENABLE_OBSERVABILITY", "1") == "1",
"FD_XPU_MOE_FFN_QUANT_TYPE_MAP": lambda: os.getenv("FD_XPU_MOE_FFN_QUANT_TYPE_MAP", ""),
# Whether to enable low latency in mixed scenario
"FD_XPU_ENABLE_MIXED_EP_MODE": lambda: bool(int(os.getenv("FD_XPU_ENABLE_MIXED_EP_MODE", "0"))),
Expand Down
Loading
Loading