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

[Feature] Sglang Tracing: Fine-Grained Tracking for Request Latency - Part 2 (#10804)

Browse Source 
Signed-off-by: Feng Su <sufeng@linux.alibaba.com>
This commit is contained in:
Feng Su
2025-10-28 16:25:46 +08:00
committed by GitHub
parent b1e13e7cea
commit ea96106000
19 changed files with 824 additions and 60 deletions
Download Patch File Download Diff File Expand all files Collapse all files

33
python/sglang/srt/managers/data_parallel_controller.py
View File

@@ -34,13 +34,21 @@ from sglang.srt.managers.io_struct import (
TokenizedGenerateReqInput,
WatchLoadUpdateReq,
)
from sglang.srt.managers.schedule_batch import Req
from sglang.srt.managers.schedule_batch import Req, RequestStage
from sglang.srt.managers.scheduler import run_scheduler_process
from sglang.srt.server_args import (
DP_ATTENTION_HANDSHAKE_PORT_DELTA,
PortArgs,
ServerArgs,
)
from sglang.srt.tracing.trace import (
process_tracing_init,
trace_get_proc_propagate_context,
trace_set_proc_propagate_context,
trace_set_thread_info,
trace_slice_end,
trace_slice_start,
)
from sglang.srt.utils import (
bind_port,
configure_logger,
@@ -170,11 +178,22 @@ class DataParallelController:
def handle_load_update_req(self, obj):
self.dp_budget.update_budget(obj)
def dispatching_with_trace(self, req: Req):
if self.server_args.enable_trace:
trace_set_proc_propagate_context(req.rid, req.trace_context)
trace_slice_start(RequestStage.DC_DISPATCH, req.rid)
req.trace_context = trace_get_proc_propagate_context(req.rid)
self.dispatching(req)
if self.server_args.enable_trace:
trace_slice_end(RequestStage.DC_DISPATCH, req.rid, thread_finish_flag=True)
def init_dispatcher(self):
self._request_dispatcher = TypeBasedDispatcher(
[
(TokenizedGenerateReqInput, self.dispatching),
(TokenizedEmbeddingReqInput, self.dispatching),
(TokenizedGenerateReqInput, self.dispatching_with_trace),
(TokenizedEmbeddingReqInput, self.dispatching_with_trace),
(BlockReqInput, self.send_to_all_workers),
(WatchLoadUpdateReq, self.handle_load_update_req),
]
@@ -487,6 +506,14 @@ def run_data_parallel_controller_process(
pipe_writer,
):
kill_itself_when_parent_died()
if server_args.enable_trace:
process_tracing_init(server_args.otlp_traces_endpoint, "sglang")
thread_label = "DP Controller"
if server_args.disaggregation_mode == "prefill":
thread_label = "Prefill DP Controller"
elif server_args.disaggregation_mode == "decode":
thread_label = "Decode DP Controller"
trace_set_thread_info(thread_label)
setproctitle.setproctitle("sglang::data_parallel_controller")
faulthandler.enable()
configure_logger(server_args)