chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 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 01:25:46 -07:00
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
python/sglang/srt/entrypoints/engine.py
+7 -4
View File
@@ -143,10 +143,13 @@ class Engine(EngineBase):
# Enable tracing
if server_args.enable_trace:
process_tracing_init(server_args.oltp_traces_endpoint, "sglang")
if server_args.disaggregation_mode == "null":
thread_label = "Tokenizer"
trace_set_thread_info(thread_label)
process_tracing_init(server_args.otlp_traces_endpoint, "sglang")
thread_label = "Tokenizer"
if server_args.disaggregation_mode == "prefill":
thread_label = "Prefill Tokenizer"
elif server_args.disaggregation_mode == "decode":
thread_label = "Decode Tokenizer"
trace_set_thread_info(thread_label)
try:
self.loop = asyncio.get_running_loop()
python/sglang/srt/entrypoints/http_server.py
+6 -3
View File
@@ -220,9 +220,12 @@ async def lifespan(fast_api_app: FastAPI):
# Init tracing
if server_args.enable_trace:
process_tracing_init(server_args.oltp_traces_endpoint, "sglang")
if server_args.disaggregation_mode == "null":
trace_set_thread_info(thread_label)
process_tracing_init(server_args.otlp_traces_endpoint, "sglang")
if server_args.disaggregation_mode == "prefill":
thread_label = "Prefill" + thread_label
elif server_args.disaggregation_mode == "decode":
thread_label = "Decode" + thread_label
trace_set_thread_info(thread_label)
# Initialize OpenAI serving handlers
fast_api_app.state.openai_serving_completion = OpenAIServingCompletion(