feat: support custom OTLP trace service name (#35802)

python/sglang/multimodal_gen/runtime/server_args/server_args.py

@@ -589,6 +589,7 @@ class ServerArgs(DisaggServerArgsMixin):
     # Tracing
     enable_trace: bool = False
     otlp_traces_endpoint: str = "localhost:4317"
+    otlp_service_name: str | None = None
 
     # SGLang backend for encoder stage
     srt_encoder_url: str | None = None
@@ -2893,6 +2894,13 @@ class ServerArgs(DisaggServerArgsMixin):
             default=ServerArgs.otlp_traces_endpoint,
             help="OTLP collector endpoint when --enable-trace is set. Format: <host>:<port>",
         )
+        parser.add_argument(
+            "--otlp-service-name",
+            type=str,
+            default=ServerArgs.otlp_service_name,
+            help="Service name for OTLP traces (displayed as 'service.name' in trace backends). "
+            "If unset, falls back to the OTEL_SERVICE_NAME env var, then to 'sglang-diffusion'.",
+        )
         parser.add_argument(
             "--log-requests",
             action="store_true",

python/sglang/multimodal_gen/runtime/utils/trace_wrapper.py

@@ -7,6 +7,7 @@ start/end bookkeeping.
 
 from __future__ import annotations
 
+import os
 from contextlib import contextmanager
 from dataclasses import dataclass
 
@@ -37,10 +38,17 @@ def init_diffusion_tracing(server_args, thread_label: str):
         trace_set_thread_info,
     )
 
+    # Priority: --otlp-service-name > OTEL_SERVICE_NAME > "sglang-diffusion"
+    service_name = (
+        server_args.otlp_service_name
+        or os.getenv("OTEL_SERVICE_NAME")
+        or "sglang-diffusion"
+    )
+
     # srt owns TraceReqContext and filters spans through its trace_modules list
     process_tracing_init(
         server_args.otlp_traces_endpoint,
-        "sglang-diffusion",
+        service_name,
         trace_modules=DIFFUSION_TRACE_MODULE,
     )
     trace_set_thread_info(thread_label)

python/sglang/srt/arg_groups/fields/observability.py

@@ -177,6 +177,11 @@ class Observability(msgspec.Struct):
         str,
         "Config opentelemetry collector endpoint if --enable-trace is set. format: <ip>:<port>",
     ] = "localhost:4317"
+    otlp_service_name: A[
+        Optional[str],
+        "Service name for OTLP traces (displayed as 'service.name' in trace backends). "
+        "If unset, falls back to the OTEL_SERVICE_NAME env var, then to 'sglang'.",
+    ] = None
     # RequestMetricsExporter configuration
     export_metrics_to_file: A[
         bool,

python/sglang/srt/disaggregation/encoder/runtime.py

@@ -1775,7 +1775,7 @@ def launch_local_runtime(server_args: ServerArgs) -> EncoderRuntime:
     if get_observability().enable_trace:
         process_tracing_init(
             get_observability().otlp_traces_endpoint,
-            "sglang",
+            get_observability().otlp_service_name,
             trace_modules=get_observability().trace_modules,
         )
         trace_set_thread_info("Encoder")

python/sglang/srt/entrypoints/engine.py

@@ -329,7 +329,7 @@ class Engine(EngineScoreMixin, EngineBase):
         if get_observability().enable_trace:
             process_tracing_init(
                 get_observability().otlp_traces_endpoint,
-                "sglang",
+                get_observability().otlp_service_name,
                 trace_modules=get_observability().trace_modules,
             )
             thread_label = "Tokenizer"

python/sglang/srt/entrypoints/http_server.py

@@ -293,7 +293,7 @@ async def lifespan(fast_api_app: FastAPI):
     if get_observability().enable_trace:
         process_tracing_init(
             get_observability().otlp_traces_endpoint,
-            "sglang",
+            get_observability().otlp_service_name,
             trace_modules=get_observability().trace_modules,
         )
         if get_disagg().disaggregation_mode == "prefill":

python/sglang/srt/managers/data_parallel_controller.py

@@ -839,7 +839,7 @@ def run_data_parallel_controller_process(
     if get_observability().enable_trace:
         process_tracing_init(
             get_observability().otlp_traces_endpoint,
-            "sglang",
+            get_observability().otlp_service_name,
             trace_modules=get_observability().trace_modules,
         )
         thread_label = "DP Controller"

python/sglang/srt/managers/scheduler.py

@@ -5852,7 +5852,7 @@ def run_scheduler_process(
     if get_observability().enable_trace:
         process_tracing_init(
             get_observability().otlp_traces_endpoint,
-            "sglang",
+            get_observability().otlp_service_name,
             trace_modules=get_observability().trace_modules,
         )
         thread_label = "Scheduler"

python/sglang/srt/observability/trace.py

@@ -227,9 +227,12 @@ def process_tracing_init(
         )
 
     try:
+        # Priority: explicit server_name > OTEL_SERVICE_NAME > "sglang"
+        service_name = server_name or os.getenv("OTEL_SERVICE_NAME", "sglang")
+
         resource = Resource.create(
             attributes={
-                SERVICE_NAME: server_name,
+                SERVICE_NAME: service_name,
             }
         )
         tracer_provider = TracerProvider(
@@ -259,7 +262,7 @@ def process_tracing_init(
     if envs.SGLANG_TRACE_ASYNC.get():
         from sglang.srt.observability.trace_async import start_trace_exporter
 
-        start_trace_exporter(otlp_endpoint, server_name, trace_modules=trace_modules)
+        start_trace_exporter(otlp_endpoint, service_name, trace_modules=trace_modules)
 
 
 def get_global_tracing_enabled():