Wrap IPv6 addresses in gRPC, bench_serving, and log messages (#21236)

Co-authored-by: hnyls2002 <lsyincs@gmail.com>
Co-authored-by: Liangsheng Yin <hnyls2002@gmail.com>

python/sglang/bench_serving.py

@@ -44,6 +44,7 @@ from sglang.benchmark.utils import (
     remove_prefix,
     set_ulimit,
 )
+from sglang.srt.utils.network import NetworkAddress
 
 _ROUTING_KEY_HEADER = "X-SMG-Routing-Key"
 
@@ -1726,10 +1727,16 @@ def run_benchmark(args_: argparse.Namespace):
             "truss": 8080,
         }.get(args.backend, 30000)
 
+    # Build base URL with proper IPv6 bracket wrapping (only when base_url is not provided)
+    if not args.base_url:
+        _na = NetworkAddress(args.host, args.port)
+        _host_base = _na.to_url()
+    else:
+        _na = None
+        _host_base = None
+
     model_url = (
-        f"{args.base_url}/v1/models"
+        f"{args.base_url}/v1/models" if args.base_url else f"{_host_base}/v1/models"
-        if args.base_url
-        else f"http://{args.host}:{args.port}/v1/models"
     )
 
     if args.backend == "sglang-embedding":
@@ -1740,43 +1747,39 @@ def run_benchmark(args_: argparse.Namespace):
         )
     elif args.backend in ["sglang", "sglang-native"]:
         api_url = (
-            f"{args.base_url}/generate"
+            f"{args.base_url}/generate" if args.base_url else f"{_host_base}/generate"
-            if args.base_url
-            else f"http://{args.host}:{args.port}/generate"
         )
     elif args.backend in ["sglang-oai", "vllm", "lmdeploy"]:
         api_url = (
             f"{args.base_url}/v1/completions"
             if args.base_url
-            else f"http://{args.host}:{args.port}/v1/completions"
+            else f"{_host_base}/v1/completions"
         )
     elif args.backend in ["sglang-oai-chat", "vllm-chat", "lmdeploy-chat"]:
         api_url = (
             f"{args.base_url}/v1/chat/completions"
             if args.base_url
-            else f"http://{args.host}:{args.port}/v1/chat/completions"
+            else f"{_host_base}/v1/chat/completions"
         )
     elif args.backend == "trt":
         api_url = (
             f"{args.base_url}/v2/models/ensemble/generate_stream"
             if args.base_url
-            else f"http://{args.host}:{args.port}/v2/models/ensemble/generate_stream"
+            else f"{_host_base}/v2/models/ensemble/generate_stream"
         )
         if args.model is None:
             print("Please provide a model using `--model` when using `trt` backend.")
             sys.exit(1)
     elif args.backend == "gserver":
-        api_url = args.base_url if args.base_url else f"{args.host}:{args.port}"
+        api_url = args.base_url if args.base_url else _na.to_host_port_str()
         args.model = args.model or "default"
     elif args.backend == "truss":
         api_url = (
             f"{args.base_url}/v1/models/model:predict"
             if args.base_url
-            else f"http://{args.host}:{args.port}/v1/models/model:predict"
+            else f"{_host_base}/v1/models/model:predict"
         )
-    base_url = (
+    base_url = _host_base if args.base_url is None else args.base_url
-        f"http://{args.host}:{args.port}" if args.base_url is None else args.base_url
-    )
 
     # Wait for server to be ready
     if args.ready_check_timeout_sec > 0:

python/sglang/multimodal_gen/benchmarks/bench_serving.py

@@ -39,6 +39,7 @@ from sglang.multimodal_gen.runtime.utils.logging_utils import (
     init_logger,
 )
 from sglang.multimodal_gen.test.test_utils import print_divider, print_value_formatted
+from sglang.srt.utils.network import NetworkAddress
 
 logger = init_logger(__name__)
 
@@ -457,7 +458,7 @@ async def benchmark(args):
 
     # Construct base_url if not provided
     if args.base_url is None:
-        args.base_url = f"http://{args.host}:{args.port}"
+        args.base_url = NetworkAddress(args.host, args.port).to_url()
 
     # Wait for service
     wait_for_service(args.base_url)

python/sglang/srt/disaggregation/decode.py

@@ -66,6 +66,7 @@ from sglang.srt.observability.req_time_stats import (
     set_schedule_time_batch,
     set_time_batch,
 )
+from sglang.srt.utils.network import NetworkAddress
 from sglang.srt.utils.torch_memory_saver_adapter import TorchMemorySaverAdapter
 
 logger = logging.getLogger(__name__)
@@ -87,7 +88,7 @@ def _is_fake_transfer(req: Req, server_args: ServerArgs) -> bool:
 
 def _bootstrap_addr(req: Req) -> str:
     # FIXME: make a property of a req
-    return f"{req.bootstrap_host}:{req.bootstrap_port}"
+    return NetworkAddress(req.bootstrap_host, req.bootstrap_port).to_host_port_str()
 
 
 class DecodeReqToTokenPool:

python/sglang/srt/disaggregation/encode_grpc_server.py

@@ -258,7 +258,7 @@ async def serve_grpc_encoder(server_args: ServerArgs):
     )
     reflection.enable_server_reflection(SERVICE_NAMES, server)
 
-    listen_addr = f"{server_args.host}:{server_args.port}"
+    listen_addr = NetworkAddress(server_args.host, server_args.port).to_host_port_str()
     server.add_insecure_port(listen_addr)
 
     await server.start()

python/sglang/srt/disaggregation/encode_receiver.py

@@ -29,7 +29,11 @@ from sglang.srt.managers.schedule_batch import Modality, Req
 from sglang.srt.server_args import ServerArgs
 from sglang.srt.utils import ImageData
 from sglang.srt.utils.hf_transformers_utils import get_processor
-from sglang.srt.utils.network import get_local_ip_auto, get_zmq_socket_on_host
+from sglang.srt.utils.network import (
+    NetworkAddress,
+    get_local_ip_auto,
+    get_zmq_socket_on_host,
+)
 
 logger = logging.getLogger(__name__)
 
@@ -447,7 +451,9 @@ class WaitingImageRequest:
                         payload = {
                             "req_id": part_req_id,  # use part_req_id to match encode request
                             "receive_count": receive_count,
-                            "receive_url": f"{host_name}:{embedding_port}",
+                            "receive_url": NetworkAddress(
+                                host_name, embedding_port
+                            ).to_host_port_str(),
                             "modality": modality.name,
                         }
                         logger.info(

python/sglang/srt/disaggregation/mooncake/conn.py

@@ -857,7 +857,8 @@ class MooncakeKVManager(CommonKVManager):
                                     )
                             self.record_failure(
                                 kv_chunk.room,
-                                f"Failed to send kv chunk of {kv_chunk.room} to {req.endpoint}:{req.dst_port}",
+                                f"Failed to send kv chunk of {kv_chunk.room} to "
+                                f"{NetworkAddress(req.endpoint, req.dst_port).to_host_port_str()}",
                             )
                             self.update_status(kv_chunk.room, KVPoll.Failed)
                             self.sync_status_to_decode_endpoint(

python/sglang/srt/utils/common.py

@@ -2380,6 +2380,8 @@ def launch_dummy_health_check_server(host, port, enable_metrics):
     import uvicorn
     from fastapi import FastAPI, Response
 
+    from sglang.srt.utils.network import NetworkAddress
+
     app = FastAPI()
 
     @app.get("/ping")
@@ -2422,14 +2424,16 @@ def launch_dummy_health_check_server(host, port, enable_metrics):
             logger.error(f"Dummy health check server failed to start: {e}")
             raise
         finally:
-            logger.info(f"Dummy health check server stopped at {host}:{port}")
+            logger.info(
+                f"Dummy health check server stopped at {NetworkAddress(host, port).to_host_port_str()}"
+            )
 
     thread = threading.Thread(
         target=run_server, daemon=True, name="health-check-server"
     )
     thread.start()
     logger.info(
-        f"Dummy health check server started in background thread at {host}:{port}"
+        f"Dummy health check server started in background thread at {NetworkAddress(host, port).to_host_port_str()}"
     )