Better PD initialization (#5751)

python/sglang/srt/server_args.py

@@ -198,6 +198,7 @@ class ServerArgs:
     disaggregation_bootstrap_port: int = 8998
     disaggregation_transfer_backend: str = "mooncake"
     disaggregation_ib_device: Optional[str] = None
+    pdlb_url: Optional[str] = None
 
     def __post_init__(self):
         # Expert parallelism
@@ -1254,6 +1255,12 @@ class ServerArgs:
             "or multiple comma-separated devices (e.g., --disaggregation-ib-device mlx5_0,mlx5_1). "
             "Default is None, which triggers automatic device detection when mooncake backend is enabled.",
         )
+        parser.add_argument(
+            "--pdlb-url",
+            type=str,
+            default=None,
+            help="The URL of the PD disaggregation load balancer. If set, the prefill/decode server will register with the load balancer.",
+        )
 
     @classmethod
     def from_cli_args(cls, args: argparse.Namespace):