update default request plane to tcp

Author: biswapanda

components/src/dynamo/frontend/main.py

@@ -245,7 +245,7 @@ def parse_args():
         "--request-plane",
         type=str,
         choices=["nats", "http", "tcp"],
-        default=os.environ.get("DYN_REQUEST_PLANE", "nats"),
+        default=os.environ.get("DYN_REQUEST_PLANE", "tcp"),
         help="Determines how requests are distributed from routers to workers. 'tcp' is fastest [nats|http|tcp]",
     )
 

components/src/dynamo/mocker/args.py

@@ -295,7 +295,7 @@ def parse_args():
         "--request-plane",
         type=str,
         choices=["nats", "http", "tcp"],
-        default=os.environ.get("DYN_REQUEST_PLANE", "nats"),
+        default=os.environ.get("DYN_REQUEST_PLANE", "tcp"),
         help="Determines how requests are distributed from routers to workers. 'tcp' is fastest [nats|http|tcp]",
     )
 

components/src/dynamo/sglang/args.py

@@ -113,7 +113,7 @@
         "flags": ["--request-plane"],
         "type": str,
         "choices": ["nats", "http", "tcp"],
-        "default": os.environ.get("DYN_REQUEST_PLANE", "nats"),
+        "default": os.environ.get("DYN_REQUEST_PLANE", "tcp"),
         "help": "Determines how requests are distributed from routers to workers. 'tcp' is fastest [nats|http|tcp]",
     },
 }

components/src/dynamo/trtllm/utils/trtllm_utils.py

@@ -300,7 +300,7 @@ def cmd_line_args():
         "--request-plane",
         type=str,
         choices=["nats", "http", "tcp"],
-        default=os.environ.get("DYN_REQUEST_PLANE", "nats"),
+        default=os.environ.get("DYN_REQUEST_PLANE", "tcp"),
         help="Determines how requests are distributed from routers to workers. 'tcp' is fastest [nats|http|tcp]",
     )
 

components/src/dynamo/vllm/args.py

@@ -201,7 +201,7 @@ def parse_args() -> Config:
         "--request-plane",
         type=str,
         choices=["nats", "http", "tcp"],
-        default=os.environ.get("DYN_REQUEST_PLANE", "nats"),
+        default=os.environ.get("DYN_REQUEST_PLANE", "tcp"),
         help="Determines how requests are distributed from routers to workers. 'tcp' is fastest [nats|http|tcp]",
     )
     parser.add_argument(