Simplify the process launch code in server.py (#2923)

python/sglang/srt/server.py

@@ -44,7 +44,6 @@ import uvloop
 from fastapi import FastAPI, File, Form, Request, UploadFile
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import ORJSONResponse, Response, StreamingResponse
-from uvicorn.config import LOGGING_CONFIG
 
 from sglang.lang.backend.runtime_endpoint import RuntimeEndpoint
 from sglang.srt.hf_transformers_utils import get_tokenizer
@@ -97,6 +96,7 @@ from sglang.srt.utils import (
     prepare_model_and_tokenizer,
     set_prometheus_multiproc_dir,
     set_ulimit,
+    set_uvicorn_logging_configs,
 )
 from sglang.utils import get_exception_traceback
 from sglang.version import __version__
@@ -474,13 +474,13 @@ def launch_engine(
         server_args.model_path, server_args.tokenizer_path
     )
 
-    memory_saver_adapter = TorchMemorySaverAdapter.create(
-        enable=server_args.enable_memory_saver
-    )
-
+    scheduler_procs = []
     if server_args.dp_size == 1:
         # Launch tensor parallel scheduler processes
-        scheduler_procs = []
+        memory_saver_adapter = TorchMemorySaverAdapter.create(
+            enable=server_args.enable_memory_saver
+        )
+
         scheduler_pipe_readers = []
         tp_size_per_node = server_args.tp_size // server_args.nnodes
         tp_rank_range = range(
@@ -498,12 +498,6 @@ def launch_engine(
                 proc.start()
             scheduler_procs.append(proc)
             scheduler_pipe_readers.append(reader)
-
-        if server_args.node_rank >= 1:
-            # For other nodes, they do not need to run tokenizer or detokenizer,
-            # so they can just wait here.
-            for proc in scheduler_procs:
-                proc.join()
     else:
         # Launch the data parallel controller
         reader, writer = mp.Pipe(duplex=False)
@@ -512,8 +506,27 @@ def launch_engine(
             target=run_data_parallel_controller_process,
             args=(server_args, port_args, writer),
         )
-        with memory_saver_adapter.configure_subprocess():
-            proc.start()
+        proc.start()
+        scheduler_procs.append(proc)
+
+    if server_args.node_rank >= 1:
+        # In multi-node cases, non-zero rank nodes do not need to run tokenizer or detokenizer,
+        # so they can just wait here.
+
+        for reader in scheduler_pipe_readers:
+            data = reader.recv()
+            assert data["status"] == "ready"
+
+        if os.getenv("SGLANG_BLOCK_NONZERO_RANK_CHILDREN") == "0":
+            # When using `Engine` as a Python API, we don't want to block here.
+            return
+
+        for proc in scheduler_procs:
+            proc.join()
+            logger.error(
+                f"Scheduler or DataParallelController {proc.pid} terminated with {proc.exitcode}"
+            )
+        return
 
     # Launch detokenizer process
     detoken_proc = mp.Process(
@@ -597,14 +610,7 @@ def launch_server(
 
     try:
         # Update logging configs
-        LOGGING_CONFIG["formatters"]["default"][
-            "fmt"
-        ] = "[%(asctime)s] %(levelprefix)s %(message)s"
-        LOGGING_CONFIG["formatters"]["default"]["datefmt"] = "%Y-%m-%d %H:%M:%S"
-        LOGGING_CONFIG["formatters"]["access"][
-            "fmt"
-        ] = '[%(asctime)s] %(levelprefix)s %(client_addr)s - "%(request_line)s" %(status_code)s'
-        LOGGING_CONFIG["formatters"]["access"]["datefmt"] = "%Y-%m-%d %H:%M:%S"
+        set_uvicorn_logging_configs()
 
         # Listen for HTTP requests
         uvicorn.run(

python/sglang/srt/utils.py

@@ -59,6 +59,7 @@ from triton.runtime.cache import (
     default_dump_dir,
     default_override_dir,
 )
+from uvicorn.config import LOGGING_CONFIG
 
 logger = logging.getLogger(__name__)
 
@@ -1404,3 +1405,14 @@ def nullable_str(val: str):
     if not val or val == "None":
         return None
     return val
+
+
+def set_uvicorn_logging_configs():
+    LOGGING_CONFIG["formatters"]["default"][
+        "fmt"
+    ] = "[%(asctime)s] %(levelprefix)s %(message)s"
+    LOGGING_CONFIG["formatters"]["default"]["datefmt"] = "%Y-%m-%d %H:%M:%S"
+    LOGGING_CONFIG["formatters"]["access"][
+        "fmt"
+    ] = '[%(asctime)s] %(levelprefix)s %(client_addr)s - "%(request_line)s" %(status_code)s'
+    LOGGING_CONFIG["formatters"]["access"]["datefmt"] = "%Y-%m-%d %H:%M:%S"