Fix data parallel perf regression (#6183)

Author: merrymercy

.github/workflows/execute-notebook.yml

@@ -22,11 +22,6 @@ jobs:
       - name: Checkout code
         uses: actions/checkout@v4
 
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: '3.9'
-
       - name: Install dependencies
         run: |
           bash scripts/ci_install_dependency.sh
@@ -35,6 +30,8 @@ jobs:
           apt-get install -y pandoc
           apt-get update && apt-get install -y parallel retry
 
+          ln -sf "$(which python3)" /usr/bin/python
+
       - name: Setup Jupyter Kernel
         run: |
           python -m ipykernel install --user --name python3 --display-name "Python 3"

python/sglang/srt/managers/data_parallel_controller.py

@@ -17,13 +17,13 @@
 import multiprocessing as mp
 import signal
 import threading
+import time
 from enum import Enum, auto
 
 import psutil
 import setproctitle
 import zmq
 
-from sglang.srt.disaggregation.utils import DisaggregationMode
 from sglang.srt.layers.dp_attention import compute_dp_attention_world_info
 from sglang.srt.managers.io_struct import (
     TokenizedEmbeddingReqInput,
@@ -158,7 +158,7 @@ def launch_tensor_parallel_group_thread(
         # This thread cannot be closed because otherwise the `kill_itself_when_parent_died`
         # function in scheduler.py will kill the scheduler.
         while True:
-            pass
+            time.sleep(30 * 24 * 3600)
 
     def launch_dp_attention_schedulers(self, server_args, port_args):
         self.launch_tensor_parallel_group(server_args, port_args, 0, None)