添加了后端监控idle_time接口

Makefile

@@ -94,7 +94,7 @@ push-visualize:
 .PHONY: clean
 clean:
 	-docker images -f "dangling=true" -q | xargs -r docker rmi
-	-docker images | grep lerobot | grep -v $(VERSION) | awk '{print $$1":"$$2}' | xargs -r docker rmi
+	-docker images | grep lerobot | grep -v $(VERSION) | grep -v $(REMOTE_VERSION) |awk '{print $$1":"$$2}' | xargs -r docker rmi
 
 # --- Help ---
 .PHONY: help

docker/Dockerfile.infer

@@ -64,7 +64,7 @@ RUN uv venv
 COPY --chown=user_lerobot:user_lerobot pyproject.toml README.md MANIFEST.in ./
 # COPY --chown=user_lerobot:user_lerobot src/ src/
 RUN uv pip install --no-cache ".[smolvla]"
-RUN uv pip install pyzmq msgpack msgpack_numpy zstandard
+RUN uv pip install pyzmq msgpack msgpack_numpy zstandard fastapi uvicorn
 
 # Set the default command - Online Inference mode
 ADD docker/infer.py /workspace/infer.py

docker/infer.py

@@ -1,11 +1,13 @@
 import json
 import os
-import signal
 import threading
 import time
 
 import torch
 
+from fastapi import FastAPI
+import uvicorn
+
 from cloud_helper import Server
 from lerobot.policies.factory import get_policy_class
 
@@ -27,6 +29,7 @@ if not checkpoint.endswith("/pretrained_model"):
 
 server_port = task_configs["online_infer"].get("port", 8080)
 
+
 class LerobotInferenceServer:
     def __init__(
         self,
@@ -35,7 +38,7 @@ class LerobotInferenceServer:
         host: str = "localhost",
         port: int = 5555,
         device="cuda",
-        timeout: int = 3600,
+        http_port: int = 80,
     ):
         self.server = Server(host, port)
         self.policy_type = policy_type
@@ -45,20 +48,8 @@ class LerobotInferenceServer:
         self.policy.to(self.device)
         print(f"Loaded {self.policy_type.upper()} policy from {checkpoint}")
 
-        self.timeout = timeout
         self.last_activity = time.time()
-        self.stop_event = threading.Event()
-        self.monitor_thread = threading.Thread(target=self.watchout, daemon=True)
-        self.monitor_thread.start()
-
-    def watchout(self):
-        while not self.stop_event.is_set():
-            time.sleep(6)  # Check every 6 seconds
-            elapsed = time.time() - self.last_activity
-            if elapsed > self.timeout:
-                print(f"No activity for {elapsed:.0f} seconds. Shutting down due to timeout.")
-                # Force exit since loop_forever might block
-                os.kill(os.getpid(), signal.SIGINT)
+        self.fastapi = FastAPI()
 
     def get_actions(self, batch):
         # batch = {
@@ -89,16 +80,33 @@ class LerobotInferenceServer:
 
         return action_chunk.cpu().numpy()  # (B, chunk_size, action_dim)
 
+    def get_idle_time(self):
+        return time.time() - self.last_activity
+
     def run(self):
         self.server.register_endpoint("get_actions", self.get_actions)
-        print(f"Lerobot {self.policy_type.upper()} Server is running...")
 
+        @self.fastapi.get("/health")
+        def health_check():
+            return {"status": 0}
+
+        @self.fastapi.get("/idle_time")
+        def idle_time():
+            return {"status": 0, "idle_time": self.get_idle_time()}
+
+        def start_fastapi(app, port: int = 80):
+            """在独立线程中启动 FastAPI"""
+            print(f"Starting FastAPI HTTP server on port {port}...")
+            uvicorn.run(app, host="0.0.0.0", port=port, log_level="info")
+
+        threading.Thread(target=start_fastapi, args=(self.fastapi,), daemon=True).start()
+
+        print(f"Lerobot {self.policy_type.upper()} Server is running...")
         self.server.loop_forever()
 
 
 if __name__ == "__main__":
-
     server = LerobotInferenceServer(
-        checkpoint=checkpoint, policy_type=model, host="0.0.0.0", port=server_port, timeout=3600
+        checkpoint=checkpoint, policy_type=model, host="0.0.0.0", port=server_port
     )
     server.run()