blanchon's picture
Update
63ed3a7
raw
history blame
11.4 kB
import logging
import os
from contextlib import asynccontextmanager
from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from inference_server.models import list_supported_policies
from inference_server.session_manager import SessionManager
# Configure logging
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(name)s - %(levelname)s - %(message)s",
)
logger = logging.getLogger(__name__)
# Global session manager
session_manager = SessionManager()
@asynccontextmanager
async def lifespan(app: FastAPI):
"""Handle app startup and shutdown."""
logger.info("🚀 Inference Server starting up...")
yield
logger.info("🔄 Inference Server shutting down...")
await session_manager.cleanup_all_sessions()
logger.info("✅ Inference Server shutdown complete")
# FastAPI app
app = FastAPI(
title="Inference Server",
description="Multi-Policy Model Inference Server for Real-time Robot Control",
version="1.0.0",
lifespan=lifespan,
)
# Add CORS middleware
app.add_middleware(
CORSMiddleware,
allow_origins=["*"], # In production, specify actual origins
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
# Request/Response models
class CreateSessionRequest(BaseModel):
session_id: str
policy_path: str
camera_names: list[str] = ["front"] # Support multiple cameras
arena_server_url: str = "http://localhost:8000"
workspace_id: str | None = None # Optional workspace ID
policy_type: str = "act" # Policy type: act, pi0, pi0fast, smolvla, diffusion
language_instruction: str | None = None # For vision-language policies
class CreateSessionResponse(BaseModel):
workspace_id: str
camera_room_ids: dict[str, str] # {camera_name: room_id}
joint_input_room_id: str
joint_output_room_id: str
class SessionStatusResponse(BaseModel):
session_id: str
status: str
policy_path: str
policy_type: str
camera_names: list[str] # Multiple camera names
workspace_id: str
rooms: dict
stats: dict
inference_stats: dict | None = None
error_message: str | None = None
# Health check
@app.get("/", tags=["Health"])
async def root():
"""Health check endpoint."""
return {"message": "Inference Server is running", "status": "healthy"}
@app.get("/health", tags=["Health"])
async def health_check():
"""Detailed health check."""
return {
"status": "healthy",
"active_sessions": len(session_manager.sessions),
"session_ids": list(session_manager.sessions.keys()),
}
@app.get("/policies", tags=["Policies"])
async def list_policies():
"""List supported policy types."""
return {
"supported_policies": list_supported_policies(),
"description": "Available policy types for inference",
}
# Session management endpoints
@app.post("/sessions", response_model=CreateSessionResponse, tags=["Sessions"])
async def create_session(request: CreateSessionRequest):
"""
Create a new inference session.
If workspace_id is provided, all rooms will be created in that workspace.
If workspace_id is not provided, a new workspace will be generated automatically.
All rooms for a session (cameras + joints) are always created in the same workspace.
"""
try:
room_ids = await session_manager.create_session(
session_id=request.session_id,
policy_path=request.policy_path,
camera_names=request.camera_names,
arena_server_url=request.arena_server_url,
workspace_id=request.workspace_id,
policy_type=request.policy_type,
language_instruction=request.language_instruction,
)
return CreateSessionResponse(**room_ids)
except ValueError as e:
raise HTTPException(status_code=400, detail=str(e))
except Exception as e:
logger.exception(f"Failed to create session {request.session_id}")
raise HTTPException(status_code=500, detail=f"Failed to create session: {e!s}")
@app.get("/sessions", response_model=list[SessionStatusResponse], tags=["Sessions"])
async def list_sessions():
"""List all sessions."""
sessions = await session_manager.list_sessions()
return [SessionStatusResponse(**session) for session in sessions]
@app.get(
"/sessions/{session_id}", response_model=SessionStatusResponse, tags=["Sessions"]
)
async def get_session_status(session_id: str):
"""Get status of a specific session."""
try:
status = await session_manager.get_session_status(session_id)
return SessionStatusResponse(**status)
except KeyError:
raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
@app.post("/sessions/{session_id}/start", tags=["Control"])
async def start_inference(session_id: str):
"""Start inference for a session."""
try:
await session_manager.start_inference(session_id)
return {"message": f"Inference started for session {session_id}"}
except KeyError:
raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
except Exception as e:
logger.exception(f"Failed to start inference for session {session_id}")
raise HTTPException(status_code=500, detail=f"Failed to start inference: {e!s}")
@app.post("/sessions/{session_id}/stop", tags=["Control"])
async def stop_inference(session_id: str):
"""Stop inference for a session."""
try:
await session_manager.stop_inference(session_id)
return {"message": f"Inference stopped for session {session_id}"}
except KeyError:
raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
@app.post("/sessions/{session_id}/restart", tags=["Control"])
async def restart_inference(session_id: str):
"""Restart inference for a session."""
try:
await session_manager.restart_inference(session_id)
return {"message": f"Inference restarted for session {session_id}"}
except KeyError:
raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
except Exception as e:
logger.exception(f"Failed to restart inference for session {session_id}")
raise HTTPException(
status_code=500, detail=f"Failed to restart inference: {e!s}"
)
@app.delete("/sessions/{session_id}", tags=["Sessions"])
async def delete_session(session_id: str):
"""Delete a session."""
try:
await session_manager.delete_session(session_id)
return {"message": f"Session {session_id} deleted"}
except KeyError:
raise HTTPException(status_code=404, detail=f"Session {session_id} not found")
# Debug endpoints for enhanced monitoring
@app.get("/debug/system", tags=["Debug"])
async def get_system_info():
"""Get system information for debugging."""
import psutil
import torch
try:
# System info
system_info = {
"cpu_percent": psutil.cpu_percent(interval=1),
"memory": {
"total": psutil.virtual_memory().total,
"available": psutil.virtual_memory().available,
"percent": psutil.virtual_memory().percent,
},
"disk": {
"total": psutil.disk_usage("/").total,
"used": psutil.disk_usage("/").used,
"percent": psutil.disk_usage("/").percent,
},
}
# GPU info if available
if torch.cuda.is_available():
system_info["gpu"] = {
"device_count": torch.cuda.device_count(),
"current_device": torch.cuda.current_device(),
"device_name": torch.cuda.get_device_name(),
"memory_allocated": torch.cuda.memory_allocated(),
"memory_cached": torch.cuda.memory_reserved(),
}
return system_info
except Exception as e:
return {"error": f"Failed to get system info: {e}"}
@app.get("/debug/logs", tags=["Debug"])
async def get_recent_logs():
"""Get recent log entries for debugging."""
try:
# This is a simple implementation - in production you might want to read from actual log files
return {
"message": "Log endpoint available",
"note": "Implement actual log reading if needed",
"active_sessions": len(session_manager.sessions),
}
except Exception as e:
return {"error": f"Failed to get logs: {e}"}
@app.post("/debug/sessions/{session_id}/reset", tags=["Debug"])
async def debug_reset_session(session_id: str):
"""Reset a session's internal state for debugging."""
try:
if session_id not in session_manager.sessions:
raise HTTPException(
status_code=404, detail=f"Session {session_id} not found"
)
session = session_manager.sessions[session_id]
# Reset inference engine if available
if session.inference_engine:
session.inference_engine.reset()
# Clear action queue
session.action_queue.clear()
# Reset flags
for camera_name in session.camera_names:
session.images_updated[camera_name] = False
session.joints_updated = False
return {"message": f"Session {session_id} state reset successfully"}
except Exception as e:
logger.exception(f"Failed to reset session {session_id}")
raise HTTPException(status_code=500, detail=f"Failed to reset session: {e}")
@app.get("/debug/sessions/{session_id}/queue", tags=["Debug"])
async def get_session_queue_info(session_id: str):
"""Get detailed information about a session's action queue."""
try:
if session_id not in session_manager.sessions:
raise HTTPException(
status_code=404, detail=f"Session {session_id} not found"
)
session = session_manager.sessions[session_id]
return {
"session_id": session_id,
"queue_length": len(session.action_queue),
"queue_maxlen": session.action_queue.maxlen,
"n_action_steps": session.n_action_steps,
"control_frequency_hz": session.control_frequency_hz,
"inference_frequency_hz": session.inference_frequency_hz,
"last_queue_cleanup": session.last_queue_cleanup,
"data_status": {
"has_joint_data": session.latest_joint_positions is not None,
"images_status": {
camera: camera in session.latest_images
for camera in session.camera_names
},
"images_updated": session.images_updated.copy(),
"joints_updated": session.joints_updated,
},
}
except Exception as e:
logger.exception(f"Failed to get queue info for session {session_id}")
raise HTTPException(status_code=500, detail=f"Failed to get queue info: {e}")
# Main entry point
if __name__ == "__main__":
import uvicorn
port = int(os.environ.get("PORT", 8001))
uvicorn.run(
"inference_server.main:app",
host="0.0.0.0",
port=port,
reload=False,
log_level="info",
)