GPU检测修改

src/main.py

@@ -366,6 +366,156 @@ def health_check():
         return jsonify({'status': 'error', 'code': 1, 'message': str(e)})
 
 
+# ============ 详细系统监控 ============
+@app.route('/api/system-info', methods=['GET'])
+def system_info():
+    """获取详细系统监控信息"""
+    import psutil
+    import os
+    try:
+        # CPU 信息
+        cpu_percent = psutil.cpu_percent(interval=None)
+        cpu_counts = psutil.cpu_count()
+        cpu_freq = psutil.cpu_freq()
+
+        # 内存信息
+        memory = psutil.virtual_memory()
+
+        # 磁盘信息
+        disk = psutil.disk_usage('/')
+        disk_io = psutil.disk_io_counters()
+
+        # 网络信息
+        net_io = psutil.net_io_counters()
+
+        # 系统启动时间
+        boot_time = psutil.boot_time()
+        uptime_seconds = time.time() - boot_time
+
+        # GPU 信息
+        gpu_list = []
+        try:
+            import pynvml
+            pynvml.nvmlInit()
+            gpu_count = pynvml.nvmlDeviceGetCount()
+            for i in range(gpu_count):
+                try:
+                    handle = pynvml.nvmlDeviceGetHandleByIndex(i)
+                    name = pynvml.nvmlDeviceGetName(handle)
+
+                    # 获取显存信息
+                    try:
+                        mem_info = pynvml.nvmlDeviceGetMemoryInfo(handle)
+                        memory_used = mem_info.used
+                        memory_total = mem_info.total
+                    except:
+                        memory_used = 0
+                        memory_total = 0
+
+                    # 获取利用率
+                    try:
+                        util = pynvml.nvmlDeviceGetUtilizationRates(handle)
+                        gpu_util = util.gpu
+                        mem_util = util.memory
+                    except:
+                        gpu_util = 0
+                        mem_util = 0
+
+                    # 获取温度 - pynvml 11.x API: 只接受handle参数
+                    try:
+                        temp = pynvml.nvmlDeviceGetTemperature(handle)
+                    except:
+                        temp = 0
+
+                    # 获取功耗
+                    try:
+                        power = pynvml.nvmlDeviceGetPowerUsage(handle)
+                    except:
+                        power = 0
+
+                    # 获取风扇转速 (百分比)
+                    try:
+                        fan_speed = pynvml.nvmlDeviceGetFanSpeed(handle)
+                    except:
+                        fan_speed = 0
+
+                    # 获取显卡时钟频率 (MHz)
+                    try:
+                        clock = pynvml.nvmlDeviceGetClockInfo(handle, pynvml.NVML_CLOCK_SM)
+                    except:
+                        clock = 0
+
+                    # 获取显存时钟频率 (MHz)
+                    try:
+                        mem_clock = pynvml.nvmlDeviceGetClockInfo(handle, pynvml.NVML_CLOCK_MEM)
+                    except:
+                        mem_clock = 0
+
+                    # 获取驱动版本信息
+                    try:
+                        version = pynvml.nvmlSystemGetDriverVersion()
+                    except:
+                        version = ''
+
+                    gpu_list.append({
+                        'name': name.decode() if isinstance(name, bytes) else name,
+                        'memory_used_gb': round(memory_used / (1024**3), 1),
+                        'memory_total_gb': round(memory_total / (1024**3), 1),
+                        'gpu_percent': gpu_util,
+                        'memory_percent': mem_util,
+                        'temperature': temp,
+                        'power_w': round(power / 1000, 1) if power > 0 else 0,
+                        'fan_speed': fan_speed,
+                        'clock_mhz': clock,
+                        'memory_clock_mhz': mem_clock,
+                        'driver_version': version.decode() if isinstance(version, bytes) else version
+                    })
+                except Exception as e:
+                    logger.debug(f"获取GPU {i} 信息失败: {e}")
+                    continue
+            pynvml.nvmlShutdown()
+        except Exception as e:
+            logger.warning(f"获取GPU信息失败: {e}")
+            gpu_list = []
+
+        return jsonify({
+            'code': 0,
+            'data': {
+                'cpu': {
+                    'percent': cpu_percent,
+                    'cores': cpu_counts,
+                    'frequency_mhz': cpu_freq.current if cpu_freq else 0
+                },
+                'memory': {
+                    'percent': memory.percent,
+                    'used_gb': round(memory.used / (1024**3), 1),
+                    'total_gb': round(memory.total / (1024**3), 1),
+                    'available_gb': round(memory.available / (1024**3), 1),
+                    'cached_gb': round(memory.cached / (1024**3), 1) if hasattr(memory, 'cached') else 0
+                },
+                'disk': {
+                    'percent': disk.percent,
+                    'used_gb': round(disk.used / (1024**3), 0),
+                    'total_gb': round(disk.total / (1024**3), 0),
+                    'read_mb': round(disk_io.read_bytes / (1024**2), 0),
+                    'write_mb': round(disk_io.write_bytes / (1024**2), 0)
+                },
+                'network': {
+                    'upload_mb': round(net_io.bytes_sent / (1024**2), 1),
+                    'download_mb': round(net_io.bytes_recv / (1024**2), 1)
+                },
+                'system': {
+                    'uptime_seconds': uptime_seconds,
+                    'process_count': len(psutil.pids())
+                },
+                'gpu': gpu_list
+            }
+        })
+    except Exception as e:
+        logger.error(f"获取系统信息失败: {e}")
+        return jsonify({'code': 1, 'message': str(e)})
+
+
 # ============ 通用 CRUD 操作 ============
 def generic_get_all(table_name, order_by='create_time DESC'):
     """通用查询所有"""