/* * health_monitor.cpp - 系统健康监控 * 包含:健康状态管理 + WiFi RSSI 监控 + 温度监控 * * 所有功能封装为 HealthMonitor 类方法。 * 所有对 m_healthState / m_lastTemperature / m_lastWifiRssi 的访问 * 均通过 m_mutex 保护,消除跨任务竞态。 */ #include #include #include "freertos/FreeRTOS.h" #include "freertos/task.h" #include "freertos/semphr.h" #include "esp_wifi.h" #include "esp_timer.h" #include "esp_err.h" #include "driver/temperature_sensor.h" #include "health_monitor.h" #include "debug_log.h" #include "server/wifi_ota.h" // ============================================================ // 单例 // ============================================================ HealthMonitor& HealthMonitor::instance() { static HealthMonitor inst; return inst; } // ============================================================ // 初始化 // ============================================================ void HealthMonitor::init() { m_mutex = xSemaphoreCreateMutex(); if (m_mutex == NULL) { DEBUG_ERR("HEALTH", "创建健康状态互斥锁失败"); } } // ============================================================ // 线程安全的健康状态读写 // ============================================================ health_state_t HealthMonitor::getState() { if (m_mutex) xSemaphoreTake(m_mutex, portMAX_DELAY); health_state_t s = m_healthState; if (m_mutex) xSemaphoreGive(m_mutex); return s; } void HealthMonitor::setState(health_state_t newState) { if (m_mutex) xSemaphoreTake(m_mutex, portMAX_DELAY); m_healthState = newState; if (m_mutex) xSemaphoreGive(m_mutex); } // ============================================================ // 获取当前健康状态(public 接口) // ============================================================ health_state_t HealthMonitor::getCurrent() { return getState(); } float HealthMonitor::getLastTemperature() { if (m_mutex) xSemaphoreTake(m_mutex, portMAX_DELAY); float t = m_lastTemperature; if (m_mutex) xSemaphoreGive(m_mutex); return t; } bool HealthMonitor::getLastWifiRssi(int8_t *out_rssi, uint32_t *out_age_ms) { if (m_mutex) xSemaphoreTake(m_mutex, portMAX_DELAY); int64_t lastMs = m_lastWifiRssiMs; int8_t rssi = m_lastWifiRssi; if (m_mutex) xSemaphoreGive(m_mutex); if (lastMs <= 0) { return false; } int64_t now_ms = esp_timer_get_time() / 1000; uint32_t age_ms = (now_ms > lastMs) ? (uint32_t)(now_ms - lastMs) : 0; if (out_rssi) *out_rssi = rssi; if (out_age_ms) *out_age_ms = age_ms; return true; } // ============================================================ // 温度监控(定时器回调) // ============================================================ void HealthMonitor::tempMonitorCallback(void *arg) { auto& self = instance(); float temp = 0.0f; esp_err_t err = temperature_sensor_get_celsius(self.m_tempHandle, &temp); if (err != ESP_OK) { DEBUG_WARN("TEMP", "读取温度失败: %s", esp_err_to_name(err)); return; } if (self.m_mutex) xSemaphoreTake(self.m_mutex, portMAX_DELAY); self.m_lastTemperature = temp; if (self.m_mutex) xSemaphoreGive(self.m_mutex); if (temp >= TEMP_CRITICAL) { self.setState(SYS_SAFE_SHUTDOWN); DEBUG_ERR("TEMP", "!!! 危险 !!! 温度 %.1f°C >= %d°C,触发安全关机", temp, TEMP_CRITICAL); DEBUG_ERR("TEMP", "健康状态变更为: SYS_SAFE_SHUTDOWN"); } else if (temp >= TEMP_WARNING) { health_state_t cur = self.getState(); if (cur < SYS_WARNING) { self.setState(SYS_WARNING); DEBUG_WARN("TEMP", "警告: 温度 %.1f°C >= %d°C", temp, TEMP_WARNING); DEBUG_WARN("TEMP", "健康状态变更为: SYS_WARNING"); } } } // ============================================================ // FreeRTOS 任务入口 // ============================================================ void HealthMonitor::taskEntry(void *pvParameters) { auto& self = instance(); DEBUG_INFO("MON", "=== 健康监控任务启动 ==="); temperature_sensor_config_t temp_config = TEMPERATURE_SENSOR_CONFIG_DEFAULT(20, 80); esp_err_t err = temperature_sensor_install(&temp_config, &self.m_tempHandle); if (err != ESP_OK) { DEBUG_ERR("MON", "温度传感器初始化失败: %s", esp_err_to_name(err)); } else { ESP_ERROR_CHECK(temperature_sensor_enable(self.m_tempHandle)); DEBUG_INFO("MON", "温度传感器已启用 (范围 20°C ~ 80°C)"); } // 定时器,每1秒读取一次温度 esp_timer_handle_t temp_timer; esp_timer_create_args_t timer_args = {}; timer_args.callback = &tempMonitorCallback; timer_args.arg = nullptr; timer_args.name = "temp_monitor"; timer_args.dispatch_method = ESP_TIMER_TASK; if (esp_timer_create(&timer_args, &temp_timer) == ESP_OK) { esp_timer_start_periodic(temp_timer, 1000000); // 1秒 DEBUG_INFO("MON", "温度监控定时器已启动 (1秒间隔)"); } else { DEBUG_ERR("MON", "创建温度监控定时器失败"); } // 周期性巡检 while (1) { uint32_t free_heap = esp_get_free_heap_size(); if (free_heap < HEAP_WARNING_MIN) { DEBUG_WARN("MON", "堆内存不足: %lu < %lu", free_heap, HEAP_WARNING_MIN); } health_state_t cur = self.getState(); if (cur >= SYS_CRITICAL) { DEBUG_ERR("MON", "健康状态危险: %d", cur); } vTaskDelay(pdMS_TO_TICKS(5000)); } } // ============================================================ // 健康事件日志 // ============================================================ void HealthMonitor::logEvent(const char *event, int value) { DEBUG_INFO("HEALTH", "[健康事件] %s: %d", event, value); } // ============================================================ // OTA 健康检查 // ============================================================ health_status_t HealthMonitor::checkBeforeOta() { health_status_t status = HEALTH_MON_OK; health_state_t cur = getState(); if (cur >= SYS_SAFE_SHUTDOWN) { DEBUG_WARN("HEALTH", "OTA 前检查: SYS_SAFE_SHUTDOWN,禁止传输"); return HEALTH_MON_SHUTDOWN; } else if (cur >= SYS_CRITICAL) { DEBUG_WARN("HEALTH", "OTA 前检查: SYS_CRITICAL,禁止传输"); return HEALTH_MON_CRITICAL; } else if (cur >= SYS_WARNING) { status = HEALTH_MON_WARNING; DEBUG_WARN("HEALTH", "OTA 前检查: SYS_WARNING,继续传输但需注意"); } if (!WifiOta::isConnected()) { DEBUG_WARN("HEALTH", "OTA 前检查: WiFi 未连接"); return HEALTH_MON_CRITICAL; } int8_t last_rssi = -127; uint32_t age_ms = 0; if (getLastWifiRssi(&last_rssi, &age_ms)) { if (age_ms <= OTA_WIFI_RSSI_SAMPLE_MAX_AGE_MS) { if (last_rssi < OTA_WIFI_RSSI_CRITICAL) { DEBUG_WARN("HEALTH", "OTA 前检查: 最近 RSSI %d < %d (样本年龄 %lu ms)", last_rssi, OTA_WIFI_RSSI_CRITICAL, (unsigned long)age_ms); return HEALTH_MON_CRITICAL; } else if (last_rssi < WIFI_RSSI_NORMAL) { if (status < HEALTH_MON_WARNING) { status = HEALTH_MON_WARNING; } DEBUG_WARN("HEALTH", "OTA 前检查: 最近 RSSI %d < %d (样本年龄 %lu ms)", last_rssi, WIFI_RSSI_NORMAL, (unsigned long)age_ms); } } } if (status == HEALTH_MON_OK) { DEBUG_DBG("HEALTH", "OTA 前检查: 正常"); } return status; } health_status_t HealthMonitor::checkBeforeOtaRelay() { health_status_t status = HEALTH_MON_OK; health_state_t cur = getState(); if (cur >= SYS_SAFE_SHUTDOWN) { DEBUG_WARN("HEALTH", "OTA 中继检查: SYS_SAFE_SHUTDOWN,禁止传输"); return HEALTH_MON_SHUTDOWN; } else if (cur >= SYS_CRITICAL) { /* SYS_CRITICAL 可能是由 WiFi RSSI 过低导致的, * 但 SPI 中继阶段不依赖 WiFi,只检查温度/硬件等本地状态。 * 温度达到 TEMP_CRITICAL(100°C) 时 m_healthState 会被设为 * SYS_SAFE_SHUTDOWN,在上面的分支已经拦截。 * 所以 SYS_CRITICAL 在中继阶段只可能是 WiFi 导致的,允许继续。 */ DEBUG_WARN("HEALTH", "OTA 中继检查: SYS_CRITICAL (WiFi 相关),SPI 中继允许继续"); } else if (cur >= SYS_WARNING) { status = HEALTH_MON_WARNING; DEBUG_WARN("HEALTH", "OTA 中继检查: SYS_WARNING,继续传输"); } /* 跳过 WiFi 连接状态和 RSSI 检查: * SPI 中继阶段固件已存储在 ESP32 Flash 中, * 不依赖 WiFi 连接,无需检查 WiFi 状态。 */ if (status == HEALTH_MON_OK) { DEBUG_DBG("HEALTH", "OTA 中继检查: 正常"); } return status; } // ============================================================ // WiFi RSSI 更新 // ============================================================ health_status_t HealthMonitor::updateFromWifiRssi(int8_t rssi) { if (rssi < WIFI_RSSI_CRITICAL) { setState(SYS_CRITICAL); DEBUG_WARN("HEALTH", "WiFi RSSI 危险: %d dBm < %d dBm", rssi, WIFI_RSSI_CRITICAL); logEvent("WIFI危险", rssi); return HEALTH_MON_CRITICAL; } else if (rssi < WIFI_RSSI_NORMAL) { health_state_t cur = getState(); if (cur < SYS_WARNING) { setState(SYS_WARNING); DEBUG_WARN("HEALTH", "WiFi RSSI 警告: %d dBm < %d dBm", rssi, WIFI_RSSI_NORMAL); logEvent("WIFI警告", rssi); } return HEALTH_MON_WARNING; } health_state_t cur = getState(); if (cur == SYS_WARNING && rssi >= WIFI_RSSI_NORMAL) { setState(SYS_NORMAL); DEBUG_INFO("HEALTH", "WiFi RSSI 恢复: %d dBm >= %d dBm", rssi, WIFI_RSSI_NORMAL); logEvent("WIFI恢复", rssi); } return HEALTH_MON_OK; }