2026-05-30 20:04:04 +08:00
|
|
|
|
/*
|
2026-04-27 02:34:47 +08:00
|
|
|
|
* health_monitor.cpp - 系统健康监控
|
2026-05-05 21:55:29 +08:00
|
|
|
|
* 包含:健康状态管理 + WiFi RSSI 监控 + 温度监控
|
2026-05-31 03:20:26 +08:00
|
|
|
|
*
|
|
|
|
|
|
* 所有功能封装为 HealthMonitor 类方法。
|
2026-04-27 00:42:26 +08:00
|
|
|
|
*/
|
|
|
|
|
|
#include <stdio.h>
|
|
|
|
|
|
#include <string.h>
|
|
|
|
|
|
#include "freertos/FreeRTOS.h"
|
|
|
|
|
|
#include "freertos/task.h"
|
2026-04-27 02:34:47 +08:00
|
|
|
|
#include "freertos/semphr.h"
|
2026-04-27 00:42:26 +08:00
|
|
|
|
#include "esp_wifi.h"
|
2026-04-27 02:34:47 +08:00
|
|
|
|
#include "esp_timer.h"
|
|
|
|
|
|
#include "esp_err.h"
|
|
|
|
|
|
#include "driver/temperature_sensor.h"
|
2026-04-27 00:42:26 +08:00
|
|
|
|
|
|
|
|
|
|
#include "health_monitor.h"
|
|
|
|
|
|
#include "debug_log.h"
|
2026-05-05 21:55:29 +08:00
|
|
|
|
#include "server/wifi_ota.h"
|
2026-04-27 00:42:26 +08:00
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
// 静态成员
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
health_state_t HealthMonitor::healthState = SYS_NORMAL;
|
2026-04-27 00:42:26 +08:00
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
// 单例
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
HealthMonitor& HealthMonitor::instance()
|
2026-04-27 00:42:26 +08:00
|
|
|
|
{
|
2026-05-31 03:20:26 +08:00
|
|
|
|
static HealthMonitor inst;
|
|
|
|
|
|
return inst;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
// 初始化
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
void HealthMonitor::init()
|
|
|
|
|
|
{
|
|
|
|
|
|
m_mutex = xSemaphoreCreateMutex();
|
|
|
|
|
|
if (m_mutex == NULL) {
|
2026-04-27 00:42:26 +08:00
|
|
|
|
DEBUG_ERR("HEALTH", "创建健康状态互斥锁失败");
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
// 获取当前健康状态
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
health_state_t HealthMonitor::getCurrent()
|
2026-04-27 02:34:47 +08:00
|
|
|
|
{
|
|
|
|
|
|
health_state_t state;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (m_mutex) {
|
|
|
|
|
|
xSemaphoreTake(m_mutex, portMAX_DELAY);
|
2026-04-27 02:34:47 +08:00
|
|
|
|
}
|
2026-05-31 03:20:26 +08:00
|
|
|
|
state = healthState;
|
|
|
|
|
|
if (m_mutex) {
|
|
|
|
|
|
xSemaphoreGive(m_mutex);
|
2026-04-27 02:34:47 +08:00
|
|
|
|
}
|
|
|
|
|
|
return state;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
float HealthMonitor::getLastTemperature()
|
2026-05-05 03:28:44 +08:00
|
|
|
|
{
|
2026-05-31 03:20:26 +08:00
|
|
|
|
return m_lastTemperature;
|
2026-05-05 03:28:44 +08:00
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
bool HealthMonitor::getLastWifiRssi(int8_t *out_rssi, uint32_t *out_age_ms)
|
2026-05-05 03:28:44 +08:00
|
|
|
|
{
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (m_lastWifiRssiMs <= 0) {
|
2026-05-05 21:55:29 +08:00
|
|
|
|
return false;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int64_t now_ms = esp_timer_get_time() / 1000;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
uint32_t age_ms = (now_ms > m_lastWifiRssiMs) ? (uint32_t)(now_ms - m_lastWifiRssiMs) : 0;
|
2026-05-05 21:55:29 +08:00
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (out_rssi) *out_rssi = m_lastWifiRssi;
|
2026-05-05 21:55:29 +08:00
|
|
|
|
if (out_age_ms) *out_age_ms = age_ms;
|
|
|
|
|
|
return true;
|
2026-05-05 03:28:44 +08:00
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
2026-04-27 02:34:47 +08:00
|
|
|
|
// 温度监控(定时器回调)
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
void HealthMonitor::tempMonitorCallback(void *arg)
|
2026-04-27 02:34:47 +08:00
|
|
|
|
{
|
2026-05-31 03:20:26 +08:00
|
|
|
|
auto& self = instance();
|
2026-04-27 02:34:47 +08:00
|
|
|
|
float temp = 0.0f;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
esp_err_t err = temperature_sensor_get_celsius(self.m_tempHandle, &temp);
|
2026-04-27 02:34:47 +08:00
|
|
|
|
if (err != ESP_OK) {
|
|
|
|
|
|
DEBUG_WARN("TEMP", "读取温度失败: %s", esp_err_to_name(err));
|
|
|
|
|
|
return;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
self.m_lastTemperature = temp;
|
2026-04-27 02:34:47 +08:00
|
|
|
|
|
|
|
|
|
|
if (temp >= TEMP_CRITICAL) {
|
2026-05-31 03:20:26 +08:00
|
|
|
|
healthState = SYS_SAFE_SHUTDOWN;
|
2026-04-28 01:06:17 +08:00
|
|
|
|
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");
|
2026-04-27 02:34:47 +08:00
|
|
|
|
} else if (temp >= TEMP_WARNING) {
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (healthState < SYS_WARNING) {
|
|
|
|
|
|
healthState = SYS_WARNING;
|
2026-04-27 02:34:47 +08:00
|
|
|
|
DEBUG_WARN("TEMP", "警告: 温度 %.1f°C >= %d°C", temp, TEMP_WARNING);
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_WARN("TEMP", "健康状态变更为: SYS_WARNING");
|
2026-04-27 02:34:47 +08:00
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
// FreeRTOS 任务入口
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
void HealthMonitor::taskEntry(void *pvParameters)
|
2026-04-27 02:34:47 +08:00
|
|
|
|
{
|
2026-05-31 03:20:26 +08:00
|
|
|
|
auto& self = instance();
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_INFO("MON", "=== 健康监控任务启动 ===");
|
2026-04-27 02:34:47 +08:00
|
|
|
|
|
|
|
|
|
|
temperature_sensor_config_t temp_config = TEMPERATURE_SENSOR_CONFIG_DEFAULT(20, 80);
|
2026-05-31 03:20:26 +08:00
|
|
|
|
esp_err_t err = temperature_sensor_install(&temp_config, &self.m_tempHandle);
|
2026-04-27 02:34:47 +08:00
|
|
|
|
if (err != ESP_OK) {
|
|
|
|
|
|
DEBUG_ERR("MON", "温度传感器初始化失败: %s", esp_err_to_name(err));
|
|
|
|
|
|
} else {
|
2026-05-31 03:20:26 +08:00
|
|
|
|
ESP_ERROR_CHECK(temperature_sensor_enable(self.m_tempHandle));
|
2026-04-27 02:34:47 +08:00
|
|
|
|
DEBUG_INFO("MON", "温度传感器已启用 (范围 20°C ~ 80°C)");
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-05 03:28:44 +08:00
|
|
|
|
// 定时器,每1秒读取一次温度
|
2026-04-27 02:34:47 +08:00
|
|
|
|
esp_timer_handle_t temp_timer;
|
|
|
|
|
|
esp_timer_create_args_t timer_args = {};
|
2026-05-31 03:20:26 +08:00
|
|
|
|
timer_args.callback = &tempMonitorCallback;
|
2026-04-27 02:34:47 +08:00
|
|
|
|
timer_args.arg = nullptr;
|
|
|
|
|
|
timer_args.name = "temp_monitor";
|
|
|
|
|
|
timer_args.dispatch_method = ESP_TIMER_TASK;
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
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");
|
|
|
|
|
|
}
|
2026-04-28 01:06:17 +08:00
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// 周期性巡检
|
2026-04-27 02:34:47 +08:00
|
|
|
|
while (1) {
|
2026-04-28 01:06:17 +08:00
|
|
|
|
uint32_t free_heap = esp_get_free_heap_size();
|
|
|
|
|
|
if (free_heap < HEAP_WARNING_MIN) {
|
2026-05-31 03:20:26 +08:00
|
|
|
|
DEBUG_WARN("MON", "堆内存不足: %lu < %lu", free_heap, HEAP_WARNING_MIN);
|
2026-04-28 01:06:17 +08:00
|
|
|
|
}
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (healthState >= SYS_CRITICAL) {
|
|
|
|
|
|
DEBUG_ERR("MON", "健康状态危险: %d", healthState);
|
2026-05-05 03:28:44 +08:00
|
|
|
|
}
|
2026-04-27 02:34:47 +08:00
|
|
|
|
vTaskDelay(pdMS_TO_TICKS(5000));
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
2026-04-27 02:34:47 +08:00
|
|
|
|
// 健康事件日志
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
void HealthMonitor::logEvent(const char *event, int value)
|
2026-04-27 02:34:47 +08:00
|
|
|
|
{
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_INFO("HEALTH", "[健康事件] %s: %d", event, value);
|
2026-04-27 02:34:47 +08:00
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
// OTA 健康检查
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
health_status_t HealthMonitor::checkBeforeOta()
|
2026-04-27 00:42:26 +08:00
|
|
|
|
{
|
|
|
|
|
|
health_status_t status = HEALTH_MON_OK;
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (healthState >= SYS_SAFE_SHUTDOWN) {
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_WARN("HEALTH", "OTA 前检查: SYS_SAFE_SHUTDOWN,禁止传输");
|
2026-04-27 00:42:26 +08:00
|
|
|
|
return HEALTH_MON_SHUTDOWN;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
} else if (healthState >= SYS_CRITICAL) {
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_WARN("HEALTH", "OTA 前检查: SYS_CRITICAL,禁止传输");
|
2026-04-27 00:42:26 +08:00
|
|
|
|
return HEALTH_MON_CRITICAL;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
} else if (healthState >= SYS_WARNING) {
|
2026-04-27 00:42:26 +08:00
|
|
|
|
status = HEALTH_MON_WARNING;
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_WARN("HEALTH", "OTA 前检查: SYS_WARNING,继续传输但需注意");
|
2026-04-27 00:42:26 +08:00
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (!WifiOta::isConnected()) {
|
2026-05-05 21:55:29 +08:00
|
|
|
|
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;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (getLastWifiRssi(&last_rssi, &age_ms)) {
|
2026-05-05 21:55:29 +08:00
|
|
|
|
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);
|
2026-04-27 00:42:26 +08:00
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-04-28 01:06:17 +08:00
|
|
|
|
if (status == HEALTH_MON_OK) {
|
|
|
|
|
|
DEBUG_DBG("HEALTH", "OTA 前检查: 正常");
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-04-27 00:42:26 +08:00
|
|
|
|
return status;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
health_status_t HealthMonitor::checkBeforeOtaRelay()
|
2026-05-30 20:04:04 +08:00
|
|
|
|
{
|
|
|
|
|
|
health_status_t status = HEALTH_MON_OK;
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (healthState >= SYS_SAFE_SHUTDOWN) {
|
2026-05-30 20:04:04 +08:00
|
|
|
|
DEBUG_WARN("HEALTH", "OTA 中继检查: SYS_SAFE_SHUTDOWN,禁止传输");
|
|
|
|
|
|
return HEALTH_MON_SHUTDOWN;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
} else if (healthState >= SYS_CRITICAL) {
|
2026-05-30 20:04:04 +08:00
|
|
|
|
/* 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 g_health_state 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");
|
2026-05-31 03:20:26 +08:00
|
|
|
|
} else if (healthState >= SYS_WARNING) {
|
2026-05-30 20:04:04 +08:00
|
|
|
|
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;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
// ============================================================
|
|
|
|
|
|
// WiFi RSSI 更新
|
|
|
|
|
|
// ============================================================
|
|
|
|
|
|
health_status_t HealthMonitor::updateFromWifiRssi(int8_t rssi)
|
2026-04-27 00:42:26 +08:00
|
|
|
|
{
|
|
|
|
|
|
if (rssi < WIFI_RSSI_CRITICAL) {
|
2026-05-31 03:20:26 +08:00
|
|
|
|
healthState = SYS_CRITICAL;
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_WARN("HEALTH", "WiFi RSSI 危险: %d dBm < %d dBm", rssi, WIFI_RSSI_CRITICAL);
|
2026-05-31 03:20:26 +08:00
|
|
|
|
logEvent("WIFI危险", rssi);
|
2026-04-27 00:42:26 +08:00
|
|
|
|
return HEALTH_MON_CRITICAL;
|
|
|
|
|
|
} else if (rssi < WIFI_RSSI_NORMAL) {
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (healthState < SYS_WARNING) {
|
|
|
|
|
|
healthState = SYS_WARNING;
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_WARN("HEALTH", "WiFi RSSI 警告: %d dBm < %d dBm", rssi, WIFI_RSSI_NORMAL);
|
2026-05-31 03:20:26 +08:00
|
|
|
|
logEvent("WIFI警告", rssi);
|
2026-04-27 00:42:26 +08:00
|
|
|
|
}
|
|
|
|
|
|
return HEALTH_MON_WARNING;
|
|
|
|
|
|
}
|
|
|
|
|
|
|
2026-05-31 03:20:26 +08:00
|
|
|
|
if (healthState == SYS_WARNING && rssi >= WIFI_RSSI_NORMAL) {
|
|
|
|
|
|
healthState = SYS_NORMAL;
|
2026-04-28 01:06:17 +08:00
|
|
|
|
DEBUG_INFO("HEALTH", "WiFi RSSI 恢复: %d dBm >= %d dBm", rssi, WIFI_RSSI_NORMAL);
|
2026-05-31 03:20:26 +08:00
|
|
|
|
logEvent("WIFI恢复", rssi);
|
2026-04-27 00:42:26 +08:00
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
return HEALTH_MON_OK;
|
2026-05-31 03:20:26 +08:00
|
|
|
|
}
|