fix(prometheus_adapter): 改进服务进程管理和告警表达式生成

JaD1ng · JaD1ng · commit 4555e0663269 · 2025-09-29T10:19:34.000+08:00
- 在deploy.sh和build.sh中添加PID文件管理，优化服务启动和停止流程
- 修复alert_service.go中告警表达式生成的标签处理逻辑
- 使用%g代替%f格式化浮点数以避免科学计数法显示
diff --git a/internal/prometheus_adapter/service/alert_service.go b/internal/prometheus_adapter/service/alert_service.go
@@ -402,7 +402,7 @@ func (s *AlertService) buildPrometheusRules(rules []model.AlertRule, ruleMetas [
 		// 构建注释
 		annotations := map[string]string{
 			"description": rule.Description,
-			"summary":     fmt.Sprintf("%s %s %f", rule.Expr, rule.Op, meta.Threshold),
+			"summary":     fmt.Sprintf("%s %s %g", rule.Expr, rule.Op, meta.Threshold),
 		}
 
 		// 计算for字段
@@ -475,29 +475,54 @@ func (s *AlertService) buildExpression(rule *model.AlertRule, meta *model.AlertR
 		if len(labelMatchers) > 0 {
 			// 如果表达式包含{，说明已经有标签选择器
 			if strings.Contains(expr, "{") {
-				expr = strings.Replace(expr, "}", ","+strings.Join(labelMatchers, ",")+"}", 1)
-			} else {
-				// 在指标名后添加标签选择器
-				// 查找第一个非字母数字下划线的字符
-				metricEnd := 0
-				for i, ch := range expr {
-					if !((ch >= 'a' && ch <= 'z') || (ch >= 'A' && ch <= 'Z') ||
-						(ch >= '0' && ch <= '9') || ch == '_') {
-						metricEnd = i
-						break
+				// 查找第一个 { 后的内容
+				start := strings.Index(expr, "{")
+				end := strings.Index(expr[start:], "}")
+				if end != -1 {
+					end += start
+					existingLabels := strings.TrimSpace(expr[start+1 : end])
+					if existingLabels == "" {
+						// 空的标签选择器，直接替换
+						expr = expr[:start+1] + strings.Join(labelMatchers, ",") + expr[end:]
+					} else {
+						// 已有标签，需要检查是否重复
+						existingLabelMap := make(map[string]bool)
+						// 解析现有标签
+						labelPairs := strings.Split(existingLabels, ",")
+						for _, pair := range labelPairs {
+							if strings.Contains(pair, "=") {
+								key := strings.TrimSpace(strings.Split(pair, "=")[0])
+								if key != "" {
+									existingLabelMap[key] = true
+								}
+							}
+						}
+						// 只添加不重复的标签
+						newLabels := []string{}
+						for k, v := range labels {
+							if !existingLabelMap[k] && k != "" && v != "" {
+								newLabels = append(newLabels, fmt.Sprintf(`%s="%s"`, k, v))
+							}
+						}
+						if len(newLabels) > 0 {
+							expr = expr[:end] + "," + strings.Join(newLabels, ",") + expr[end:]
+						}
 					}
 				}
-				if metricEnd == 0 {
-					metricEnd = len(expr)
+			} else {
+				// 对于没有标签的简单指标，只处理单个单词的情况
+				// 如果表达式包含空格、括号等，不进行标签注入
+				if !strings.ContainsAny(expr, " ()[]{}") {
+					// 只有单个指标名，可以安全添加标签
+					expr = expr + "{" + strings.Join(labelMatchers, ",") + "}"
 				}
-				expr = expr[:metricEnd] + "{" + strings.Join(labelMatchers, ",") + "}" + expr[metricEnd:]
 			}
 		}
 	}
 
 	// 添加比较操作符和阈值
 	if meta.Threshold != 0 {
-		expr = fmt.Sprintf("%s %s %f", expr, rule.Op, meta.Threshold)
+		expr = fmt.Sprintf("%s %s %g", expr, rule.Op, meta.Threshold)
 	}
 
 	return expr
diff --git a/scripts/prometheus_adapter/build.sh b/scripts/prometheus_adapter/build.sh
@@ -118,13 +118,26 @@ cat > "$BUILD_DIR/start.sh" << 'EOF'
 SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
 BIN_PATH="$SCRIPT_DIR/bin/prometheus_adapter"
 CONFIG_FILE="$SCRIPT_DIR/config/prometheus_adapter.yml"
+PID_FILE="$SCRIPT_DIR/prometheus_adapter.pid"
+LOG_FILE="$SCRIPT_DIR/prometheus_adapter.log"
 
 # 检查二进制文件
 if [ ! -f "$BIN_PATH" ]; then
     echo "错误: 找不到可执行文件 $BIN_PATH"
     exit 1
 fi
 
+# 检查是否已在运行
+if [ -f "$PID_FILE" ]; then
+    PID=$(cat "$PID_FILE")
+    if kill -0 "$PID" 2>/dev/null; then
+        echo "Prometheus Adapter已在运行 (PID: $PID)"
+        exit 1
+    else
+        rm -f "$PID_FILE"
+    fi
+fi
+
 # 检查配置文件
 if [ -f "$CONFIG_FILE" ]; then
     echo "使用配置文件: $CONFIG_FILE"
@@ -140,11 +153,23 @@ fi
 
 echo "启动 Prometheus Adapter..."
 
-# 切换到 bin 目录，以便程序能正确找到相对路径的配置文件
+# 切换到脚本目录
 cd "$SCRIPT_DIR"
 
-# 启动服务
-exec "$BIN_PATH"
+# 后台启动服务
+nohup "$BIN_PATH" > "$LOG_FILE" 2>&1 &
+PID=$!
+
+# 保存PID
+echo $PID > "$PID_FILE"
+
+echo "Prometheus Adapter已启动"
+echo "PID: $PID"
+echo "日志文件: $LOG_FILE"
+echo "PID文件: $PID_FILE"
+echo ""
+echo "查看日志: tail -f $LOG_FILE"
+echo "停止服务: ./stop.sh"
 EOF
 chmod +x "$BUILD_DIR/start.sh"
 
@@ -155,26 +180,54 @@ cat > "$BUILD_DIR/stop.sh" << 'EOF'
 
 # Prometheus Adapter 停止脚本
 
+# 获取脚本所在目录
+SCRIPT_DIR=$(cd "$(dirname "$0")" && pwd)
+PID_FILE="$SCRIPT_DIR/prometheus_adapter.pid"
 APP_NAME="prometheus_adapter"
 
-# 查找进程
-PID=$(ps aux | grep -v grep | grep "$APP_NAME" | awk '{print $2}')
+# 优先从PID文件读取
+if [ -f "$PID_FILE" ]; then
+    PID=$(cat "$PID_FILE" 2>/dev/null)
+    if [ -n "$PID" ] && kill -0 "$PID" 2>/dev/null; then
+        echo "从PID文件获取进程ID: $PID"
+    else
+        echo "PID文件中的进程已不存在，清理PID文件"
+        rm -f "$PID_FILE"
+        PID=""
+    fi
+else
+    PID=""
+fi
+
+# 如果PID文件不存在或进程已死，通过进程名查找
+if [ -z "$PID" ]; then
+    PID=$(ps aux | grep -v grep | grep "$APP_NAME" | awk '{print $2}')
+fi
 
 if [ -z "$PID" ]; then
     echo "没有找到运行中的 $APP_NAME 进程"
     exit 0
 fi
 
 echo "停止 $APP_NAME (PID: $PID)..."
-kill -TERM $PID
+kill -TERM $PID 2>/dev/null || true
 
 # 等待进程退出
-sleep 2
+count=0
+while [ $count -lt 10 ] && ps -p "$PID" > /dev/null 2>&1; do
+    sleep 1
+    count=$((count + 1))
+done
+
+# 检查是否已退出
+if ps -p "$PID" > /dev/null 2>&1; then
+    echo "强制停止 $APP_NAME..."
+    kill -KILL "$PID" 2>/dev/null || true
+fi
 
-# 检查是否还在运行
-if ps -p $PID > /dev/null 2>&1; then
-    echo "强制停止进程..."
-    kill -KILL $PID
+# 清理PID文件
+if [ -f "$PID_FILE" ]; then
+    rm -f "$PID_FILE"
 fi
 
 echo "$APP_NAME 已停止"
diff --git a/scripts/prometheus_adapter/deploy.sh b/scripts/prometheus_adapter/deploy.sh
@@ -132,6 +132,16 @@ fi
 
 # 检查是否有运行中的服务
 check_running_service() {
+    # 优先从PID文件读取
+    if [ -f "$DEPLOY_DIR/prometheus_adapter.pid" ]; then
+        local pid=$(cat "$DEPLOY_DIR/prometheus_adapter.pid" 2>/dev/null)
+        if [ -n "$pid" ] && kill -0 "$pid" 2>/dev/null; then
+            echo "$pid"
+            return
+        fi
+    fi
+
+    # 如果PID文件不存在或进程已死，通过进程名查找
     local pid=$(ps aux | grep -v grep | grep "prometheus_adapter" | grep -v "$0" | awk '{print $2}')
     if [ -n "$pid" ]; then
         echo "$pid"
@@ -158,6 +168,11 @@ stop_service() {
             kill -KILL "$pid" 2>/dev/null || true
         fi
 
+        # 清理PID文件
+        if [ -f "$DEPLOY_DIR/prometheus_adapter.pid" ]; then
+            rm -f "$DEPLOY_DIR/prometheus_adapter.pid"
+        fi
+
         log_info "服务已停止"
     fi
 }
@@ -340,17 +355,27 @@ if [ "$START_SERVICE" = true ] || [ "$RESTART_SERVICE" = true ]; then
 
     # 启动服务
     cd "$DEPLOY_DIR"
-    nohup ./start.sh > prometheus_adapter.log 2>&1 &
+
+    # 直接启动二进制文件而不是通过start.sh脚本
+    nohup ./bin/prometheus_adapter > prometheus_adapter.log 2>&1 &
+    PID=$!
+
+    # 保存PID到文件
+    echo $PID > prometheus_adapter.pid
+
+    log_info "服务已启动 (PID: $PID)"
+    echo "PID文件: $DEPLOY_DIR/prometheus_adapter.pid"
+    echo "日志文件: $DEPLOY_DIR/prometheus_adapter.log"
 
     # 等待服务启动
     sleep 2
 
     # 检查是否启动成功
-    NEW_PID=$(check_running_service)
-    if [ -n "$NEW_PID" ]; then
-        log_info "服务已启动 (PID: $NEW_PID)"
+    if kill -0 "$PID" 2>/dev/null; then
+        log_info "服务启动成功，正在运行"
         echo ""
         echo "查看日志: tail -f $DEPLOY_DIR/prometheus_adapter.log"
+        echo "停止服务: kill \$(cat $DEPLOY_DIR/prometheus_adapter.pid)"
     else
         log_error "服务启动失败，请检查日志"
         exit 1
@@ -359,10 +384,11 @@ else
     echo ""
     echo "手动启动服务:"
     echo "  cd $DEPLOY_DIR"
-    echo "  ./start.sh"
+    echo "  nohup ./bin/prometheus_adapter > prometheus_adapter.log 2>&1 &"
+    echo "  echo \$! > prometheus_adapter.pid"
     echo ""
-    echo "或使用后台模式:"
-    echo "  nohup ./start.sh > prometheus_adapter.log 2>&1 &"
+    echo "停止服务:"
+    echo "  kill \$(cat prometheus_adapter.pid)"
 fi
 
 log_info "部署完成!"
