rlwj-lab/aiops
0
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
de6f9454fcb06af7710bd9866749049179d63ef3
aiops/backend/controller/alert_metric_catalog.go
T
LoganandClaude 471353e3d7 feat: 监控告警体系统一重构 + 自动化任务调度引擎 + 监控大屏拆分 
告警中心
- 新增统一告警(alert_unified)、监控告警规则(monitor_alert_rule)、指标目录(alert_metric_catalog)
- 通知策略与消息分发(alert_notify_policy/dispatch/message)、静默规则
- 告警引擎按阈值/SNMP 拆分(alert_engine_monitor/snmp)

监控采集
- SNMP 常量/模板与历史指标(snmp templates/constants、snmp_metrics_history、snmp_collect)
- 中间件监控 Redis 采集与种子(middleware_monitor_redis/seed)
- 站点探测 pkg(siteprobe) + 站点视图(site_view)

自动化任务
- cron 调度器(cronscheduler)、脚本运行器(scriptrunner)、系统任务(systask)及运行记录(cron_job_run、system_task)

审计与凭证
- 审计设置(audit_setting) + 历史清理(auditcleaner)

前端
- 监控页面(host/middleware/site/snmp)拆分为子模块并大屏化
- 告警中心多 Tab(规则/指标目录/通知策略/静默)、任务管理组件重构、Agent CLI Markdown 渲染
- 移除旧脚本卡片/详情/日志组件，统一为 ScriptTable/ScriptCodeEditor/TerminalOutput

Co-Authored-By: Claude <noreply@anthropic.com>
2026-07-01 23:44:45 +08:00

107 lines
12 KiB
Go
Raw Blame History

package controller
import (
"net/http"
"ezops/model"
"github.com/gin-gonic/gin"
)
// 指标目录静态定义，供监控中心与告警中心共用 // Logan.Li
// MetricCatalogItem 可采集/可告警指标定义 // Logan.Li
type MetricCatalogItem struct {
Key string `json:"key"`
Label string `json:"label"`
Unit string `json:"unit"`
Type string `json:"type"` // gauge / counter / status
DefaultOp string `json:"default_op"`
DefaultWarn float64 `json:"default_warn"`
DefaultCrit float64 `json:"default_crit"`
Min float64 `json:"min"`
Max float64 `json:"max"`
CollectSource string `json:"collect_source"`
Description string `json:"description"`
Operators []string `json:"operators"`
}
// MetricCatalogCategory 指标分类 // Logan.Li
type MetricCatalogCategory struct {
ID string `json:"id"`
Label string `json:"label"`
Description string `json:"description"`
Metrics []MetricCatalogItem `json:"metrics"`
}
var metricCatalog = []MetricCatalogCategory{
{
ID: "host", Label: "主机监控", Description: "Agent 主动采集 /proc 指标，默认 10s 上报",
Metrics: []MetricCatalogItem{
{Key: "cpu", Label: "CPU 使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 80, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "agent", Description: "整体 CPU 占用百分比", Operators: []string{">", ">=", "<", "<="}},
{Key: "cpu_iowait", Label: "CPU IO 等待", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 20, DefaultCrit: 40, Min: 0, Max: 100, CollectSource: "agent", Description: "iowait 占比过高表示磁盘瓶颈", Operators: []string{">", ">="}},
{Key: "mem", Label: "内存使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 80, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "agent", Description: "物理内存占用百分比", Operators: []string{">", ">=", "<", "<="}},
{Key: "mem_available", Label: "可用内存", Unit: "GB", Type: "gauge", DefaultOp: "<", DefaultWarn: 1, DefaultCrit: 0.5, Min: 0, Max: 1024, CollectSource: "agent", Description: "剩余可用内存低于阈值告警", Operators: []string{"<", "<="}},
{Key: "swap", Label: "Swap 使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 50, DefaultCrit: 80, Min: 0, Max: 100, CollectSource: "agent", Description: "Swap 使用过高影响性能", Operators: []string{">", ">="}},
{Key: "disk", Label: "磁盘使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 85, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "agent", Description: "各挂载点最大使用率", Operators: []string{">", ">="}},
{Key: "disk_inode", Label: "Inode 使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 80, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "agent", Description: "Inode 耗尽导致无法创建文件", Operators: []string{">", ">="}},
{Key: "disk_io_util", Label: "磁盘 IO 利用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 80, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "agent", Description: "磁盘繁忙程度", Operators: []string{">", ">="}},
{Key: "load1", Label: "1分钟负载", Unit: "", Type: "gauge", DefaultOp: ">", DefaultWarn: 4, DefaultCrit: 8, Min: 0, Max: 256, CollectSource: "agent", Description: "系统 1 分钟平均负载", Operators: []string{">", ">="}},
{Key: "net_rx_bps", Label: "入网带宽", Unit: "B/s", Type: "gauge", DefaultOp: ">", DefaultWarn: 104857600, DefaultCrit: 524288000, Min: 0, Max: 1e12, CollectSource: "agent", Description: "接收字节速率", Operators: []string{">", ">="}},
{Key: "net_tx_bps", Label: "出网带宽", Unit: "B/s", Type: "gauge", DefaultOp: ">", DefaultWarn: 104857600, DefaultCrit: 524288000, Min: 0, Max: 1e12, CollectSource: "agent", Description: "发送字节速率", Operators: []string{">", ">="}},
{Key: "net_rx_errors", Label: "入网错误", Unit: "次", Type: "counter", DefaultOp: ">", DefaultWarn: 10, DefaultCrit: 100, Min: 0, Max: 1e9, CollectSource: "agent", Description: "网卡接收错误累计", Operators: []string{">", ">="}},
{Key: "tcp_established", Label: "TCP 已建立连接", Unit: "个", Type: "gauge", DefaultOp: ">", DefaultWarn: 5000, DefaultCrit: 20000, Min: 0, Max: 1e6, CollectSource: "agent", Description: "ESTABLISHED 连接数", Operators: []string{">", ">="}},
{Key: "tcp_time_wait", Label: "TCP TIME_WAIT", Unit: "个", Type: "gauge", DefaultOp: ">", DefaultWarn: 15000, DefaultCrit: 30000, Min: 0, Max: 1e6, CollectSource: "agent", Description: "TIME_WAIT 过多可能端口耗尽", Operators: []string{">", ">="}},
{Key: "tcp_close_wait", Label: "TCP CLOSE_WAIT", Unit: "个", Type: "gauge", DefaultOp: ">", DefaultWarn: 200, DefaultCrit: 1000, Min: 0, Max: 1e6, CollectSource: "agent", Description: "CLOSE_WAIT 泄漏检测", Operators: []string{">", ">="}},
{Key: "fd", Label: "文件描述符使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 80, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "agent", Description: "进程 FD 占用/system limit", Operators: []string{">", ">="}},
{Key: "offline", Label: "主机离线", Unit: "分钟", Type: "status", DefaultOp: ">", DefaultWarn: 3, DefaultCrit: 5, Min: 0, Max: 1440, CollectSource: "agent", Description: "超过 N 分钟无上报", Operators: []string{">", ">="}},
{Key: "port_down", Label: "端口不通", Unit: "", Type: "status", DefaultOp: "==", DefaultWarn: 0, DefaultCrit: 0, Min: 0, Max: 65535, CollectSource: "agent", Description: "监听端口消失", Operators: []string{"=="}},
{Key: "health_probe", Label: "健康探针失败", Unit: "", Type: "status", DefaultOp: "==", DefaultWarn: 1, DefaultCrit: 1, Min: 0, Max: 1, CollectSource: "agent", Description: "HTTP/TCP/Exec 探针失败", Operators: []string{"=="}},
},
},
{
ID: "middleware", Label: "中间件监控", Description: "平台直连 Redis/MySQL 采集，默认 30s",
Metrics: []MetricCatalogItem{
{Key: "redis_connected_clients", Label: "Redis 连接数", Unit: "个", Type: "gauge", DefaultOp: ">", DefaultWarn: 500, DefaultCrit: 2000, Min: 0, Max: 1e5, CollectSource: "middleware", Description: "当前客户端连接数", Operators: []string{">", ">="}},
{Key: "redis_used_memory", Label: "Redis 内存", Unit: "bytes", Type: "gauge", DefaultOp: ">", DefaultWarn: 536870912, DefaultCrit: 1073741824, Min: 0, Max: 1e12, CollectSource: "middleware", Description: "已用内存字节数", Operators: []string{">", ">="}},
{Key: "redis_hit_rate", Label: "Redis 命中率", Unit: "ratio", Type: "gauge", DefaultOp: "<", DefaultWarn: 0.9, DefaultCrit: 0.8, Min: 0, Max: 1, CollectSource: "middleware", Description: "keyspace 命中率 (0~1)", Operators: []string{"<", "<="}},
{Key: "redis_qps", Label: "Redis QPS", Unit: "ops/s", Type: "gauge", DefaultOp: ">", DefaultWarn: 10000, DefaultCrit: 50000, Min: 0, Max: 1e7, CollectSource: "middleware", Description: "每秒命令数", Operators: []string{">", ">="}},
{Key: "redis_blocked_clients", Label: "Redis 阻塞客户端", Unit: "个", Type: "gauge", DefaultOp: ">", DefaultWarn: 1, DefaultCrit: 10, Min: 0, Max: 1e4, CollectSource: "middleware", Description: "BLPOP 等阻塞连接", Operators: []string{">", ">="}},
{Key: "redis_evicted_keys", Label: "Redis 驱逐键", Unit: "个", Type: "counter", DefaultOp: ">", DefaultWarn: 100, DefaultCrit: 1000, Min: 0, Max: 1e12, CollectSource: "middleware", Description: "内存不足驱逐 key 数", Operators: []string{">", ">="}},
{Key: "mysql_connections", Label: "MySQL 连接数", Unit: "个", Type: "gauge", DefaultOp: ">", DefaultWarn: 200, DefaultCrit: 500, Min: 0, Max: 1e4, CollectSource: "middleware", Description: "Threads_connected", Operators: []string{">", ">="}},
{Key: "mysql_threads_running", Label: "MySQL 活跃线程", Unit: "个", Type: "gauge", DefaultOp: ">", DefaultWarn: 20, DefaultCrit: 50, Min: 0, Max: 1e4, CollectSource: "middleware", Description: "正在执行的线程", Operators: []string{">", ">="}},
{Key: "mysql_slow_queries", Label: "MySQL 慢查询", Unit: "次", Type: "counter", DefaultOp: ">", DefaultWarn: 10, DefaultCrit: 100, Min: 0, Max: 1e12, CollectSource: "middleware", Description: "慢查询累计数", Operators: []string{">", ">="}},
{Key: "mysql_buffer_hit_rate", Label: "MySQL 缓冲命中率", Unit: "ratio", Type: "gauge", DefaultOp: "<", DefaultWarn: 0.95, DefaultCrit: 0.9, Min: 0, Max: 1, CollectSource: "middleware", Description: "InnoDB buffer pool 命中率 (0~1)", Operators: []string{"<", "<="}},
{Key: "mysql_qps", Label: "MySQL QPS", Unit: "ops/s", Type: "gauge", DefaultOp: ">", DefaultWarn: 5000, DefaultCrit: 20000, Min: 0, Max: 1e7, CollectSource: "middleware", Description: "每秒查询数", Operators: []string{">", ">="}},
{Key: "middleware_offline", Label: "中间件离线", Unit: "", Type: "status", DefaultOp: "==", DefaultWarn: 1, DefaultCrit: 1, Min: 0, Max: 1, CollectSource: "middleware", Description: "采集失败或不可达", Operators: []string{"=="}},
},
},
{
ID: "snmp", Label: "SNMP 监控", Description: "网络设备 / BMC 采集，默认 5 分钟周期",
Metrics: []MetricCatalogItem{
{Key: "snmp_cpu_usage", Label: "CPU 使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 80, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "snmp", Description: "各厂商 CPU 利用率最大值", Operators: []string{">", ">="}},
{Key: "snmp_mem_usage", Label: "内存使用率", Unit: "%", Type: "gauge", DefaultOp: ">", DefaultWarn: 85, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "snmp", Description: "各厂商内存利用率最大值", Operators: []string{">", ">="}},
{Key: "snmp_temperature", Label: "设备温度", Unit: "°C", Type: "gauge", DefaultOp: ">", DefaultWarn: 65, DefaultCrit: 80, Min: 0, Max: 120, CollectSource: "snmp", Description: "温度传感器读数最大值", Operators: []string{">", ">="}},
{Key: "snmp_if_in_bps", Label: "入方向流量", Unit: "bps", Type: "gauge", DefaultOp: ">", DefaultWarn: 1e9, DefaultCrit: 5e9, Min: 0, Max: 1e12, CollectSource: "snmp", Description: "接口入方向比特速率", Operators: []string{">", ">="}},
{Key: "snmp_if_out_bps", Label: "出方向流量", Unit: "bps", Type: "gauge", DefaultOp: ">", DefaultWarn: 1e9, DefaultCrit: 5e9, Min: 0, Max: 1e12, CollectSource: "snmp", Description: "接口出方向比特速率", Operators: []string{">", ">="}},
{Key: "snmp_offline", Label: "设备离线", Unit: "", Type: "status", DefaultOp: "==", DefaultWarn: 1, DefaultCrit: 1, Min: 0, Max: 1, CollectSource: "snmp", Description: "设备离线或采集异常", Operators: []string{"=="}},
},
},
{
ID: "site", Label: "站点拨测", Description: "HTTP/SSL/TCP/DNS/PING 主动拨测",
Metrics: []MetricCatalogItem{
{Key: "probe_fail", Label: "拨测失败", Unit: "", Type: "status", DefaultOp: "==", DefaultWarn: 1, DefaultCrit: 1, Min: 0, Max: 1, CollectSource: "site", Description: "连续拨测失败", Operators: []string{"=="}},
{Key: "probe_timeout", Label: "拨测超时", Unit: "", Type: "status", DefaultOp: "==", DefaultWarn: 1, DefaultCrit: 1, Min: 0, Max: 1, CollectSource: "site", Description: "请求超时", Operators: []string{"=="}},
{Key: "response_time", Label: "响应延迟", Unit: "ms", Type: "gauge", DefaultOp: ">", DefaultWarn: 500, DefaultCrit: 2000, Min: 0, Max: 60000, CollectSource: "site", Description: "拨测响应时间", Operators: []string{">", ">="}},
{Key: "availability", Label: "可用率", Unit: "%", Type: "gauge", DefaultOp: "<", DefaultWarn: 99, DefaultCrit: 95, Min: 0, Max: 100, CollectSource: "site", Description: "近 1h 成功率", Operators: []string{"<", "<="}},
{Key: "http_status_error", Label: "HTTP 状态码异常", Unit: "", Type: "status", DefaultOp: "==", DefaultWarn: 1, DefaultCrit: 1, Min: 0, Max: 1, CollectSource: "site", Description: "非期望状态码", Operators: []string{"=="}},
{Key: "ssl_expiry_days", Label: "SSL 证书剩余天数", Unit: "天", Type: "gauge", DefaultOp: "<", DefaultWarn: 30, DefaultCrit: 7, Min: -365, Max: 365, CollectSource: "site", Description: "SSL 协议拨测证书即将过期", Operators: []string{"<", "<="}},
},
},
}
// GetMetricCatalog 返回监控指标目录
// GET /api/v1/alerts/metric-catalog
func (h *AlertHandler) GetMetricCatalog(c *gin.Context) {
c.JSON(http.StatusOK, model.Response{Code: 200, Data: gin.H{"categories": metricCatalog}})
}
Reference in new issue View Git Blame Copy Permalink