feat(monitor): F6 日志告警引擎 — AlertEngine 支持 type=log 规则查询Loki/ES+触发/恢复 + 数据源认证升级(none/basic/bearer/custom) + 从查询创建告警真实API对接
This commit is contained in:
1 parent
894cb16a05
commit
6bba6144f3
17 files changed
+657
-74
No files matched your search
@@ -11,7 +11,6 @@ import (
|
||||
"log"
|
||||
"net/http"
|
||||
"net/url"
|
||||
"strconv"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
@@ -67,6 +66,7 @@ func (e *AlertEngine) Check() {
|
||||
e.checkPortAlerts()
|
||||
e.checkLogAlertRules()
|
||||
e.autoResolveAlerts()
|
||||
e.checkLogAlertRecovery()
|
||||
}
|
||||
|
||||
// tryAcquireLock 尝试获取分布式锁,返回是否获得 Leader 角色
|
||||
@@ -386,6 +386,143 @@ func (e *AlertEngine) checkPortAlerts() {
|
||||
}
|
||||
}
|
||||
|
||||
// checkLogAlertRules 执行 type == "log" 的告警规则
|
||||
// 通过数据源查询 Loki/ES,结果条数 > 0 即触发。// Logan.Li
|
||||
func (e *AlertEngine) checkLogAlertRules() {
|
||||
if e.ds == nil {
|
||||
return
|
||||
}
|
||||
var rules []model.AlertRule
|
||||
e.db.Where("enabled = ? AND type = ?", true, "log").Find(&rules)
|
||||
|
||||
for _, rule := range rules {
|
||||
ds, err := loadLogDS(e.ds, rule.DatasourceID)
|
||||
if err != nil || ds.Type != rule.Source {
|
||||
log.Printf("[AlertEngine] 日志告警规则 %s 数据源无效: id=%d source=%s", rule.Name, rule.DatasourceID, rule.Source)
|
||||
continue
|
||||
}
|
||||
|
||||
count, err := e.queryLogDataSource(ds, rule.Expression)
|
||||
if err != nil {
|
||||
log.Printf("[AlertEngine] 日志告警规则 %s 查询失败: %v", rule.Name, err)
|
||||
continue
|
||||
}
|
||||
|
||||
// 简单策略:查询结果数量 > 0 即触发(后续可扩展为 count_over_time 阈值)// Logan.Li
|
||||
if count == 0 {
|
||||
continue
|
||||
}
|
||||
|
||||
// 避免重复告警:最近 10 分钟已有同规则未恢复事件,不重复
|
||||
var existing model.AlertEvent
|
||||
tenMinAgo := time.Now().Add(-10 * time.Minute)
|
||||
e.db.Where("rule_id = ? AND status = ? AND started_at > ?", rule.ID, model.AlertEventFiring, tenMinAgo).First(&existing)
|
||||
if existing.ID > 0 {
|
||||
continue
|
||||
}
|
||||
|
||||
if e.isSilenced(rule.Name, "", rule.ID) {
|
||||
continue
|
||||
}
|
||||
|
||||
now := time.Now()
|
||||
alert := model.AlertEvent{
|
||||
RuleID: rule.ID,
|
||||
RuleName: rule.Name,
|
||||
Severity: rule.Severity,
|
||||
Status: string(model.AlertEventFiring),
|
||||
Target: fmt.Sprintf("%s/%s", ds.Type, ds.Name),
|
||||
Message: fmt.Sprintf("日志告警触发: %s 在 %s 中匹配到 %d 条日志", rule.Name, ds.Name, count),
|
||||
Value: float64(count),
|
||||
Threshold: 0,
|
||||
StartedAt: now,
|
||||
LastEvalAt: now,
|
||||
}
|
||||
e.db.Create(&alert)
|
||||
|
||||
e.notifyAlertRule(rule.Name, alert.Target, "log", rule.Severity, alert.Message, float64(count), 0)
|
||||
|
||||
e.db.Model(&rule).Update("last_triggered", now)
|
||||
}
|
||||
}
|
||||
|
||||
// queryLogDataSource 根据数据源类型查询日志条数 // Logan.Li
|
||||
func (e *AlertEngine) queryLogDataSource(ds *model.LogDataSource, expression string) (int, error) {
|
||||
end := time.Now()
|
||||
start := end.Add(-5 * time.Minute)
|
||||
client := newLogHTTPClient(ds.SkipTLSVerify, 30*time.Second)
|
||||
|
||||
switch strings.ToLower(ds.Type) {
|
||||
case "loki":
|
||||
params := url.Values{}
|
||||
params.Set("query", expression)
|
||||
params.Set("limit", "1")
|
||||
params.Set("direction", "backward")
|
||||
params.Set("start", fmt.Sprintf("%d", start.UnixNano()))
|
||||
params.Set("end", fmt.Sprintf("%d", end.UnixNano()))
|
||||
target := ds.URL + "/loki/api/v1/query_range?" + params.Encode()
|
||||
req, _ := http.NewRequest(http.MethodGet, target, nil)
|
||||
applyLogDSAuthV2(req, ds)
|
||||
resp, err := client.Do(req)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
defer resp.Body.Close()
|
||||
body, _ := io.ReadAll(resp.Body)
|
||||
if resp.StatusCode >= 400 {
|
||||
return 0, fmt.Errorf("HTTP %s: %s", resp.Status, string(body))
|
||||
}
|
||||
var raw map[string]interface{}
|
||||
_ = json.Unmarshal(body, &raw)
|
||||
entries := flattenLokiStreams(raw)
|
||||
return len(entries), nil
|
||||
case "elasticsearch":
|
||||
searchBody := map[string]interface{}{
|
||||
"size": 0,
|
||||
"query": map[string]interface{}{
|
||||
"bool": map[string]interface{}{
|
||||
"must": []interface{}{
|
||||
map[string]interface{}{"query_string": map[string]interface{}{"query": defaultESQuery(expression)}},
|
||||
},
|
||||
},
|
||||
},
|
||||
}
|
||||
rangeCond := map[string]interface{}{
|
||||
"bool": map[string]interface{}{
|
||||
"should": []interface{}{
|
||||
map[string]interface{}{"range": map[string]interface{}{"@timestamp": map[string]string{"gte": start.Format(time.RFC3339Nano), "lte": end.Format(time.RFC3339Nano)}}},
|
||||
map[string]interface{}{"range": map[string]interface{}{"timestamp": map[string]string{"gte": start.Format(time.RFC3339Nano), "lte": end.Format(time.RFC3339Nano)}}},
|
||||
},
|
||||
"minimum_should_match": 1,
|
||||
},
|
||||
}
|
||||
searchBody["query"].(map[string]interface{})["bool"].(map[string]interface{})["must"] = append(
|
||||
searchBody["query"].(map[string]interface{})["bool"].(map[string]interface{})["must"].([]interface{}),
|
||||
rangeCond,
|
||||
)
|
||||
payload, _ := json.Marshal(searchBody)
|
||||
target := fmt.Sprintf("%s/%s/_search", ds.URL, "*")
|
||||
req, _ := http.NewRequest(http.MethodPost, target, bytes.NewReader(payload))
|
||||
req.Header.Set("Content-Type", "application/json")
|
||||
applyLogDSAuthV2(req, ds)
|
||||
resp, err := client.Do(req)
|
||||
if err != nil {
|
||||
return 0, err
|
||||
}
|
||||
defer resp.Body.Close()
|
||||
body, _ := io.ReadAll(resp.Body)
|
||||
if resp.StatusCode >= 400 {
|
||||
return 0, fmt.Errorf("HTTP %s: %s", resp.Status, string(body))
|
||||
}
|
||||
var raw map[string]interface{}
|
||||
_ = json.Unmarshal(body, &raw)
|
||||
total := esHitsTotal(raw)
|
||||
return int(total), nil
|
||||
default:
|
||||
return 0, fmt.Errorf("未知日志数据源类型: %s", ds.Type)
|
||||
}
|
||||
}
|
||||
|
||||
// 自动恢复告警(指标恢复正常)
|
||||
func (e *AlertEngine) autoResolveAlerts() {
|
||||
var alerts []model.AgentAlert
|
||||
@@ -478,6 +615,43 @@ func (e *AlertEngine) autoResolveAlerts() {
|
||||
}
|
||||
}
|
||||
|
||||
// checkLogAlertRecovery 恢复已解决的日志告警事件 // Logan.Li
|
||||
func (e *AlertEngine) checkLogAlertRecovery() {
|
||||
if e.ds == nil {
|
||||
return
|
||||
}
|
||||
var events []model.AlertEvent
|
||||
e.db.Where("status = ?", model.AlertEventFiring).Find(&events)
|
||||
|
||||
for _, evt := range events {
|
||||
if evt.RuleID == 0 {
|
||||
continue
|
||||
}
|
||||
var rule model.AlertRule
|
||||
if err := e.db.First(&rule, evt.RuleID).Error; err != nil || rule.Type != "log" {
|
||||
continue
|
||||
}
|
||||
ds, err := loadLogDS(e.ds, rule.DatasourceID)
|
||||
if err != nil {
|
||||
continue
|
||||
}
|
||||
count, err := e.queryLogDataSource(ds, rule.Expression)
|
||||
if err != nil {
|
||||
continue
|
||||
}
|
||||
// 连续两个周期无日志匹配则恢复(简化策略)// Logan.Li
|
||||
if count > 0 {
|
||||
continue
|
||||
}
|
||||
now := time.Now()
|
||||
e.db.Model(&evt).Updates(map[string]interface{}{
|
||||
"status": model.AlertEventResolved,
|
||||
"resolved_at": now,
|
||||
})
|
||||
e.notifyRecovery(rule.Name, evt.Target, "log", evt.Message)
|
||||
}
|
||||
}
|
||||
|
||||
func (e *AlertEngine) compareOperator(op string, val, threshold float64) bool {
|
||||
switch op {
|
||||
case ">":
|
||||
|
||||
Reference in new issue
Block a user