rlwj-lab/aiops
0
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: dashboard metrics 慢查询优化(index+UNIX_TIMESTAMP分组) + 批量升级版本号修复

Browse Source
This commit is contained in:
attacker committed 2026-06-24 14:27:58 +00:00
1 parent c515f6ed03
commit 20f64e3851
5 files changed
+222 -19
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

backend/internal/api/handlers/agent_install.go
+11 -4
View File
@@ -6,6 +6,7 @@ import (
"net"
"net/http"
"strconv"
"strings"
"time"
"ezops/internal/models"
@@ -423,9 +424,9 @@ func (h *AgentInstallHandler) BatchUpgradeAgent(c *gin.Context) {
binaryFile = "aiops-agent-linux-arm64"
}
// 下载新二进制 + 重启服务
// 下载新二进制 + 获取版本号 + 重启服务
upgradeScript := fmt.Sprintf(
`curl -fsSL -o /tmp/aiops-agent %s/api/agent/bin/%s && chmod +x /tmp/aiops-agent && systemctl stop aiops-agent 2>/dev/null; mv /tmp/aiops-agent /usr/local/bin/aiops-agent && systemctl start aiops-agent && echo "UPGRADE_OK"`,
`curl -fsSL -o /tmp/aiops-agent %s/api/agent/bin/%s && chmod +x /tmp/aiops-agent && systemctl stop aiops-agent 2>/dev/null; mv /tmp/aiops-agent /usr/local/bin/aiops-agent && systemctl start aiops-agent; ver=$(/usr/local/bin/aiops-agent output 2>/dev/null | grep 'agent.version' | awk -F'[|[|]' '{print $2}' | tr -d ' '); echo "UPGRADE_OK ver=$ver"`,
serverURL, binaryFile,
)
@@ -437,9 +438,15 @@ func (h *AgentInstallHandler) BatchUpgradeAgent(c *gin.Context) {
continue
}
// 更新版本号
// 从输出提取版本号
agentVersion := "unknown"
parts := strings.Split(out, "ver=")
if len(parts) > 1 {
agentVersion = strings.TrimSpace(parts[1])
}
h.db.Model(&server).Updates(map[string]interface{}{
"agent_version": "v0.1.0",
"agent_version": agentVersion,
"last_check_at": time.Now(),
})
log.Printf("[BatchUpgrade] %s Agent 升级成功", server.Hostname)
backend/internal/api/handlers/agent_task.go
+1 -1
View File
@@ -191,7 +191,7 @@ func (h *AgentTaskHandler) QuickSetupRules(c *gin.Context) {
{Name: "CPU使用率(Warn)", Metric: "cpu", Operator: ">", Threshold: 80},
{Name: "内存使用率(Warn)", Metric: "mem", Operator: ">", Threshold: 80},
{Name: "磁盘使用率(Warn)", Metric: "disk", Operator: ">", Threshold: 80},
{Name: "网络带宽使用率(Warn)", Metric: "net", Operator: ">", Threshold: 80},
{Name: "FD使用率(Warn)", Metric: "fd", Operator: ">", Threshold: 80},
}
created := 0
backend/internal/api/handlers/alert_engine.go
+10 -2
View File
@@ -30,7 +30,7 @@ func (e *AlertEngine) Check() {
// 检查 CPU/内存/磁盘/进程/TCP连接/端口 阈值
func (e *AlertEngine) checkResourceAlerts() {
var rules []models.AgentRule
e.db.Where("enabled = ? AND metric IN ?", true, []string{"cpu", "mem", "disk", "service_process", "tcp_time_wait", "tcp_close_wait"}).Find(&rules)
e.db.Where("enabled = ? AND metric IN ?", true, []string{"cpu", "mem", "disk", "service_process", "tcp_time_wait", "tcp_close_wait", "fd"}).Find(&rules)
for _, rule := range rules {
// 获取目标主机最新上报
@@ -81,6 +81,10 @@ func (e *AlertEngine) checkResourceAlerts() {
continue // 进程存在，不触发
}
value = 0 // 进程不存在时触发
case "fd":
var fdStat models.FDMetric
json.Unmarshal([]byte(report.FDInfo), &fdStat)
value = fdStat.UsagePct
}
// 检查是否超阈值
@@ -288,7 +292,7 @@ func (e *AlertEngine) autoResolveAlerts() {
if report.ID > 0 && time.Since(report.ReportTime) < 3*time.Minute {
resolved = true
}
case "cpu", "mem", "disk", "tcp_time_wait", "tcp_close_wait":
case "cpu", "mem", "disk", "tcp_time_wait", "tcp_close_wait", "fd":
// 获取对应规则，检查是否恢复正常
var rule models.AgentRule
e.db.Where("name = ?", alert.RuleName).First(&rule)
@@ -326,6 +330,10 @@ func (e *AlertEngine) autoResolveAlerts() {
var netStat models.NetworkStatModel
json.Unmarshal([]byte(report.NetInfo), &netStat)
value = float64(netStat.TCPCloseWait)
case "fd":
var fdStat models.FDMetric
json.Unmarshal([]byte(report.FDInfo), &fdStat)
value = fdStat.UsagePct
}
if !e.compareOperator(rule.Operator, value, rule.Threshold) {
resolved = true
backend/internal/api/handlers/dashboard.go
+8 -12
View File
@@ -294,28 +294,24 @@ func (h *DashboardHandler) GetMetrics(c *gin.Context) {
}
}
// 从 agent_reports 表查最近 N 小时的真实数据（MySQL 聚合）
// 从 agent_reports 表查最近 N 小时的真实数据（整数分组，可用索引）
since := time.Now().Add(-time.Duration(hours) * time.Hour)
type avgPoint struct {
Bucket string
AvgCPU float64
AvgMem float64
BucketTS int64
AvgCPU float64
AvgMem float64
}
var rows []avgPoint
h.db.Model(&models.AgentReport{}).
Select("DATE_FORMAT(report_time, '%Y-%m-%d %H:00') as bucket, ROUND(AVG(cpu_usage),2) as avg_cpu, ROUND(AVG(mem_usage),2) as avg_mem").
Select("UNIX_TIMESTAMP(report_time) DIV 3600 as bucket_ts, ROUND(AVG(cpu_usage),2) as avg_cpu, ROUND(AVG(mem_usage),2) as avg_mem").
Where("report_time >= ?", since).
Group("bucket").
Order("bucket ASC").
Group("bucket_ts").
Order("bucket_ts ASC").
Scan(&rows)
var metrics []models.MetricData
for _, r := range rows {
// 手动解析 SQLite strftime 字符串为 time.Time（GORM 无法自动转换） // Logan.Li
ts, err := time.ParseInLocation("2006-01-02 15:00", r.Bucket, time.Local)
if err != nil {
ts = time.Now()
}
ts := time.Unix(r.BucketTS*3600, 0)
metrics = append(metrics,
models.MetricData{Name: "cpu", Value: r.AvgCPU, Timestamp: ts},
models.MetricData{Name: "memory", Value: r.AvgMem, Timestamp: ts},
backend/internal/api/handlers/kubernetes.go
+192
View File
@@ -1806,6 +1806,32 @@ func (h *KubernetesHandler) resolveClusterCode(code string) int {
return int(cluster.ID)
}
// formatAge converts a K8s creationTimestamp (RFC3339) to a human-readable relative age
func formatAge(creationTimestamp string) string {
if creationTimestamp == "" {
return "-"
}
t, err := time.Parse(time.RFC3339, creationTimestamp)
if err != nil {
return creationTimestamp
}
d := time.Since(t)
switch {
case d < time.Minute:
return fmt.Sprintf("%ds", int(d.Seconds()))
case d < time.Hour:
return fmt.Sprintf("%dm", int(d.Minutes()))
case d < 24*time.Hour:
return fmt.Sprintf("%dh", int(d.Hours()))
case d < 30*24*time.Hour:
return fmt.Sprintf("%dd", int(d.Hours()/24))
case d < 365*24*time.Hour:
return fmt.Sprintf("%dM", int(d.Hours()/(24*30)))
default:
return fmt.Sprintf("%dy", int(d.Hours()/(24*365)))
}
}
// getK8sClientByCode resolves cluster by code, falls back to ID
func (h *KubernetesHandler) getK8sClientByCode(c *gin.Context) (*k8sclient.Client, *models.KubernetesCluster, error) {
clusterID, _ := strconv.Atoi(c.Param("id"))
@@ -2571,6 +2597,12 @@ func (h *KubernetesHandler) GetResourceYAMLNS(c *gin.Context) {
apiPath = fmt.Sprintf("/apis/autoscaling/v2/namespaces/%s/horizontalpodautoscalers/%s", ns, name)
case "cronjobs", "cronjob":
apiPath = fmt.Sprintf("/apis/batch/v1/namespaces/%s/cronjobs/%s", ns, name)
case "serviceaccounts", "serviceaccount":
apiPath = fmt.Sprintf("/api/v1/namespaces/%s/serviceaccounts/%s", ns, name)
case "roles", "role":
apiPath = fmt.Sprintf("/apis/rbac.authorization.k8s.io/v1/namespaces/%s/roles/%s", ns, name)
case "rolebindings", "rolebinding":
apiPath = fmt.Sprintf("/apis/rbac.authorization.k8s.io/v1/namespaces/%s/rolebindings/%s", ns, name)
default:
c.JSON(400, models.Response{Code: 400, Message: "unsupported resource type: " + resource}); return
}
@@ -2627,6 +2659,12 @@ func (h *KubernetesHandler) UpdateResourceYAMLNS(c *gin.Context) {
apiPath = fmt.Sprintf("/apis/autoscaling/v2/namespaces/%s/horizontalpodautoscalers/%s", ns, name)
case "cronjobs", "cronjob":
apiPath = fmt.Sprintf("/apis/batch/v1/namespaces/%s/cronjobs/%s", ns, name)
case "serviceaccounts", "serviceaccount":
apiPath = fmt.Sprintf("/api/v1/namespaces/%s/serviceaccounts/%s", ns, name)
case "roles", "role":
apiPath = fmt.Sprintf("/apis/rbac.authorization.k8s.io/v1/namespaces/%s/roles/%s", ns, name)
case "rolebindings", "rolebinding":
apiPath = fmt.Sprintf("/apis/rbac.authorization.k8s.io/v1/namespaces/%s/rolebindings/%s", ns, name)
default:
c.JSON(400, models.Response{Code: 400, Message: "unsupported resource type: " + resource}); return
}
@@ -2663,6 +2701,12 @@ func resourceAPIPath(ns, resource, name string) (string, error) {
return fmt.Sprintf("/apis/autoscaling/v2/namespaces/%s/horizontalpodautoscalers/%s", ns, name), nil
case "cronjobs", "cronjob":
return fmt.Sprintf("/apis/batch/v1/namespaces/%s/cronjobs/%s", ns, name), nil
case "serviceaccounts", "serviceaccount":
return fmt.Sprintf("/api/v1/namespaces/%s/serviceaccounts/%s", ns, name), nil
case "roles", "role":
return fmt.Sprintf("/apis/rbac.authorization.k8s.io/v1/namespaces/%s/roles/%s", ns, name), nil
case "rolebindings", "rolebinding":
return fmt.Sprintf("/apis/rbac.authorization.k8s.io/v1/namespaces/%s/rolebindings/%s", ns, name), nil
default:
return "", fmt.Errorf("unsupported resource type: %s", resource)
}
@@ -3744,3 +3788,151 @@ func (h *KubernetesHandler) GetMultiClusterComparison(c *gin.Context) {
},
})
}
// ============================================================================
// RBAC — ServiceAccount / Role / RoleBinding (code-based)
// ============================================================================
// ListServiceAccountsNS lists ServiceAccounts in a namespace
func (h *KubernetesHandler) ListServiceAccountsNS(c *gin.Context) {
clusterID := h.resolveClusterCode(c.Param("clusterCode"))
if clusterID == 0 {
c.JSON(500, models.Response{Code: 500, Message: "cluster not found"})
return
}
client, _, err := h.getK8sClient(clusterID)
if err != nil {
c.JSON(500, models.Response{Code: 500, Message: err.Error()})
return
}
sas, err := client.ListServiceAccounts(c.Param("namespace"))
if err != nil {
c.JSON(500, models.Response{Code: 500, Message: "K8s API error: " + err.Error()})
return
}
type SASummary struct {
Namespace string `json:"namespace"`
Name string `json:"name"`
SecretCount int `json:"secret_count"`
ImagePullSecrets []string `json:"image_pull_secrets"`
Age string `json:"age"`
}
result := make([]SASummary, 0, len(sas))
for _, sa := range sas {
ipsNames := make([]string, 0, len(sa.ImagePullSecrets))
for _, ips := range sa.ImagePullSecrets {
ipsNames = append(ipsNames, ips.Name)
}
result = append(result, SASummary{
Namespace: sa.Metadata.Namespace,
Name: sa.Metadata.Name,
SecretCount: len(sa.Secrets),
ImagePullSecrets: ipsNames,
Age: formatAge(sa.Metadata.CreationTimestamp),
})
}
c.JSON(200, models.Response{Code: 200, Data: result})
}
// ListRolesNS lists Roles in a namespace
func (h *KubernetesHandler) ListRolesNS(c *gin.Context) {
clusterID := h.resolveClusterCode(c.Param("clusterCode"))
if clusterID == 0 {
c.JSON(500, models.Response{Code: 500, Message: "cluster not found"})
return
}
client, _, err := h.getK8sClient(clusterID)
if err != nil {
c.JSON(500, models.Response{Code: 500, Message: err.Error()})
return
}
roles, err := client.ListRoles(c.Param("namespace"))
if err != nil {
c.JSON(500, models.Response{Code: 500, Message: "K8s API error: " + err.Error()})
return
}
type RoleSummary struct {
Namespace string `json:"namespace"`
Name string `json:"name"`
RuleCount int `json:"rule_count"`
Verbs []string `json:"verbs"`
Resources []string `json:"resources"`
Age string `json:"age"`
}
result := make([]RoleSummary, 0, len(roles.Items))
for _, r := range roles.Items {
allVerbs := []string{}
allResources := []string{}
seen := map[string]bool{}
for _, rule := range r.Rules {
for _, v := range rule.Verbs {
if !seen[v] {
allVerbs = append(allVerbs, v)
seen[v] = true
}
}
for _, res := range rule.Resources {
if !seen[res] {
allResources = append(allResources, res)
seen[res] = true
}
}
}
result = append(result, RoleSummary{
Namespace: r.Metadata.Namespace,
Name: r.Metadata.Name,
RuleCount: len(r.Rules),
Verbs: allVerbs,
Resources: allResources,
Age: formatAge(r.Metadata.CreationTimestamp),
})
}
c.JSON(200, models.Response{Code: 200, Data: result})
}
// ListRoleBindingsNS lists RoleBindings in a namespace
func (h *KubernetesHandler) ListRoleBindingsNS(c *gin.Context) {
clusterID := h.resolveClusterCode(c.Param("clusterCode"))
if clusterID == 0 {
c.JSON(500, models.Response{Code: 500, Message: "cluster not found"})
return
}
client, _, err := h.getK8sClient(clusterID)
if err != nil {
c.JSON(500, models.Response{Code: 500, Message: err.Error()})
return
}
rbs, err := client.ListRoleBindings(c.Param("namespace"))
if err != nil {
c.JSON(500, models.Response{Code: 500, Message: "K8s API error: " + err.Error()})
return
}
type RBSummary struct {
Namespace string `json:"namespace"`
Name string `json:"name"`
RoleRefKind string `json:"role_ref_kind"`
RoleRefName string `json:"role_ref_name"`
Subjects []string `json:"subjects"`
Age string `json:"age"`
}
result := make([]RBSummary, 0, len(rbs.Items))
for _, rb := range rbs.Items {
subjStrs := make([]string, 0, len(rb.Subjects))
for _, s := range rb.Subjects {
if s.Namespace != "" && s.Namespace != rb.Metadata.Namespace {
subjStrs = append(subjStrs, fmt.Sprintf("%s/%s (%s)", s.Kind, s.Name, s.Namespace))
} else {
subjStrs = append(subjStrs, fmt.Sprintf("%s/%s", s.Kind, s.Name))
}
}
result = append(result, RBSummary{
Namespace: rb.Metadata.Namespace,
Name: rb.Metadata.Name,
RoleRefKind: rb.RoleRef.Kind,
RoleRefName: rb.RoleRef.Name,
Subjects: subjStrs,
Age: formatAge(rb.Metadata.CreationTimestamp),
})
}
c.JSON(200, models.Response{Code: 200, Data: result})
}