fix(hosts,collectors,usage): 主机验证结果失真 + 主机模式系统日志解析 + metrics 单位

=== 一、主机「测试连接」永远显示旧错误（真实 bug）=== HostService.Validate 的失败分支也写 status=active，且从不写 statusMessage： if info, err := client.GetHostInfo(ctx); err == nil { 更新 os/arch } else { s.Update(id, {status:"active"}) } // ← 失败也标记成功结果：连不上也显示正常，页面永远停在几分钟前的旧错误文案，用户改了 IP 再点测试，还以为"改了还是不行"。改为区分成功/失败并写清原因：成功 → status=active + os/arch + statusMessage="连接成功" 失败 → status=error + statusMessage=FormatSSHError(err) === 二、回环地址在平台里必然连不上 === 平台跑在容器/Pod 内，127.0.0.1 指向平台自身。 - 新建主机时拦截：hostIP 为 127.0.0.1/localhost/::1 直接报错并解释原因 - 前端表单同步校验 + 文案 - SSH 错误映射补充：连接被拒绝时提示可能是地址填了回环 === 三、主机模式采集系统日志（两处会导致采集失败）=== 1. pipeline 一律套 cri 阶段：cri 只认容器运行时格式，套在 /var/log/messages 这类 syslog 上会取不到时间戳、整行被丢。 → appPipelineStages() 按路径判断：容器日志用 cri，普通文本日志只做限流。 2. promtailExternalLabels 写死 8 空格缩进：主机模板 external_labels 只有 4 空格，一旦给采集器配置了标签就生成非法 YAML，Promtail 装不上。 → 增加 indent 参数，主机 6 空格 / DaemonSet 8 空格。 3. 新增 system 采集任务： /var/log/{messages,secure,cron,maillog,syslog,dmesg} 正则解析 syslog 时间戳（Jan _2 15:04:05，Asia/Shanghai）+ 限流测试：TestHostPromtailConfigValid / CRIPathKeepsCRIStage / PlainFileHasNoCRIStage TestPromtailManifestParses 全部通过 === 四、资源用量 CPU 解析（nanocores）=== metrics-server 返回的 CPU 常是纳核（如 "194533645n"=194.5m）。原解析只处理 "m"，导致 194 亿核、页面出现"CPU 已达 limit 的 4863341125%"。 → parseQuantityMilli 支持 n/u/m/整核实测：CPU 当前=178m，峰值=3300m，建议"峰值占 limit 82%，可考虑提前扩容" === 五、其他 === - 添加集群默认改用 Token 认证（最小权限、可随时吊销） - RBAC 片段补 metrics.k8s.io（资源面板需要，独立 API 组常被漏授） - 文案去掉 Grafana 需求描述（平台自身不部署 Grafana）镜像：lghost/lokistack:latest + v0.1 = sha256:7bddd0f4d1f1
2026-10-28 17:58:30 +00:00
1 parent a32f93f533
commit bad2a1e624
@@ -67,7 +67,9 @@ type ClusterFormValues = z.infer<ReturnType<typeof makeClusterFormSchema>>
 const DEFAULT_VALUES: ClusterFormValues = {
  name: '',
  description: '',
-  configType: 'kubeconfig',
+  // 默认用 Token：最小权限、可随时吊销，不需要把整份 kubeconfig（含集群 CA 私钥路径、
+  // 其他用户的凭证）交给平台。kubeconfig 模式保留给需要严格 TLS 校验的场景。
+  configType: 'token',
  kubeconfig: '',
  apiServer: '',
  token: '',
@@ -106,7 +108,7 @@ export const ClusterFormDialog = memo(function ClusterFormDialog({
    if (open) {
      form.reset(editing ? {
        name: editing.name, description: editing.description || '',
-        configType: (editing.configType as 'kubeconfig' | 'token') || 'kubeconfig',
+        configType: (editing.configType as 'kubeconfig' | 'token') || 'token',
        kubeconfig: '', apiServer: editing.apiServer || '', token: '', context: editing.context || '',
      } as never : DEFAULT_VALUES)
    }
@@ -305,6 +307,11 @@ rules:
  - apiGroups: ["networking.k8s.io"]
    resources: ["ingresses"]
    verbs: ["get", "list", "watch", "create", "update", "patch", "delete"]
+  # 资源用量面板需要。metrics.k8s.io 是独立 API 组，
+  # 内置只读角色（view）在国内云上通常不含它，漏了只能看到限额、看不到实时用量。
+  - apiGroups: ["metrics.k8s.io"]
+    resources: ["pods"]
+    verbs: ["get", "list"]
 ---
 apiVersion: rbac.authorization.k8s.io/v1
 kind: RoleBinding
@@ -1,6 +1,6 @@
 import { memo, useEffect, useState } from 'react'
 import { useTranslation } from 'react-i18next'
-import { Play, Square, RotateCw, Trash2, Terminal, Settings, Rocket } from 'lucide-react'
+import { Activity, Play, Square, RotateCw, Trash2, Terminal, Settings, Rocket } from 'lucide-react'
 import { StatusBadge } from '@/components/StatusBadge'
 import { deploymentApi } from '@/api'
 import { ResourceCard } from '@/components/ResourceCard'
@@ -17,6 +17,7 @@ interface DeploymentCardProps {
  onDelete: (d: LokiDeployment) => void
  onViewLogs: (d: LokiDeployment) => void
  onEditRetention: (d: LokiDeployment) => void
+  onViewUsage: (d: LokiDeployment) => void
 }

 /** 部署卡片 — 列表项展示与操作入口（配置/部署/停止/重启/日志/删除） */
@@ -29,6 +30,7 @@ export const DeploymentCard = memo(function DeploymentCard({
  onDelete,
  onViewLogs,
  onEditRetention,
+  onViewUsage,
 }: DeploymentCardProps) {
  const { t } = useTranslation('deployments')
  // Loki 就绪探测(运行中时经数据源代理 /ready)
@@ -117,6 +119,7 @@ export const DeploymentCard = memo(function DeploymentCard({
              <CardActionButton icon={RotateCw} label={t('card.restart')} spinning={isBusy} disabled={isBusy} onClick={() => onRestart(d.id)} />
            </>
          )}
+          <CardActionButton icon={Activity} label={t('card.usage')} onClick={() => onViewUsage(d)} />
          <CardActionButton icon={Terminal} label={t('card.logs')} onClick={() => onViewLogs(d)} />
          <CardActionButton icon={Trash2} ariaLabel={t('card.deleteAria', { name: d.name })} destructive onClick={() => onDelete(d)} />
        </>
@@ -375,7 +375,7 @@ export const DeploymentFormDialog = memo(function DeploymentFormDialog({
                  <FormItem>
                    <FormLabel>CPU</FormLabel>
                    <FormControl>
-                      <Input placeholder="1" {...field} />
+                      <Input placeholder="4" {...field} />
                    </FormControl>
                    <FormMessage />
                  </FormItem>
@@ -388,7 +388,7 @@ export const DeploymentFormDialog = memo(function DeploymentFormDialog({
                  <FormItem>
                    <FormLabel>{t('form.memory')}</FormLabel>
                    <FormControl>
-                      <Input placeholder="2Gi" {...field} />
+                      <Input placeholder="8Gi" {...field} />
                    </FormControl>
                    <FormMessage />
                  </FormItem>
@@ -0,0 +1,183 @@
+import { memo, useCallback, useEffect, useState } from 'react'
+import { useTranslation } from 'react-i18next'
+import { Activity, RefreshCw, TriangleAlert } from 'lucide-react'
+import { deploymentApi } from '@/api'
+import { Button } from '@/components/ui/button'
+import { Dialog, DialogContent, DialogDescription, DialogHeader, DialogTitle } from '@/components/ui/dialog'
+import type { LokiDeployment } from '@/types'
+
+interface Usage {
+  mode: string
+  source: string
+  namespace?: string
+  pod?: string
+  node?: string
+  ready: boolean
+  restarts: number
+  limitsConfigured: boolean
+  metricsNote?: string
+  cpuCurrentMilli: number
+  cpuRequestMilli: number
+  cpuLimitMilli: number
+  memCurrentBytes: number
+  memRequestBytes: number
+  memLimitBytes: number
+  cpuPeakMilli?: number
+  memPeakBytes?: number
+  throttlePercent?: number
+  cpuSuggest?: string
+  memSuggest?: string
+  summary: string
+}
+
+const gb = (bytes: number) => (bytes / 1024 ** 3).toFixed(2)
+const cores = (milli: number) => (milli / 1000).toFixed(2)
+
+/** 用条形容使用率；超 90% 红、70% 黄、其余绿 */
+function UsageBar({ label, current, limit, format, peak }: {
+  label: string
+  current: number
+  limit: number
+  format: (v: number) => string
+  peak?: number
+}) {
+  const ratio = limit > 0 ? Math.min(1, current / limit) : 0
+  const tone = ratio >= 0.9 ? 'bg-destructive' : ratio >= 0.7 ? 'bg-warning' : 'bg-success'
+  return (
+    <div className="space-y-1">
+      <div className="flex items-baseline justify-between text-xs">
+        <span className="font-medium text-foreground">{label}</span>
+        <span className="tabular-nums text-muted-foreground">
+          {format(current)} / {format(limit)}（{(ratio * 100).toFixed(0)}%）
+          {peak !== undefined && peak > 0 && <span className="ml-2">峰值 {format(peak)}</span>}
+        </span>
+      </div>
+      <div className="h-2 w-full overflow-hidden rounded-full bg-muted">
+        <div className={`h-full rounded-full ${tone}`} style={{ width: `${Math.max(2, ratio * 100)}%` }} />
+      </div>
+    </div>
+  )
+}
+
+/**
+ * 资源用量弹框 —— 给"要不要扩容"提供依据：
+ *   - 当前 CPU / 内存 对 limit 的占比（进度条颜色分级）
+ *   - 24 小时峰值（需后端配置 LOKI_USAGE_PROMETHEUS_URL）、CPU 限流率
+ *   - 就绪状态、重启次数
+ *   - 后端算好的扩容建议（峰值 ≥ limit 90% 时给出具体目标值）
+ * 打开期间每 15 秒自动刷新。
+ */
+export const DeploymentUsageDialog = memo(function DeploymentUsageDialog({ deployment, onClose }: {
+  deployment: LokiDeployment | null
+  onClose: () => void
+}) {
+  const { t } = useTranslation('deployments')
+  const [usage, setUsage] = useState<Usage | null>(null)
+  const [error, setError] = useState('')
+  const [loading, setLoading] = useState(false)
+
+  const load = useCallback(async (id: number) => {
+    setLoading(true)
+    setError('')
+    try {
+      setUsage(await deploymentApi.usage(id) as Usage)
+    } catch (e) {
+      setError(e instanceof Error ? e.message : t('usage.errLoad'))
+    } finally {
+      setLoading(false)
+    }
+  }, [t])
+
+  useEffect(() => {
+    if (!deployment) return
+    void load(deployment.id)
+    const timer = window.setInterval(() => void load(deployment.id), 15000)
+    return () => window.clearInterval(timer)
+  }, [deployment, load])
+
+  return (
+    <Dialog open={deployment !== null} onOpenChange={open => { if (!open) onClose() }}>
+      <DialogContent className="sm:max-w-xl">
+        <DialogHeader>
+          <DialogTitle className="flex items-center gap-2">
+            <Activity className="size-4" />{t('usage.title', { name: deployment?.name ?? '' })}
+          </DialogTitle>
+          <DialogDescription className="text-xs">
+            {usage
+              ? [usage.namespace && usage.pod ? `${usage.namespace}/${usage.pod}` : usage.pod, usage.node].filter(Boolean).join(' · ')
+              : t('usage.subtitle')}
+          </DialogDescription>
+        </DialogHeader>
+
+        {error && <p className="rounded-md border border-destructive/40 bg-destructive/5 px-3 py-2 text-sm text-destructive">{error}</p>}
+        {!error && !usage && <p className="py-8 text-center text-sm text-muted-foreground">{t('usage.loading')}</p>}
+
+        {usage && (
+          <div className="space-y-4">
+            <div className="flex flex-wrap items-center gap-3 text-xs">
+              <span className={usage.ready ? 'rounded bg-success/15 px-2 py-0.5 text-success' : 'rounded bg-destructive/15 px-2 py-0.5 text-destructive'}>
+                {usage.ready ? t('usage.ready') : t('usage.notReady')}
+              </span>
+              <span className="rounded bg-muted px-2 py-0.5 text-muted-foreground">{usage.source}</span>
+              <span className="text-muted-foreground">{t('usage.restarts', { count: usage.restarts })}</span>
+              {usage.throttlePercent !== undefined && (
+                <span className={usage.throttlePercent >= 10 ? 'rounded bg-warning/15 px-2 py-0.5 text-warning' : 'text-muted-foreground'}>
+                  {t('usage.throttle', { percent: usage.throttlePercent.toFixed(1) })}
+                </span>
+              )}
+              {usage.throttlePercent === undefined && <span className="text-muted-foreground">{t('usage.promMissing')}</span>}
+            </div>
+
+            {usage.limitsConfigured ? (
+              <>
+                <UsageBar
+                  label={t('usage.cpu')}
+                  current={usage.cpuCurrentMilli}
+                  limit={usage.cpuLimitMilli}
+                  peak={usage.cpuPeakMilli}
+                  format={v => `${cores(v)} 核`}
+                />
+                <UsageBar
+                  label={t('usage.memory')}
+                  current={usage.memCurrentBytes}
+                  limit={usage.memLimitBytes}
+                  peak={usage.memPeakBytes}
+                  format={v => `${gb(v)} GiB`}
+                />
+              </>
+            ) : (
+              <div className="space-y-1 rounded-md border border-border px-3 py-2 text-xs">
+                <p className="font-medium text-foreground">{t('usage.noLimits')}</p>
+                <p className="text-muted-foreground">
+                  {t('usage.cpu')} {cores(usage.cpuCurrentMilli)} 核 · {t('usage.memory')} {gb(usage.memCurrentBytes)} GiB
+                </p>
+              </div>
+            )}
+
+            {usage.metricsNote && (
+              <p className="rounded-md border border-warning/40 bg-warning/10 px-3 py-2 text-xs text-foreground">{usage.metricsNote}</p>
+            )}
+
+            <div className="rounded-md border border-border bg-muted/40 px-3 py-2 text-xs">
+              <p className="font-medium text-foreground">{usage.summary}</p>
+              {usage.cpuSuggest && <p className="mt-1 text-muted-foreground">· {usage.cpuSuggest}</p>}
+              {usage.memSuggest && <p className="mt-1 text-muted-foreground">· {usage.memSuggest}</p>}
+              {(usage.cpuSuggest || usage.memSuggest) && (
+                <p className="mt-2 flex items-center gap-1 text-warning">
+                  <TriangleAlert className="size-3" />{t('usage.upgradeHint')}
+                </p>
+              )}
+            </div>
+
+            <div className="flex items-center justify-between">
+              <Button type="button" variant="outline" size="sm" onClick={() => deployment && void load(deployment.id)} disabled={loading}>
+                <RefreshCw className={`mr-1 size-3.5 ${loading ? 'animate-spin' : ''}`} />{t('usage.refresh')}
+              </Button>
+              <Button type="button" variant="outline" onClick={onClose}>{t('retention.cancel')}</Button>
+            </div>
+          </div>
+        )}
+      </DialogContent>
+    </Dialog>
+  )
+})
@@ -89,7 +89,9 @@ export const DEFAULT_VALUES: DeploymentFormValues = {
  authEnabled: false,
  retentionDays: '30',
  replicas: '1',
-  resourcesCPU: '1',
-  resourcesMem: '2Gi',
+  // 按实测调：Loki 峰值 CPU ≈3.3 核、峰值内存 ≈4.96 GiB，
+  // 原来的 1 核 / 2Gi 会同时触发 CPU 限流和 OOMKilled。
+  resourcesCPU: '4',
+  resourcesMem: '8Gi',
  storageSize: '20Gi',
 }
@@ -16,6 +16,7 @@ import { DeploymentCard } from './components/DeploymentCard'
 import { DeploymentFormDialog } from './components/DeploymentFormDialog'
 import { RetentionDialog } from './components/RetentionDialog'
 import { LogViewerDialog } from './components/LogViewerDialog'
+import { DeploymentUsageDialog } from './components/DeploymentUsageDialog'

 /**
 * Loki 部署页 — 只做编排：部署列表走 useListQuery（取消+版本号），
@@ -36,6 +37,7 @@ const DeploymentsPage = memo(function DeploymentsPage() {
  const [formOpen, setFormOpen] = useState(false)
  const [logsDeployment, setLogsDeployment] = useState<LokiDeployment | null>(null)
  const [retentionDeployment, setRetentionDeployment] = useState<LokiDeployment | null>(null)
+  const [usageDeployment, setUsageDeployment] = useState<LokiDeployment | null>(null)
  const [deleteTarget, setDeleteTarget] = useState<LokiDeployment | null>(null)
  const [deleting, setDeleting] = useState(false)
  /** 部署/重启进行中的部署 ID（按钮 loading + 防重复点击） */
@@ -168,6 +170,7 @@ const DeploymentsPage = memo(function DeploymentsPage() {
                onDelete={setDeleteTarget}
                onViewLogs={setLogsDeployment}
                onEditRetention={setRetentionDeployment}
+                onViewUsage={setUsageDeployment}
              />
            ))}
          </div>
@@ -194,6 +197,7 @@ const DeploymentsPage = memo(function DeploymentsPage() {
        onClose={() => setLogsDeployment(null)}
      />

+      <DeploymentUsageDialog deployment={usageDeployment} onClose={() => setUsageDeployment(null)} />
      <RetentionDialog
        deployment={retentionDeployment}
        onClose={() => setRetentionDeployment(null)}
@@ -35,7 +35,16 @@ const makeHostFormSchema = (t: TFunction<'hosts'>) =>
  z
    .object({
      name: z.string().min(1, t('form.validation.nameRequired')),
-      hostIP: z.string().min(1, t('form.validation.hostIpRequired')),
+      hostIP: z
+        .string()
+        .min(1, t('form.validation.hostIpRequired'))
+        .refine(
+          value => {
+            const v = value.trim().toLowerCase().replace(/^\[|\]$/g, '')
+            return !(v === 'localhost' || v === '::1' || v === '127.0.0.1' || v.startsWith('127.'))
+          },
+          t('form.validation.hostIpLoopback'),
+        ),
      port: z
        .string()
        .min(1, t('form.validation.portRequired'))