feat(services): считать сервис живым при одном OK и писать историю failover

Сервис up, если жив хотя бы один IP. Failover показывает Down по FQDN и журнал add/remove A-записей. Co-authored-by: Cursor <[email protected]>
2026-08-20 15:49:17 +07:00
parent 7938d2f707
commit fa7d2b6df3
23 changed files with 1030 additions and 107 deletions
@@ -13,7 +13,11 @@ import {
 import { HealthCheckBadge } from '@/components/health-check-badge'
 import { EmptyState } from '@/components/empty-state'
 import { formatDate, formatRelative, sqliteUtcToIso } from '@/lib/format'
-import type { FailoverEvent } from '@/lib/failover-events'
+import {
+  failoverEventCopy,
+  type FailoverEvent,
+} from '@/lib/failover-events'
+import type { FailoverLogEntry } from '@/lib/schemas'
 import { cn } from '@cfdm/ui/lib/utils'
 import type { ComponentProps } from 'react'

@@ -29,32 +33,44 @@ function failStreakLabel(count: number): string {
  return `${count} ошибок подряд`
 }

-function indicatorClass(status: string): string {
-  if (status === 'checking') {
-    return 'border-warning bg-warning/15 group-data-completed/timeline-item:border-warning'
+function indicatorClass(tone: 'down' | 'added' | 'removed'): string {
+  if (tone === 'added') {
+    return 'border-success bg-success/15 group-data-completed/timeline-item:border-success'
  }
  return 'border-destructive bg-destructive/15 group-data-completed/timeline-item:border-destructive'
 }

-function separatorClass(status: string): string {
-  if (status === 'checking') return 'bg-warning/25'
+function separatorClass(tone: 'down' | 'added' | 'removed'): string {
+  if (tone === 'added') return 'bg-success/25'
  return 'bg-destructive/25'
 }

+function failoverHistoryCopy(item: FailoverLogEntry): string {
+  return item.action === 'removed'
+    ? `${item.ip} убрана с ${item.fqdn}`
+    : `${item.ip} добавлена на ${item.fqdn}`
+}
+
 /**
- * Failover как sibling «Смены статуса»: ReUI Timeline + Badge, не степпер.
+ * Failover как sibling «Смены статуса»: ReUI Timeline + Badge.
 * Preview: https://reui.io/preview/base/components/c-timeline-10
 * Preview: https://reui.io/preview/base/empty-state-12
 * Docs: https://reui.io/docs/components/base/timeline
 * Docs: https://reui.io/docs/components/base/badge
 */
-export function FailoverTimeline({ events }: { events: FailoverEvent[] }) {
-  if (events.length === 0) {
+export function FailoverTimeline({
+  events,
+  history = [],
+}: {
+  events: FailoverEvent[]
+  history?: readonly FailoverLogEntry[]
+}) {
+  if (events.length === 0 && history.length === 0) {
    return (
      <EmptyState
        icon={ShieldCheckIcon}
-        title="Пул в DNS на месте"
-        description="Standby и last-resort не красные — только down, снятые с A-записей"
+        title="Нет инцидентов Failover"
+        description="Нет Down и нет смен A-записей по FQDN"
        stackedIcon={false}
        centered={false}
      />
@@ -62,55 +78,88 @@ export function FailoverTimeline({ events }: { events: FailoverEvent[] }) {
  }

  return (
-    <Timeline defaultValue={0} className="gap-0">
-      {events.map((event, index) => {
-        const checkedIso = event.lastCheckAt
-          ? (sqliteUtcToIso(event.lastCheckAt) ?? event.lastCheckAt)
-          : null
-        const isChecking = event.status === 'checking'
+    <div className="flex flex-col gap-4">
+      {events.length > 0 ? (
+        <Timeline defaultValue={0} className="gap-0">
+          {events.map((event, index) => {
+            const checkedIso = event.lastCheckAt
+              ? (sqliteUtcToIso(event.lastCheckAt) ?? event.lastCheckAt)
+              : null

-        return (
-          <TimelineItem key={event.id} step={index + 1}>
-            <TimelineSeparator className={separatorClass(event.status)} />
-            <TimelineIndicator className={indicatorClass(event.status)} />
-            <TimelineHeader>
-              <TimelineTitle className="flex flex-wrap items-center gap-2">
-                <span className="font-mono text-sm">{event.address}</span>
-                <HealthCheckBadge
-                  status={event.status as HealthBadgeStatus}
-                  lastError={event.lastFailureReason}
-                  lastCheckedAt={checkedIso}
-                  size="xs"
-                />
-              </TimelineTitle>
-              <TimelineDate>
-                {event.consecutiveFailures > 0
-                  ? failStreakLabel(event.consecutiveFailures)
-                  : null}
-                {checkedIso
-                  ? `${event.consecutiveFailures > 0 ? ' · ' : ''}${formatRelative(checkedIso)} · ${formatDate(checkedIso)}`
-                  : null}
-              </TimelineDate>
-            </TimelineHeader>
-            <TimelineContent className="flex flex-col gap-2">
-              <p className="text-foreground text-sm">
-                {isChecking
-                  ? 'Снята с DNS, идёт восстановление'
-                  : 'Снята с DNS — в A-записях этого адреса нет'}
-              </p>
-              {event.lastFailureReason ? (
-                <code
-                  className={cn(
-                    'bg-muted block overflow-x-auto rounded-md px-2 py-1.5 font-mono text-xs',
-                  )}
-                >
-                  {event.lastFailureReason}
-                </code>
-              ) : null}
-            </TimelineContent>
-          </TimelineItem>
-        )
-      })}
-    </Timeline>
+            return (
+              <TimelineItem key={event.id} step={index + 1}>
+                <TimelineSeparator className={separatorClass('down')} />
+                <TimelineIndicator className={indicatorClass('down')} />
+                <TimelineHeader>
+                  <TimelineTitle className="flex flex-wrap items-center gap-2">
+                    <span className="font-mono text-sm">{event.address}</span>
+                    <HealthCheckBadge
+                      status={event.status as HealthBadgeStatus}
+                      lastError={event.lastFailureReason}
+                      lastCheckedAt={checkedIso}
+                      size="xs"
+                    />
+                  </TimelineTitle>
+                  <TimelineDate>
+                    {event.consecutiveFailures > 0
+                      ? failStreakLabel(event.consecutiveFailures)
+                      : null}
+                    {checkedIso
+                      ? `${event.consecutiveFailures > 0 ? ' · ' : ''}${formatRelative(checkedIso)} · ${formatDate(checkedIso)}`
+                      : null}
+                  </TimelineDate>
+                </TimelineHeader>
+                <TimelineContent className="flex flex-col gap-2">
+                  <p className="text-foreground text-sm">
+                    {failoverEventCopy(event)}
+                  </p>
+                  {event.lastFailureReason ? (
+                    <code
+                      className={cn(
+                        'bg-muted block overflow-x-auto rounded-md px-2 py-1.5 font-mono text-xs',
+                      )}
+                    >
+                      {event.lastFailureReason}
+                    </code>
+                  ) : null}
+                </TimelineContent>
+              </TimelineItem>
+            )
+          })}
+        </Timeline>
+      ) : null}
+
+      {history.length > 0 ? (
+        <Timeline defaultValue={0} className="gap-0">
+          {history.map((item, index) => {
+            const checkedIso = sqliteUtcToIso(item.created_at) ?? item.created_at
+            const tone = item.action === 'added' ? 'added' : 'removed'
+
+            return (
+              <TimelineItem key={item.id} step={index + 1}>
+                <TimelineSeparator className={separatorClass(tone)} />
+                <TimelineIndicator className={indicatorClass(tone)} />
+                <TimelineHeader>
+                  <TimelineTitle className="flex flex-wrap items-center gap-2">
+                    <span className="font-mono text-sm">{item.ip}</span>
+                    <span className="text-muted-foreground text-xs">
+                      {item.fqdn}
+                    </span>
+                  </TimelineTitle>
+                  <TimelineDate>
+                    {formatRelative(checkedIso)} · {formatDate(checkedIso)}
+                  </TimelineDate>
+                </TimelineHeader>
+                <TimelineContent>
+                  <p className="text-foreground text-sm">
+                    {failoverHistoryCopy(item)}
+                  </p>
+                </TimelineContent>
+              </TimelineItem>
+            )
+          })}
+        </Timeline>
+      ) : null}
+    </div>
  )
 }
@@ -3,8 +3,10 @@ import { UnplugIcon } from 'lucide-react'
 import { FailoverTimeline } from '@/components/failover-timeline'
 import {
  toFailoverEvents,
+  type FailoverBindingPool,
  type FailoverHealthInput,
 } from '@/lib/failover-events'
+import type { FailoverLogEntry } from '@/lib/schemas'
 import { Badge } from '@/components/reui/badge'
 import {
  Frame,
@@ -22,15 +24,15 @@ import {
 function failoverCountLabel(count: number): string {
  const mod10 = count % 10
  const mod100 = count % 100
-  if (mod10 === 1 && mod100 !== 11) return `${count} нода вне пула`
+  if (mod10 === 1 && mod100 !== 11) return `${count} адрес Down`
  if (mod10 >= 2 && mod10 <= 4 && (mod100 < 12 || mod100 > 14)) {
-    return `${count} ноды вне пула`
+    return `${count} адреса Down`
  }
-  return `${count} нод вне пула`
+  return `${count} адресов Down`
 }

 /**
- * Failover — sibling ServiceHealthMonitor: Frame stacked + Alert + Timeline.
+ * Failover — текущие Down + журнал add/remove по FQDN.
 * Preview: https://reui.io/preview/base/components/c-timeline-10
 * Preview: https://reui.io/preview/base/empty-state-12
 * Docs: https://reui.io/docs/components/base/frame
@@ -40,12 +42,15 @@ function failoverCountLabel(count: number): string {
 */
 export function ServiceFailoverPanel({
  ipHealth,
-  activeAddresses,
+  bindings,
+  history,
 }: {
  ipHealth: readonly FailoverHealthInput[]
-  activeAddresses: readonly string[]
+  bindings: readonly FailoverBindingPool[]
+  history: readonly FailoverLogEntry[]
 }) {
-  const events = toFailoverEvents(ipHealth, activeAddresses)
+  const events = toFailoverEvents(ipHealth, bindings)
+  const removedCount = events.filter((event) => event.kind === 'removed').length

  return (
    <Frame stacked spacing="sm" className="min-w-0 w-full">
@@ -64,7 +69,7 @@ export function ServiceFailoverPanel({
            )}
          </FrameTitle>
          <FrameDescription>
-            Только down из IP Health вне DNS-пула · как таблица активов
+            Текущие Down и история A-записей по FQDN
          </FrameDescription>
        </FrameHeader>

@@ -73,12 +78,14 @@ export function ServiceFailoverPanel({
            <UnplugIcon aria-hidden="true" />
            <AlertTitle>{failoverCountLabel(events.length)}</AlertTitle>
            <AlertDescription>
-              Эти IP сняты с A-записей
+              {removedCount > 0
+                ? 'Сняты с FQDN или остались last-resort'
+                : 'Остались в A-записях как last-resort'}
            </AlertDescription>
          </Alert>
        ) : null}

-        <FailoverTimeline events={events} />
+        <FailoverTimeline events={events} history={history} />
      </FramePanel>
    </Frame>
  )