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]>

apps/web/src/lib/failover-events.test.ts

@@ -1,8 +1,10 @@
 import { describe, expect, it } from 'vitest'
 
 import {
+  failoverEventCopy,
   isFailoverEventStatus,
   toFailoverEvents,
+  type FailoverBindingPool,
   type FailoverHealthInput,
 } from '@/lib/failover-events'
 
@@ -18,6 +20,24 @@ function row(
   }
 }
 
+const mskHip: FailoverBindingPool[] = [
+  {
+    fqdn: 'gt.rkns.top',
+    configured: ['130.49.213.153', '93.115.203.183'],
+    active: ['93.115.203.183'],
+  },
+  {
+    fqdn: 'nsgt.rkns.top',
+    configured: ['130.49.213.153'],
+    active: ['130.49.213.153'],
+  },
+  {
+    fqdn: 'rutg.rkns.top',
+    configured: ['93.115.203.183'],
+    active: ['93.115.203.183'],
+  },
+]
+
 describe('toFailoverEvents', () => {
   it('инцидент только при binding down, не при unhealthy ноды', () => {
     expect(isFailoverEventStatus('down')).toBe(true)
@@ -27,7 +47,7 @@ describe('toFailoverEvents', () => {
     expect(isFailoverEventStatus('unhealthy')).toBe(false)
   })
 
-  it('без DNS-пула ничего не показывает — нельзя врать про вывод', () => {
+  it('Down всегда виден, даже без A-записей', () => {
     const events = toFailoverEvents([
       row({
         ip: '130.49.213.153',
@@ -36,7 +56,19 @@ describe('toFailoverEvents', () => {
         last_error: 'fetch failed',
       }),
     ])
-    expect(events).toEqual([])
+    expect(events).toEqual([
+      {
+        id: '130.49.213.153',
+        address: '130.49.213.153',
+        status: 'down',
+        kind: 'last-resort',
+        fqdns: [],
+        consecutiveFailures: 9,
+        lastFailureReason: 'fetch failed',
+        lastCheckAt: null,
+      },
+    ])
+    expect(failoverEventCopy(events[0]!)).toBe('Down, в A-записях last-resort')
   })
 
   it('OK вне пула не инцидент (standby)', () => {
@@ -45,7 +77,13 @@ describe('toFailoverEvents', () => {
         row({ ip: '10.0.0.1', status: 'up' }),
         row({ ip: '130.49.213.153', status: 'up' }),
       ],
-      ['10.0.0.1'],
+      [
+        {
+          fqdn: 'pool.example.com',
+          configured: ['10.0.0.1', '130.49.213.153'],
+          active: ['10.0.0.1'],
+        },
+      ],
     )
     expect(events).toEqual([])
   })
@@ -57,12 +95,18 @@ describe('toFailoverEvents', () => {
         row({ ip: '10.0.0.2', status: 'unknown' }),
         row({ ip: '10.0.0.3', status: 'degraded' }),
       ],
-      ['10.0.0.1'],
+      [
+        {
+          fqdn: 'pool.example.com',
+          configured: ['10.0.0.1', '10.0.0.2', '10.0.0.3'],
+          active: ['10.0.0.1'],
+        },
+      ],
     )
     expect(events).toEqual([])
   })
 
-  it('down в DNS-пуле не инцидент (last-resort / ещё в A-записи)', () => {
+  it('Down на доп. FQDN last-resort, снятая с общего пула', () => {
     const events = toFailoverEvents(
       [
         row({
@@ -70,15 +114,46 @@ describe('toFailoverEvents', () => {
           status: 'down',
           consecutive_failures: 9,
           last_error: 'fetch failed',
+          last_checked_at: '2026-08-20 07:00:00',
         }),
-        row({ ip: '10.0.0.3', status: 'down' }),
+        row({ ip: '93.115.203.183', status: 'up' }),
       ],
-      ['130.49.213.153', '10.0.0.2'],
+      mskHip,
     )
-    expect(events.map((event) => event.address)).toEqual(['10.0.0.3'])
+    expect(events).toEqual([
+      {
+        id: '130.49.213.153',
+        address: '130.49.213.153',
+        status: 'down',
+        kind: 'removed',
+        fqdns: ['gt.rkns.top'],
+        consecutiveFailures: 9,
+        lastFailureReason: 'fetch failed',
+        lastCheckAt: '2026-08-20 07:00:00',
+      },
+    ])
+    expect(failoverEventCopy(events[0]!)).toBe('Снята с gt.rkns.top')
   })
 
-  it('down без A-записи — реальный вывод из пула', () => {
+  it('Down только last-resort на своём FQDN', () => {
+    const events = toFailoverEvents(
+      [row({ ip: '130.49.213.153', status: 'down' })],
+      [
+        {
+          fqdn: 'nsgt.rkns.top',
+          configured: ['130.49.213.153'],
+          active: ['130.49.213.153'],
+        },
+      ],
+    )
+    expect(events[0]?.kind).toBe('last-resort')
+    expect(events[0]?.fqdns).toEqual(['nsgt.rkns.top'])
+    expect(failoverEventCopy(events[0]!)).toBe(
+      'Down, в A-записях last-resort на nsgt.rkns.top',
+    )
+  })
+
+  it('down без A-записи на configured FQDN — снятие', () => {
     const events = toFailoverEvents(
       [
         row({
@@ -89,13 +164,21 @@ describe('toFailoverEvents', () => {
           last_checked_at: '2026-08-20 07:00:00',
         }),
       ],
-      ['10.0.0.1'],
+      [
+        {
+          fqdn: 'pool.example.com',
+          configured: ['10.0.0.1', '130.49.213.153'],
+          active: ['10.0.0.1'],
+        },
+      ],
     )
     expect(events).toEqual([
       {
         id: '130.49.213.153',
         address: '130.49.213.153',
         status: 'down',
+        kind: 'removed',
+        fqdns: ['pool.example.com'],
         consecutiveFailures: 9,
         lastFailureReason: 'fetch failed',
         lastCheckAt: '2026-08-20 07:00:00',

apps/web/src/lib/failover-events.ts

@@ -1,7 +1,11 @@
+export type FailoverEventKind = 'removed' | 'last-resort'
+
 export interface FailoverEvent {
   id: string
   address: string
   status: string
+  kind: FailoverEventKind
+  fqdns: string[]
   consecutiveFailures: number
   lastFailureReason: string | null
   lastCheckAt?: string | null
@@ -16,31 +20,56 @@ export interface FailoverHealthInput {
   last_checked_at?: string | null
 }
 
+export interface FailoverBindingPool {
+  fqdn: string
+  configured: readonly string[]
+  active: readonly string[]
+}
+
 /** Инцидент только при down. up / degraded / unknown — не вывод из пула. */
 export function isFailoverEventStatus(status: string): boolean {
   return status === 'down'
 }
 
+export function failoverEventCopy(event: FailoverEvent): string {
+  if (event.kind === 'removed') {
+    return event.fqdns.length > 0
+      ? `Снята с ${event.fqdns.join(', ')}`
+      : 'Снята с DNS'
+  }
+  return event.fqdns.length > 0
+    ? `Down, в A-записях last-resort на ${event.fqdns.join(', ')}`
+    : 'Down, в A-записях last-resort'
+}
+
 /**
- * Инцидент failover = binding health down и адреса нет в DNS-пуле.
- * OK / unknown / degraded вне пула — standby, не инцидент.
- * Down в activeAddresses (last-resort) — не «снята с DNS».
+ * Текущие Down всегда в панели. Per-FQDN: снята с hostname или last-resort.
+ * Standby (up / degraded / unknown) — не инцидент.
  */
 export function toFailoverEvents(
   ipHealth: readonly FailoverHealthInput[],
-  activeAddresses: readonly string[] = [],
+  bindings: readonly FailoverBindingPool[] = [],
 ): FailoverEvent[] {
-  const pool = new Set(activeAddresses)
-  if (pool.size === 0) return []
-
-  return ipHealth
-    .filter((row) => isFailoverEventStatus(row.status) && !pool.has(row.ip))
-    .map((row) => ({
+  return ipHealth.filter((row) => isFailoverEventStatus(row.status)).map((row) => {
+    const removedFqdns: string[] = []
+    const lastResortFqdns: string[] = []
+    for (const binding of bindings) {
+      const configured = binding.configured.includes(row.ip)
+      const active = binding.active.includes(row.ip)
+      if (configured && !active) removedFqdns.push(binding.fqdn)
+      else if (configured && active) lastResortFqdns.push(binding.fqdn)
+    }
+    const kind: FailoverEventKind =
+      removedFqdns.length > 0 ? 'removed' : 'last-resort'
+    return {
       id: row.ip,
       address: row.ip,
       status: row.status,
+      kind,
+      fqdns: kind === 'removed' ? removedFqdns : lastResortFqdns,
       consecutiveFailures: row.consecutive_failures ?? 0,
       lastFailureReason: row.last_error ?? null,
       lastCheckAt: row.last_checked_at ?? null,
-    }))
+    }
+  })
 }

apps/web/src/lib/health-log.test.ts

@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest'
 
 import {
+  bestAliveHealthStatus,
   collapseStatusChanges,
   enabledHealthProviders,
   providerHealthStatuses,
@@ -73,7 +74,7 @@ describe('enabledHealthProviders', () => {
 })
 
 describe('providerHealthStatuses', () => {
-  it('uses worst latest-per-ip status', () => {
+  it('uses any-up among latest-per-ip statuses', () => {
     const items = [
       probe({ id: 1, ip: '1.1.1.1', status: 'up', checked_at: '2026-01-01T00:02:00Z' }),
       probe({ id: 2, ip: '2.2.2.2', status: 'down', checked_at: '2026-01-01T00:01:00Z' }),
@@ -84,7 +85,7 @@ describe('providerHealthStatuses', () => {
         checked_at: '2026-01-01T00:00:00Z',
       }),
     ]
-    expect(providerHealthStatuses(items, ['local']).local).toBe('down')
+    expect(providerHealthStatuses(items, ['local']).local).toBe('up')
   })
 })
 
@@ -95,3 +96,12 @@ describe('worstHealthStatus', () => {
     expect(worstHealthStatus([])).toBe('unknown')
   })
 })
+
+describe('bestAliveHealthStatus', () => {
+  it('is up if any IP is up', () => {
+    expect(bestAliveHealthStatus(['up', 'down'])).toBe('up')
+    expect(bestAliveHealthStatus(['degraded', 'down'])).toBe('degraded')
+    expect(bestAliveHealthStatus(['down'])).toBe('down')
+    expect(bestAliveHealthStatus([])).toBe('unknown')
+  })
+})

apps/web/src/lib/health-log.ts

@@ -98,7 +98,18 @@ export function worstHealthStatus(statuses: readonly HealthLogStatus[]): HealthL
   )
 }
 
-/** Latest probe per IP for a provider, then worst among those IPs. */
+/** Any up → up; else degraded, then down, then unknown. */
+export function bestAliveHealthStatus(
+  statuses: readonly HealthLogStatus[],
+): HealthLogStatus {
+  if (statuses.length === 0) return 'unknown'
+  if (statuses.some((status) => status === 'up')) return 'up'
+  if (statuses.some((status) => status === 'degraded')) return 'degraded'
+  if (statuses.some((status) => status === 'down')) return 'down'
+  return 'unknown'
+}
+
+/** Latest probe per IP for a provider, then any-up among those IPs. */
 export function providerHealthStatuses(
   items: readonly HealthLogProbe[],
   providers: readonly HealthCheckProvider[],
@@ -120,7 +131,7 @@ export function providerHealthStatuses(
     const statuses = [...latestByIp.values()]
       .filter((item) => item.provider === provider)
       .map((item) => item.status)
-    result[provider] = worstHealthStatus(statuses)
+    result[provider] = bestAliveHealthStatus(statuses)
   }
 
   return result

apps/web/src/lib/schemas.ts

@@ -84,6 +84,7 @@ export const serviceDomainBindingSchema = z
     health_check_aggregate: z.enum(['any', 'all', 'majority']).catch('majority'),
     cert_monitoring: z.enum(['auto', 'required', 'skipped']).default('auto'),
     sync_status: z.string().nullable().default(null),
+    active_ips: z.array(z.string()).default([]),
   })
   .transform((binding) => ({
     ...binding,
@@ -125,6 +126,18 @@ export const healthProbeLogSchema = z.object({
   checked_at: z.string(),
 })
 
+export const failoverLogSchema = z.object({
+  id: z.number(),
+  service_id: z.number(),
+  binding_id: z.number(),
+  fqdn: z.string(),
+  ip: z.string(),
+  action: z.enum(['added', 'removed']),
+  created_at: z.string(),
+})
+
+export type FailoverLogEntry = z.infer<typeof failoverLogSchema>
+
 export const serviceViewSchema = serviceSchema.extend({
   subdomain: z.string().default(''),
   enabled: z.coerce.boolean().default(false),