fix(services): показывать живой health и историю выхода из пула

Таблица IP брала hysteresis unknown вместо последней пробы. Failover теперь показывает выход и возврат в пул, а не только DNS add.

Co-authored-by: Cursor <cursoragent@cursor.com>

apps/api/src/routes/services.ts

@@ -78,6 +78,18 @@ export async function serviceRoutes(app: FastifyInstance) {
     };
   });
 
+  app.get("/services/:id/failover-log", async (request) => {
+    const { id } = request.params as { id: string };
+    repos.getService(request.server.db, Number(id));
+    return {
+      items: repos.listFailoverLogForService(
+        request.server.db,
+        Number(id),
+        200,
+      ),
+    };
+  });
+
   app.get("/services/:id/certificates", async (request) => {
     const { id } = request.params as { id: string };
     return certificateService.listServiceCertificates(

apps/api/src/services/service-config-service.ts

@@ -39,6 +39,43 @@ import {
 export type { LbIpRow, LbTargetConfig };
 export { selectActiveIpsByMode };
 
+export function failoverARecordDiff(
+  existingA: readonly string[],
+  desiredIps: readonly string[],
+): { added: string[]; removed: string[] } {
+  const before = new Set(existingA);
+  const after = new Set(desiredIps);
+  return {
+    added: desiredIps.filter((ip) => !before.has(ip)),
+    removed: existingA.filter((ip) => !after.has(ip)),
+  };
+}
+
+function recordFailoverDnsDiff(
+  db: Db,
+  bindingId: number,
+  hostname: string,
+  zoneName: string,
+  existingRecords: DnsRecord[],
+  desiredIps: string[],
+): void {
+  const existingA = existingRecords
+    .filter((record) => record.record_type.toUpperCase() === "A")
+    .map((record) => record.content);
+  const { added, removed } = failoverARecordDiff(existingA, desiredIps);
+  if (added.length === 0 && removed.length === 0) return;
+  const binding = repos.getBinding(db, bindingId);
+  repos.insertFailoverLog(db, {
+    serviceId: binding.service_id,
+    bindingId,
+    fqdn: fqdnToDisplay(hostname, zoneName),
+    entries: [
+      ...added.map((ip) => ({ ip, action: "added" as const })),
+      ...removed.map((ip) => ({ ip, action: "removed" as const })),
+    ],
+  });
+}
+
 export interface ServiceDomainInput {
   fqdn: string;
   target_ips?: string[];
@@ -285,6 +322,11 @@ async function buildView(db: Db, serviceId: number): Promise<ServiceView> {
       if (target_ip_priorities[ip] === undefined) target_ip_priorities[ip] = 1;
     }
 
+    const { config, rows } = getBindingLbState(db, binding.id);
+    const bindingActiveIps = targetCname
+      ? []
+      : selectActiveIpsByMode(config, rows);
+
     return {
       binding_id: binding.id,
       domain_id: binding.domain_id,
@@ -312,13 +354,13 @@ async function buildView(db: Db, serviceId: number): Promise<ServiceView> {
       health_check_aggregate: binding.health_check_aggregate ?? "majority",
       cert_monitoring: binding.cert_monitoring ?? "auto",
       sync_status: aggregateSyncStatus(statuses),
+      active_ips: bindingActiveIps,
     };
   });
 
   const activeIps = new Set<string>();
-  for (const binding of bindings) {
-    const { config, rows } = getBindingLbState(db, binding.id);
-    for (const ip of selectActiveIpsByMode(config, rows)) {
+  for (const domain of domainViews) {
+    for (const ip of domain.active_ips) {
       activeIps.add(ip);
     }
   }
@@ -677,6 +719,14 @@ async function syncBindingADns(
 
   if (desiredIps.length === 0) {
     repos.setBindingDnsRecordId(db, bindingId, null);
+    recordFailoverDnsDiff(
+      db,
+      bindingId,
+      hostname,
+      zoneName,
+      existingRecords,
+      desiredIps,
+    );
     return;
   }
 
@@ -725,6 +775,14 @@ async function syncBindingADns(
   }
 
   repos.setBindingDnsRecordId(db, bindingId, primaryId);
+  recordFailoverDnsDiff(
+    db,
+    bindingId,
+    hostname,
+    zoneName,
+    existingRecords,
+    desiredIps,
+  );
 }
 
 async function cleanupBindingDns(

apps/api/test/failover-log.test.ts

@@ -0,0 +1,112 @@
+import { describe, expect, it } from "vitest";
+import { createMemoryDb, repos, runMigrations } from "@cfdm/db";
+import { buildApp } from "../src/app.js";
+import { loadConfig } from "../src/config.js";
+import { failoverARecordDiff } from "../src/services/service-config-service.js";
+
+async function authHeaders(app: Awaited<ReturnType<typeof buildApp>>) {
+  const config = loadConfig();
+  const res = await app.inject({
+    method: "POST",
+    url: "/api/v1/auth/login",
+    payload: { username: config.adminUsername, password: "admin" },
+  });
+  expect(res.statusCode).toBe(200);
+  const { token } = res.json() as { token: string };
+  return { authorization: `Bearer ${token}` };
+}
+
+describe("failoverARecordDiff", () => {
+  it("diffs added and removed A contents", () => {
+    expect(
+      failoverARecordDiff(
+        ["130.49.213.153", "93.115.203.183"],
+        ["93.115.203.183"],
+      ),
+    ).toEqual({
+      added: [],
+      removed: ["130.49.213.153"],
+    });
+    expect(failoverARecordDiff(["10.0.0.1"], ["10.0.0.1", "10.0.0.2"])).toEqual({
+      added: ["10.0.0.2"],
+      removed: [],
+    });
+  });
+});
+
+describe("GET /services/:id/failover-log", () => {
+  it("returns add/remove rows for the service", async () => {
+    const app = await buildApp({
+      config: { ...loadConfig(), staticDir: null },
+      memory: true,
+    });
+    const headers = await authHeaders(app);
+
+    const domain = repos.createDomain(app.db, null, "rkns.top", "zone-1");
+    const service = repos.createService(app.db, "MSK Hip", "msk-hip");
+    const binding = repos.insertBinding(app.db, domain.id, service.id, "gt", null);
+
+    repos.insertFailoverLog(app.db, {
+      serviceId: service.id,
+      bindingId: binding.id,
+      fqdn: "gt.rkns.top",
+      entries: [
+        { ip: "130.49.213.153", action: "removed" },
+        { ip: "93.115.203.183", action: "added" },
+      ],
+    });
+
+    const res = await app.inject({
+      method: "GET",
+      url: `/api/v1/services/${service.id}/failover-log`,
+      headers,
+    });
+    expect(res.statusCode).toBe(200);
+    const body = res.json() as {
+      items: Array<{ ip: string; fqdn: string; action: string }>;
+    };
+    expect(body.items).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          ip: "130.49.213.153",
+          fqdn: "gt.rkns.top",
+          action: "removed",
+        }),
+        expect.objectContaining({
+          ip: "93.115.203.183",
+          fqdn: "gt.rkns.top",
+          action: "added",
+        }),
+      ]),
+    );
+
+    await app.close();
+  });
+});
+
+describe("failover_log table", () => {
+  it("lists newest first", () => {
+    const { db, sqlite } = createMemoryDb();
+    runMigrations(sqlite);
+
+    const domain = repos.createDomain(db, null, "example.com", "zone-1");
+    const service = repos.createService(db, "Panel", "panel");
+    const binding = repos.insertBinding(db, domain.id, service.id, "panel", null);
+
+    repos.insertFailoverLog(db, {
+      serviceId: service.id,
+      bindingId: binding.id,
+      fqdn: "panel.example.com",
+      entries: [{ ip: "1.1.1.1", action: "removed" }],
+    });
+    repos.insertFailoverLog(db, {
+      serviceId: service.id,
+      bindingId: binding.id,
+      fqdn: "panel.example.com",
+      entries: [{ ip: "1.1.1.1", action: "added" }],
+    });
+
+    const rows = repos.listFailoverLogForService(db, service.id);
+    expect(rows.map((row) => row.action)).toEqual(["added", "removed"]);
+  });
+});

apps/api/test/health-check.test.ts

@@ -397,4 +397,43 @@ describe("CNAME health mapped onto service IPs", () => {
       }),
     ]);
   });
+
+  it("getView is up when any binding IP is up", async () => {
+    const { createMemoryDb, repos, runMigrations } = await import("@cfdm/db");
+    const { getView } = await import("../src/services/service-config-service.js");
+    const { db, sqlite } = createMemoryDb();
+    runMigrations(sqlite);
+
+    const domain = repos.createDomain(db, null, "rkns.top", "zone-id");
+    const service = repos.createService(db, "MSK Hip", "msk-hip");
+    repos.replaceServiceIps(db, service.id, ["10.0.0.1", "10.0.0.2"]);
+    const binding = repos.insertBinding(db, domain.id, service.id, "gt", null);
+    repos.replaceBindingIpsWithMeta(db, binding.id, [
+      { ip: "10.0.0.1", weight: 1, priority: 1 },
+      { ip: "10.0.0.2", weight: 1, priority: 1 },
+    ]);
+    repos.upsertIpHealthStatus(
+      db,
+      "binding",
+      binding.id,
+      "10.0.0.1",
+      "up",
+      12,
+      0,
+      null,
+    );
+    repos.upsertIpHealthStatus(
+      db,
+      "binding",
+      binding.id,
+      "10.0.0.2",
+      "down",
+      null,
+      5,
+      "timeout",
+    );
+
+    const view = await getView(db, service.id);
+    expect(view.health_status).toBe("up");
+  });
 });

apps/web/src/components/failover-timeline.tsx

@@ -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,
+  type FailoverHistoryItem,
+} from '@/lib/failover-events'
 import { cn } from '@cfdm/ui/lib/utils'
 import type { ComponentProps } from 'react'
 
@@ -29,32 +33,38 @@ 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'
 }
 
 /**
- * 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 FailoverHistoryItem[]
+}) {
+  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 выходов из пула"
         stackedIcon={false}
         centered={false}
       />
@@ -62,55 +72,90 @@ 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>
+                    <HealthCheckBadge
+                      status={item.action === 'added' ? 'up' : 'down'}
+                      size="xs"
+                    />
+                    <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">{item.copy}</p>
+                </TimelineContent>
+              </TimelineItem>
+            )
+          })}
+        </Timeline>
+      ) : null}
+    </div>
   )
 }

apps/web/src/components/reui-kit/service-failover-panel.tsx

@@ -2,9 +2,18 @@ import { UnplugIcon } from 'lucide-react'
 
 import { FailoverTimeline } from '@/components/failover-timeline'
 import {
+  mergeFailoverHistory,
   toFailoverEvents,
+  type FailoverBindingPool,
   type FailoverHealthInput,
 } from '@/lib/failover-events'
+import {
+  latestHealthByIp,
+  resolveIpDisplayHealth,
+  type HealthLogProbe,
+  type HealthLogStatus,
+} from '@/lib/health-log'
+import type { FailoverLogEntry } from '@/lib/schemas'
 import { Badge } from '@/components/reui/badge'
 import {
   Frame,
@@ -22,15 +31,15 @@ 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 + кто вышел из пула и кто вернулся.
  * 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 +49,35 @@ function failoverCountLabel(count: number): string {
  */
 export function ServiceFailoverPanel({
   ipHealth,
-  activeAddresses,
+  bindings,
+  history,
+  probes = [],
 }: {
   ipHealth: readonly FailoverHealthInput[]
-  activeAddresses: readonly string[]
+  bindings: readonly FailoverBindingPool[]
+  history: readonly FailoverLogEntry[]
+  probes?: readonly HealthLogProbe[]
 }) {
-  const events = toFailoverEvents(ipHealth, activeAddresses)
+  const liveByIp = latestHealthByIp(probes)
+  const overlayHealth = ipHealth.map((row) => {
+    const live = liveByIp.get(row.ip)
+    return {
+      ...row,
+      status: resolveIpDisplayHealth(
+        row.status as HealthLogStatus,
+        live?.status,
+      ),
+      last_error:
+        live && live.status !== 'unknown' ? live.last_error : row.last_error,
+      last_checked_at:
+        live && live.status !== 'unknown'
+          ? live.last_checked_at
+          : row.last_checked_at,
+    }
+  })
+  const events = toFailoverEvents(overlayHealth, bindings)
+  const mergedHistory = mergeFailoverHistory(history, probes, bindings)
+  const removedCount = events.filter((event) => event.kind === 'removed').length
 
   return (
     <Frame stacked spacing="sm" className="min-w-0 w-full">
@@ -64,7 +96,7 @@ export function ServiceFailoverPanel({
             )}
           </FrameTitle>
           <FrameDescription>
-            Только down из IP Health вне DNS-пула · как таблица активов
+            Текущие Down и история: кто вышел из пула и кто вернулся
           </FrameDescription>
         </FrameHeader>
 
@@ -73,12 +105,14 @@ 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={mergedHistory} />
       </FramePanel>
     </Frame>
   )

apps/web/src/components/reui-kit/uptime-chart.test.ts

@@ -45,6 +45,28 @@ describe('toAlignedSeries', () => {
     expect(keys).toEqual(['local', 'cloudflare', 'globalping'])
   })
 
+  it('keeps live latency when another IP in the same bucket is down', () => {
+    const { points } = toAlignedSeries([
+      probe({
+        id: 1,
+        latency_ms: 18,
+        checked_at: '2026-01-01T00:00:10.000Z',
+      }),
+      probe({
+        id: 2,
+        status: 'down',
+        ok: false,
+        latency_ms: null,
+        checked_at: '2026-01-01T00:00:12.000Z',
+      }),
+    ])
+
+    expect(points).toHaveLength(1)
+    expect(points[0]?.local).toBe(18)
+    expect(points[0]?.localOk).toBe(true)
+    expect(points[0]?.ok).toBe(true)
+  })
+
   it('does not plot down probes as latency 0', () => {
     const { points } = toAlignedSeries([
       probe({

apps/web/src/components/reui-kit/uptime-chart.tsx

@@ -124,8 +124,13 @@ export function toAlignedSeries(items: UptimeProbe[]): {
     }
 
     const ok = probeOk(item)
-    row[`${key}Ok`] = ok
-    row[key] = ok ? item.latency_ms : null
+    const prevOk = row[`${key}Ok`]
+    row[`${key}Ok`] = prevOk === true || ok
+    if (ok && item.latency_ms != null) {
+      row[key] = item.latency_ms
+    } else if (row[key] === undefined) {
+      row[key] = null
+    }
   }
 
   const points = [...buckets.entries()]

apps/web/src/components/services/service-detail-grid.tsx

@@ -18,6 +18,11 @@ import { DataGridColumnHeader } from '@/components/reui/data-grid/data-grid-colu
 import { IconTile } from '@/components/reui/icon-tile'
 import { createFilter, type Filter, type FilterFieldConfig } from '@/components/reui/filters'
 import { ResourcePage } from '@/components/reui-kit'
+import {
+  latestHealthByIp,
+  resolveIpDisplayHealth,
+  type HealthLogProbe,
+} from '@/lib/health-log'
 import { certRelativeBadge } from '@/components/columns/certificates-columns'
 import { certMonitoringOptions } from '@/lib/cert-monitoring'
 import { formatDate } from '@/lib/format'
@@ -99,8 +104,12 @@ function mapNodeHealth(status: string): HealthStatus {
   return 'unknown'
 }
 
-function buildIpRows(service: ServiceView): ServiceIpRow[] {
+function buildIpRows(
+  service: ServiceView,
+  probes: readonly HealthLogProbe[] = [],
+): ServiceIpRow[] {
   const healthByIp = new Map(service.ip_health.map((row) => [row.ip, row]))
+  const liveByIp = latestHealthByIp(probes)
   const weights = Object.assign(
     {},
     ...service.domains.map((domain) => domain.target_ip_weights ?? {}),
@@ -113,19 +122,25 @@ function buildIpRows(service: ServiceView): ServiceIpRow[] {
 
   return service.ips.map((ip) => {
     const health = healthByIp.get(ip)
+    const live = liveByIp.get(ip)
+    const status = resolveIpDisplayHealth(health?.status, live?.status)
+    const extras = live && live.status !== 'unknown' ? live : health
     return {
       id: ip,
       ip,
-      status: health?.status ?? 'unknown',
+      status,
       enabled: service.ip_enabled[ip] !== false,
       active: activeSet.has(ip),
       weight: weights[ip] ?? 1,
       priority: priorities[ip] ?? 1,
-      latency_ms: health?.latency_ms ?? null,
-      last_checked_at: health?.last_checked_at ?? null,
-      last_error: health?.last_error ?? null,
-      colo: health?.colo ?? null,
-      provider: health?.provider ?? null,
+      latency_ms: extras?.latency_ms ?? null,
+      last_checked_at: extras?.last_checked_at ?? null,
+      last_error:
+        live && live.status !== 'unknown'
+          ? live.last_error
+          : (health?.last_error ?? null),
+      colo: extras?.colo ?? null,
+      provider: extras?.provider ?? null,
     }
   })
 }
@@ -151,17 +166,23 @@ function buildNodeRows(
     weight: number
     priority: number
   }>,
+  probes: readonly HealthLogProbe[] = [],
 ): ServiceNodeRow[] {
-  return nodes.map((node) => ({
-    id: String(node.id),
-    nodeId: node.id,
-    address: node.address,
-    protocol: node.protocol,
-    port: node.port,
-    health_status: mapNodeHealth(node.health_status),
-    weight: node.weight,
-    priority: node.priority,
-  }))
+  const liveByIp = latestHealthByIp(probes)
+  return nodes.map((node) => {
+    const stored = mapNodeHealth(node.health_status)
+    const live = liveByIp.get(node.address)
+    return {
+      id: String(node.id),
+      nodeId: node.id,
+      address: node.address,
+      protocol: node.protocol,
+      port: node.port,
+      health_status: resolveIpDisplayHealth(stored, live?.status),
+      weight: node.weight,
+      priority: node.priority,
+    }
+  })
 }
 
 function NameCell({
@@ -199,6 +220,7 @@ interface ServiceDetailGridProps {
     weight: number
     priority: number
   }>
+  probes?: readonly HealthLogProbe[]
   togglingIp: string | null
   onToggleIp: (ip: string, enabled: boolean) => void
   onChangeIp: (row: ServiceFqdnRow) => void
@@ -211,6 +233,7 @@ interface ServiceDetailGridProps {
 export function ServiceDetailGrid({
   service,
   nodes,
+  probes = [],
   togglingIp,
   onToggleIp,
   onChangeIp,
@@ -283,9 +306,9 @@ export function ServiceDetailGrid({
     },
   })
 
-  const ipRows = useMemo(() => buildIpRows(service), [service])
+  const ipRows = useMemo(() => buildIpRows(service, probes), [service, probes])
   const fqdnRows = useMemo(() => buildFqdnRows(service), [service])
-  const nodeRows = useMemo(() => buildNodeRows(nodes), [nodes])
+  const nodeRows = useMemo(() => buildNodeRows(nodes, probes), [nodes, probes])
   const markActive = service.lb_mode === 'failover' || service.lb_mode === 'weighted'
 
   const tabs = TABS.map((entry) => ({

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

@@ -1,10 +1,16 @@
 import { describe, expect, it } from 'vitest'
 
 import {
+  failoverEventCopy,
   isFailoverEventStatus,
+  mergeFailoverHistory,
   toFailoverEvents,
+  toIpAliveTransitions,
+  type FailoverBindingPool,
   type FailoverHealthInput,
 } from '@/lib/failover-events'
+import type { FailoverLogEntry } from '@/lib/schemas'
+import type { HealthLogProbe } from '@/lib/health-log'
 
 function row(
   overrides: Partial<FailoverHealthInput> & Pick<FailoverHealthInput, 'ip'>,
@@ -18,6 +24,24 @@ 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 +51,7 @@ 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 +60,19 @@ 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 +81,13 @@ 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 +99,18 @@ 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 +118,46 @@ 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 +168,21 @@ 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',
@@ -103,3 +190,105 @@ describe('toFailoverEvents', () => {
     ])
   })
 })
+
+function probe(
+  overrides: Partial<HealthLogProbe> & Pick<HealthLogProbe, 'id' | 'status' | 'checked_at'>,
+): HealthLogProbe {
+  return {
+    ip: '130.49.213.153',
+    provider: 'local',
+    ok: overrides.status === 'up',
+    latency_ms: 12,
+    colo: null,
+    error: null,
+    ...overrides,
+  }
+}
+
+function dns(
+  overrides: Partial<FailoverLogEntry> & Pick<FailoverLogEntry, 'id' | 'action' | 'created_at'>,
+): FailoverLogEntry {
+  return {
+    service_id: 13,
+    binding_id: 1,
+    fqdn: 'gt.rkns.top',
+    ip: '130.49.213.153',
+    ...overrides,
+  }
+}
+
+describe('toIpAliveTransitions', () => {
+  it('emits leave then return, not the initial up', () => {
+    const transitions = toIpAliveTransitions([
+      probe({ id: 1, status: 'up', checked_at: '2026-08-20T09:00:00Z' }),
+      probe({ id: 2, status: 'up', checked_at: '2026-08-20T09:01:00Z' }),
+      probe({ id: 3, status: 'down', checked_at: '2026-08-20T09:02:00Z' }),
+      probe({ id: 4, status: 'down', checked_at: '2026-08-20T09:03:00Z' }),
+      probe({ id: 5, status: 'up', checked_at: '2026-08-20T09:04:00Z' }),
+    ])
+    expect(transitions).toEqual([
+      { id: 'probe:3', ip: '130.49.213.153', alive: false, at: '2026-08-20T09:02:00Z' },
+      { id: 'probe:5', ip: '130.49.213.153', alive: true, at: '2026-08-20T09:04:00Z' },
+    ])
+  })
+
+  it('any-up across providers: leave only when every source is down', () => {
+    const transitions = toIpAliveTransitions([
+      probe({
+        id: 1,
+        provider: 'local',
+        status: 'up',
+        checked_at: '2026-08-20T09:00:00Z',
+      }),
+      probe({
+        id: 2,
+        provider: 'cloudflare',
+        status: 'down',
+        checked_at: '2026-08-20T09:01:00Z',
+      }),
+      probe({
+        id: 3,
+        provider: 'local',
+        status: 'down',
+        checked_at: '2026-08-20T09:02:00Z',
+      }),
+      probe({
+        id: 4,
+        provider: 'local',
+        status: 'up',
+        checked_at: '2026-08-20T09:03:00Z',
+      }),
+    ])
+    expect(transitions.map((item) => item.id)).toEqual(['probe:3', 'probe:4'])
+  })
+})
+
+describe('mergeFailoverHistory', () => {
+  it('fills leave/return from probes when DNS has only the add', () => {
+    const history = mergeFailoverHistory(
+      [dns({ id: 10, action: 'added', created_at: '2026-08-20 09:26:00' })],
+      [
+        probe({ id: 1, status: 'up', checked_at: '2026-08-20T09:00:00Z' }),
+        probe({ id: 2, status: 'down', checked_at: '2026-08-20T09:10:00Z' }),
+        probe({ id: 3, status: 'up', checked_at: '2026-08-20T09:26:30Z' }),
+      ],
+      mskHip,
+    )
+    expect(history.map((item) => `${item.action}:${item.source}`)).toEqual([
+      'added:dns',
+      'removed:probe',
+    ])
+    expect(history[0]?.copy).toBe('130.49.213.153 добавлена на gt.rkns.top')
+    expect(history[1]?.copy).toContain('вышла из пула')
+  })
+
+  it('keeps DNS over a probe return in the same 2-minute window', () => {
+    const history = mergeFailoverHistory(
+      [dns({ id: 10, action: 'added', created_at: '2026-08-20 09:26:00' })],
+      [probe({ id: 3, status: 'up', checked_at: '2026-08-20T09:26:30Z' })],
+      mskHip,
+    )
+    expect(history).toHaveLength(1)
+    expect(history[0]?.source).toBe('dns')
+  })
+})

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

@@ -1,7 +1,29 @@
+import {
+  bestAliveHealthStatus,
+  probeTime,
+  type HealthLogProbe,
+  type HealthLogStatus,
+} from '@/lib/health-log'
+import type { FailoverLogEntry } from '@/lib/schemas'
+
+export type FailoverEventKind = 'removed' | 'last-resort'
+
+export interface FailoverHistoryItem {
+  id: string
+  ip: string
+  fqdn: string
+  action: 'added' | 'removed'
+  created_at: string
+  copy: string
+  source: 'dns' | 'probe'
+}
+
 export interface FailoverEvent {
   id: string
   address: string
   status: string
+  kind: FailoverEventKind
+  fqdns: string[]
   consecutiveFailures: number
   lastFailureReason: string | null
   lastCheckAt?: string | null
@@ -16,31 +38,194 @@ 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,
-    }))
+    }
+  })
+}
+
+const DEDUPE_WINDOW_MS = 2 * 60 * 1000
+
+function fqdnsForIp(
+  ip: string,
+  bindings: readonly FailoverBindingPool[],
+): string[] {
+  return bindings.filter((binding) => binding.configured.includes(ip)).map((binding) => binding.fqdn)
+}
+
+function fqdnLabel(fqdns: string[]): string {
+  return fqdns.join(', ') || 'пул'
+}
+
+function isAliveStatus(status: HealthLogStatus): boolean {
+  return status === 'up' || status === 'degraded'
+}
+
+/**
+ * Per-IP any-up flips: down → вышла из пула, up после down → вернулась.
+ * Initial state is not an event.
+ */
+export function toIpAliveTransitions(
+  probes: readonly HealthLogProbe[],
+): Array<{ id: string; ip: string; alive: boolean; at: string }> {
+  const byIp = new Map<string, HealthLogProbe[]>()
+  for (const item of probes) {
+    const list = byIp.get(item.ip)
+    if (list) list.push(item)
+    else byIp.set(item.ip, [item])
+  }
+
+  const out: Array<{ id: string; ip: string; alive: boolean; at: string }> = []
+  for (const [ip, list] of byIp) {
+    list.sort(
+      (a, b) => probeTime(a.checked_at) - probeTime(b.checked_at) || a.id - b.id,
+    )
+    const latestByProvider = new Map<string, HealthLogProbe>()
+    let prevAlive: boolean | undefined
+    for (const probe of list) {
+      latestByProvider.set(probe.provider, probe)
+      const status = bestAliveHealthStatus(
+        [...latestByProvider.values()].map((item) => item.status),
+      )
+      if (status === 'unknown') continue
+      const alive = isAliveStatus(status)
+      if (prevAlive !== undefined && alive !== prevAlive) {
+        out.push({
+          id: `probe:${probe.id}`,
+          ip,
+          alive,
+          at: probe.checked_at,
+        })
+      }
+      prevAlive = alive
+    }
+  }
+  return out
+}
+
+function dnsHistoryItem(item: FailoverLogEntry): FailoverHistoryItem {
+  return {
+    id: `dns:${item.id}`,
+    ip: item.ip,
+    fqdn: item.fqdn,
+    action: item.action,
+    created_at: item.created_at,
+    copy:
+      item.action === 'removed'
+        ? `${item.ip} убрана с ${item.fqdn}`
+        : `${item.ip} добавлена на ${item.fqdn}`,
+    source: 'dns',
+  }
+}
+
+function probeHistoryItem(
+  transition: { id: string; ip: string; alive: boolean; at: string },
+  bindings: readonly FailoverBindingPool[],
+): FailoverHistoryItem {
+  const fqdns = fqdnsForIp(transition.ip, bindings)
+  const fqdn = fqdnLabel(fqdns)
+  const action = transition.alive ? 'added' : 'removed'
+  return {
+    id: transition.id,
+    ip: transition.ip,
+    fqdn,
+    action,
+    created_at: transition.at,
+    copy:
+      action === 'removed'
+        ? `${transition.ip} вышла из пула (${fqdn})`
+        : `${transition.ip} вернулась в пул (${fqdn})`,
+    source: 'probe',
+  }
+}
+
+function eventTime(value: string): number {
+  return probeTime(value)
+}
+
+/**
+ * DNS add/remove + health leave/return, newest first.
+ * Same IP+action within 2 minutes: keep the DNS row (it has a concrete FQDN).
+ */
+export function mergeFailoverHistory(
+  dns: readonly FailoverLogEntry[],
+  probes: readonly HealthLogProbe[],
+  bindings: readonly FailoverBindingPool[] = [],
+): FailoverHistoryItem[] {
+  const merged = [
+    ...dns.map(dnsHistoryItem),
+    ...toIpAliveTransitions(probes).map((transition) =>
+      probeHistoryItem(transition, bindings),
+    ),
+  ]
+  merged.sort(
+    (a, b) => eventTime(b.created_at) - eventTime(a.created_at) || a.id.localeCompare(b.id),
+  )
+
+  const kept: FailoverHistoryItem[] = []
+  for (const item of merged) {
+    const duplicate = kept.find(
+      (other) =>
+        other.ip === item.ip &&
+        other.action === item.action &&
+        Math.abs(eventTime(other.created_at) - eventTime(item.created_at)) <=
+          DEDUPE_WINDOW_MS,
+    )
+    if (!duplicate) {
+      kept.push(item)
+      continue
+    }
+    if (duplicate.source === 'probe' && item.source === 'dns') {
+      kept[kept.indexOf(duplicate)] = item
+    }
+  }
+  return kept
 }

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

@@ -1,9 +1,12 @@
 import { describe, expect, it } from 'vitest'
 
 import {
+  bestAliveHealthStatus,
   collapseStatusChanges,
   enabledHealthProviders,
+  latestHealthByIp,
   providerHealthStatuses,
+  resolveIpDisplayHealth,
   worstHealthStatus,
   type HealthLogProbe,
 } from '@/lib/health-log'
@@ -73,7 +76,7 @@ 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 +87,7 @@ 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 +98,51 @@ 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')
+  })
+})
+
+describe('latestHealthByIp', () => {
+  it('any-up among latest-per-provider probes', () => {
+    const items = [
+      probe({
+        id: 1,
+        ip: '130.49.213.153',
+        provider: 'local',
+        status: 'up',
+        checked_at: '2026-08-20T09:00:00Z',
+      }),
+      probe({
+        id: 2,
+        ip: '130.49.213.153',
+        provider: 'cloudflare',
+        status: 'down',
+        checked_at: '2026-08-20T09:00:01Z',
+      }),
+      probe({
+        id: 3,
+        ip: '93.115.203.183',
+        status: 'unknown',
+        checked_at: '2026-08-20T08:59:00Z',
+      }),
+    ]
+    const byIp = latestHealthByIp(items)
+    expect(byIp.get('130.49.213.153')?.status).toBe('up')
+    expect(byIp.get('93.115.203.183')?.status).toBe('unknown')
+  })
+})
+
+describe('resolveIpDisplayHealth', () => {
+  it('prefers a live probe over stored unknown', () => {
+    expect(resolveIpDisplayHealth('unknown', 'up')).toBe('up')
+    expect(resolveIpDisplayHealth('down', 'up')).toBe('up')
+    expect(resolveIpDisplayHealth('up', 'unknown')).toBe('up')
+    expect(resolveIpDisplayHealth('unknown', undefined)).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[],
@@ -120,8 +131,66 @@ 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
 }
+
+export interface IpDisplayHealth {
+  status: HealthLogStatus
+  latency_ms: number | null
+  last_checked_at: string
+  last_error: string | null
+  colo: string | null
+  provider: HealthCheckProvider
+}
+
+/**
+ * Latest probe per provider+IP, then any-up among those providers.
+ * Used by the IP table so hysteresis `unknown` in ip_health does not hide a live OK.
+ */
+export function latestHealthByIp(
+  items: readonly HealthLogProbe[],
+): Map<string, IpDisplayHealth> {
+  const latest = new Map<string, HealthLogProbe>()
+  const sorted = [...items].sort(
+    (a, b) => probeTime(b.checked_at) - probeTime(a.checked_at) || b.id - a.id,
+  )
+  for (const item of sorted) {
+    const key = `${item.provider}\0${item.ip}`
+    if (!latest.has(key)) latest.set(key, item)
+  }
+
+  const byIp = new Map<string, HealthLogProbe[]>()
+  for (const item of latest.values()) {
+    const list = byIp.get(item.ip)
+    if (list) list.push(item)
+    else byIp.set(item.ip, [item])
+  }
+
+  const result = new Map<string, IpDisplayHealth>()
+  for (const [ip, probes] of byIp) {
+    const status = bestAliveHealthStatus(probes.map((probe) => probe.status))
+    const preferred =
+      probes.find((probe) => probe.status === status) ?? probes[0]!
+    result.set(ip, {
+      status,
+      latency_ms: preferred.latency_ms,
+      last_checked_at: preferred.checked_at,
+      last_error: preferred.error,
+      colo: preferred.colo,
+      provider: preferred.provider,
+    })
+  }
+  return result
+}
+
+/** Prefer a concrete live probe over stored hysteresis `unknown`. */
+export function resolveIpDisplayHealth(
+  stored: HealthLogStatus | undefined,
+  live: HealthLogStatus | undefined,
+): HealthLogStatus {
+  if (live && live !== 'unknown') return live
+  return stored ?? live ?? 'unknown'
+}

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),

apps/web/src/queries/services.ts

@@ -1,6 +1,7 @@
 import { queryOptions } from '@tanstack/react-query'
 import { api } from '@/lib/api-client'
 import {
+  failoverLogSchema,
   healthProbeLogSchema,
   serviceBindingSchema,
   serviceGroupsResponseSchema,
@@ -89,6 +90,7 @@ export const serviceDetailKeys = {
   overview: (id: number) => [...serviceKeys.all, id, 'overview'] as const,
   nodes: (id: number) => [...serviceKeys.all, id, 'nodes'] as const,
   healthLog: (id: number) => [...serviceKeys.all, id, 'health-log'] as const,
+  failoverLog: (id: number) => [...serviceKeys.all, id, 'failover-log'] as const,
   view: (id: number) => [...serviceKeys.all, id, 'view'] as const,
 }
 
@@ -110,6 +112,15 @@ export const serviceHealthLogQueryOptions = (id: number) =>
     },
   })
 
+export const serviceFailoverLogQueryOptions = (id: number) =>
+  queryOptions({
+    queryKey: serviceDetailKeys.failoverLog(id),
+    queryFn: async () => {
+      const data = await api.get<unknown>(`/api/v1/services/${id}/failover-log`)
+      return z.object({ items: z.array(failoverLogSchema) }).parse(data)
+    },
+  })
+
 export const serviceOverviewQueryOptions = (id: number) =>
   queryOptions({
     queryKey: serviceDetailKeys.overview(id),

apps/web/src/routes/_auth/services/$serviceId/index.tsx

@@ -44,6 +44,7 @@ import {
   domainsListQueryOptions,
   serviceBindingKeys,
   serviceDetailKeys,
+  serviceFailoverLogQueryOptions,
   serviceGroupKeys,
   serviceGroupsQueryOptions,
   serviceHealthLogQueryOptions,
@@ -90,6 +91,7 @@ function ServiceDetailPage() {
   const viewQuery = useQuery(serviceViewQueryOptions(id))
   const overviewQuery = useQuery(serviceOverviewQueryOptions(id))
   const logQuery = useQuery(serviceHealthLogQueryOptions(id))
+  const failoverLogQuery = useQuery(serviceFailoverLogQueryOptions(id))
   const nodesQuery = useQuery(serviceNodesQueryOptions(id))
   const groupsQuery = useQuery(serviceGroupsQueryOptions())
   const domainsQuery = useQuery(domainsListQueryOptions())
@@ -100,6 +102,19 @@ function ServiceDetailPage() {
     () => logQuery.data?.items ?? [],
     [logQuery.data?.items],
   )
+  const failoverHistory = useMemo(
+    () => failoverLogQuery.data?.items ?? [],
+    [failoverLogQuery.data?.items],
+  )
+  const failoverBindings = useMemo(
+    () =>
+      (service?.domains ?? []).map((domain) => ({
+        fqdn: domain.fqdn,
+        configured: domain.target_ips,
+        active: domain.active_ips ?? [],
+      })),
+    [service?.domains],
+  )
   const nodes = (nodesQuery.data as OverviewPayload['nodes']) ?? []
 
   const [editOpen, setEditOpen] = useState(false)
@@ -129,6 +144,7 @@ function ServiceDetailPage() {
       queryClient.invalidateQueries({ queryKey: serviceDetailKeys.overview(id) }),
       queryClient.invalidateQueries({ queryKey: serviceDetailKeys.nodes(id) }),
       queryClient.invalidateQueries({ queryKey: serviceDetailKeys.healthLog(id) }),
+      queryClient.invalidateQueries({ queryKey: serviceDetailKeys.failoverLog(id) }),
     ])
   }
 
@@ -313,7 +329,9 @@ function ServiceDetailPage() {
             />
             <ServiceFailoverPanel
               ipHealth={service.ip_health}
-              activeAddresses={overview?.active_addresses ?? service.active_ips}
+              bindings={failoverBindings}
+              history={failoverHistory}
+              probes={logItems}
             />
           </section>
 
@@ -327,6 +345,7 @@ function ServiceDetailPage() {
             <ServiceDetailGrid
               service={service}
               nodes={nodes}
+              probes={logItems}
               togglingIp={togglingIp}
               onToggleIp={(ip, enabled) => {
                 setTogglingIp(ip)

apps/web/src/routes/_auth/services/$serviceId/route.tsx

@@ -1,6 +1,7 @@
 import { createFileRoute, Outlet } from '@tanstack/react-router'
 import { PageShell } from '@/components/page-shell'
 import {
+  serviceFailoverLogQueryOptions,
   serviceHealthLogQueryOptions,
   serviceNodesQueryOptions,
   serviceOverviewQueryOptions,
@@ -14,6 +15,7 @@ export const Route = createFileRoute('/_auth/services/$serviceId')({
       queryClient.ensureQueryData(serviceViewQueryOptions(id)),
       queryClient.ensureQueryData(serviceOverviewQueryOptions(id)),
       queryClient.ensureQueryData(serviceHealthLogQueryOptions(id)),
+      queryClient.ensureQueryData(serviceFailoverLogQueryOptions(id)),
       queryClient.ensureQueryData(serviceNodesQueryOptions(id)),
     ])
     return { breadcrumb: view.name }

packages/db/dist/index.js

@@ -355,6 +355,15 @@ var notificationLog = sqliteTable("notification_log", {
   message: text("message").notNull(),
   created_at: text("created_at").notNull().default(sql`datetime('now')`)
 });
+var failoverLog = sqliteTable("failover_log", {
+  id: integer("id").primaryKey({ autoIncrement: true }),
+  service_id: integer("service_id").notNull().references(() => services.id, { onDelete: "cascade" }),
+  binding_id: integer("binding_id").notNull().references(() => serviceBindings.id, { onDelete: "cascade" }),
+  fqdn: text("fqdn").notNull(),
+  ip: text("ip").notNull(),
+  action: text("action").notNull(),
+  created_at: text("created_at").notNull().default(sql`datetime('now')`)
+});
 var auditLog = sqliteTable("audit_log", {
   id: text("id").primaryKey(),
   event_id: text("event_id"),
@@ -395,6 +404,7 @@ var schema = {
   domainMonitorResults,
   healthProbeLog,
   notificationLog,
+  failoverLog,
   auditLog
 };
 
@@ -715,6 +725,7 @@ __export(repos_exports, {
   getSyncJob: () => getSyncJob,
   insertBinding: () => insertBinding,
   insertDnsRecord: () => insertDnsRecord,
+  insertFailoverLog: () => insertFailoverLog,
   insertHealthProbeLog: () => insertHealthProbeLog,
   insertNotificationLog: () => insertNotificationLog,
   linkBindingRecord: () => linkBindingRecord,
@@ -738,6 +749,7 @@ __export(repos_exports, {
   listDomains: () => listDomains,
   listDomainsEnriched: () => listDomainsEnriched,
   listEnabledDomainMonitors: () => listEnabledDomainMonitors,
+  listFailoverLogForService: () => listFailoverLogForService,
   listGroupDnsRecords: () => listGroupDnsRecords,
   listGroups: () => listGroups,
   listHealthCheckTargets: () => listHealthCheckTargets,
@@ -1995,6 +2007,12 @@ var WORST_HEALTH_SQL = sql2.raw(`CASE
   END) = 1 THEN 'up'
   ELSE 'unknown'
 END`);
+var BEST_ALIVE_HEALTH_SQL = sql2.raw(`CASE
+  WHEN MAX(CASE WHEN status = 'up' THEN 1 ELSE 0 END) = 1 THEN 'up'
+  WHEN MAX(CASE WHEN status = 'degraded' THEN 1 ELSE 0 END) = 1 THEN 'degraded'
+  WHEN MAX(CASE WHEN status = 'down' THEN 1 ELSE 0 END) = 1 THEN 'down'
+  ELSE 'unknown'
+END`);
 function aggregateIpHealthByRefs(db, scope, refIds) {
   const result = /* @__PURE__ */ new Map();
   if (refIds.length === 0) return result;
@@ -2055,7 +2073,7 @@ function aggregateIpHealthByServiceIds(db, serviceIds) {
   );
   const rows = db.all(sql2`
     SELECT sb.service_id AS service_id,
-      ${WORST_HEALTH_SQL} AS health_status,
+      ${BEST_ALIVE_HEALTH_SQL} AS health_status,
       MAX(ihs.latency_ms) AS health_latency_ms
     FROM ip_health_status ihs
     INNER JOIN service_bindings sb
@@ -2545,6 +2563,26 @@ function listNotificationLog(db, limit = 50) {
     LIMIT ${limit}
   `);
 }
+function insertFailoverLog(db, input) {
+  for (const entry of input.entries) {
+    db.insert(failoverLog).values({
+      service_id: input.serviceId,
+      binding_id: input.bindingId,
+      fqdn: input.fqdn,
+      ip: entry.ip,
+      action: entry.action
+    }).run();
+  }
+}
+function listFailoverLogForService(db, serviceId, limit = 100) {
+  return db.all(sql2`
+    SELECT id, service_id, binding_id, fqdn, ip, action, created_at
+    FROM failover_log
+    WHERE service_id = ${serviceId}
+    ORDER BY created_at DESC, id DESC
+    LIMIT ${limit}
+  `);
+}
 export {
   ConflictError,
   NotFoundError,
@@ -2560,6 +2598,7 @@ export {
   domainMonitors,
   domainTags,
   domains,
+  failoverLog,
   getAppSettings,
   getAppSettingsSecrets,
   groups,

packages/db/migrations/025_failover_log.sql

@@ -0,0 +1,11 @@
+CREATE TABLE failover_log (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  service_id INTEGER NOT NULL REFERENCES services(id) ON DELETE CASCADE,
+  binding_id INTEGER NOT NULL REFERENCES service_bindings(id) ON DELETE CASCADE,
+  fqdn TEXT NOT NULL,
+  ip TEXT NOT NULL,
+  action TEXT NOT NULL CHECK (action IN ('added', 'removed')),
+  created_at TEXT NOT NULL DEFAULT (datetime('now'))
+);
+
+CREATE INDEX idx_failover_log_service ON failover_log(service_id, created_at DESC);

packages/db/src/repos.ts

@@ -49,6 +49,7 @@ import {
   nodes,
   bindingNodes,
   notificationLog,
+  failoverLog,
   serviceBindingIps,
   serviceBindingRecords,
   serviceBindings,
@@ -2163,6 +2164,14 @@ const WORST_HEALTH_SQL = sql.raw(`CASE
   ELSE 'unknown'
 END`);
 
+/** Service is up if any binding IP is up. */
+const BEST_ALIVE_HEALTH_SQL = sql.raw(`CASE
+  WHEN MAX(CASE WHEN status = 'up' THEN 1 ELSE 0 END) = 1 THEN 'up'
+  WHEN MAX(CASE WHEN status = 'degraded' THEN 1 ELSE 0 END) = 1 THEN 'degraded'
+  WHEN MAX(CASE WHEN status = 'down' THEN 1 ELSE 0 END) = 1 THEN 'down'
+  ELSE 'unknown'
+END`);
+
 /** Worst status across ip_health_status rows for each ref_id in a scope. */
 export function aggregateIpHealthByRefs(
   db: Db,
@@ -2236,7 +2245,7 @@ export function aggregateGroupScopeHealthByIds(
   return result;
 }
 
-/** Worst binding-scope health rolled up per service_id. */
+/** Binding-scope health rolled up per service_id: any up → service up. */
 export function aggregateIpHealthByServiceIds(
   db: Db,
   serviceIds: number[],
@@ -2253,7 +2262,7 @@ export function aggregateIpHealthByServiceIds(
     health_latency_ms: number | null;
   }>(sql`
     SELECT sb.service_id AS service_id,
-      ${WORST_HEALTH_SQL} AS health_status,
+      ${BEST_ALIVE_HEALTH_SQL} AS health_status,
       MAX(ihs.latency_ms) AS health_latency_ms
     FROM ip_health_status ihs
     INNER JOIN service_bindings sb
@@ -3055,3 +3064,51 @@ export function listNotificationLog(
   `);
 }
 
+export type FailoverLogAction = "added" | "removed";
+
+export type FailoverLogRow = {
+  id: number;
+  service_id: number;
+  binding_id: number;
+  fqdn: string;
+  ip: string;
+  action: FailoverLogAction;
+  created_at: string;
+};
+
+export function insertFailoverLog(
+  db: Db,
+  input: {
+    serviceId: number;
+    bindingId: number;
+    fqdn: string;
+    entries: Array<{ ip: string; action: FailoverLogAction }>;
+  },
+): void {
+  for (const entry of input.entries) {
+    db.insert(failoverLog)
+      .values({
+        service_id: input.serviceId,
+        binding_id: input.bindingId,
+        fqdn: input.fqdn,
+        ip: entry.ip,
+        action: entry.action,
+      })
+      .run();
+  }
+}
+
+export function listFailoverLogForService(
+  db: Db,
+  serviceId: number,
+  limit = 100,
+): FailoverLogRow[] {
+  return db.all<FailoverLogRow>(sql`
+    SELECT id, service_id, binding_id, fqdn, ip, action, created_at
+    FROM failover_log
+    WHERE service_id = ${serviceId}
+    ORDER BY created_at DESC, id DESC
+    LIMIT ${limit}
+  `);
+}
+

packages/db/src/schema.ts

@@ -497,6 +497,22 @@ export const notificationLog = sqliteTable("notification_log", {
     .default(sql`datetime('now')`),
 });
 
+export const failoverLog = sqliteTable("failover_log", {
+  id: integer("id").primaryKey({ autoIncrement: true }),
+  service_id: integer("service_id")
+    .notNull()
+    .references(() => services.id, { onDelete: "cascade" }),
+  binding_id: integer("binding_id")
+    .notNull()
+    .references(() => serviceBindings.id, { onDelete: "cascade" }),
+  fqdn: text("fqdn").notNull(),
+  ip: text("ip").notNull(),
+  action: text("action").notNull(),
+  created_at: text("created_at")
+    .notNull()
+    .default(sql`datetime('now')`),
+});
+
 export const auditLog = sqliteTable("audit_log", {
   id: text("id").primaryKey(),
   event_id: text("event_id"),
@@ -540,5 +556,6 @@ export const schema = {
   domainMonitorResults,
   healthProbeLog,
   notificationLog,
+  failoverLog,
   auditLog,
 };

packages/shared/dist/index.d.ts

@@ -143,6 +143,7 @@ interface ServiceDomainBindingView {
     fqdn: string;
     record_type: "A" | "CNAME";
     target_ips: string[];
+    active_ips: string[];
     target_ip_weights: Record<string, number>;
     target_ip_priorities: Record<string, number>;
     target_cname: string | null;
@@ -505,6 +506,19 @@ declare const healthProbeLogSchema: z.ZodObject<{
     checked_at: z.ZodString;
 }, z.core.$strip>;
 type HealthProbeLog = z.infer<typeof healthProbeLogSchema>;
+declare const failoverLogSchema: z.ZodObject<{
+    id: z.ZodNumber;
+    service_id: z.ZodNumber;
+    binding_id: z.ZodNumber;
+    fqdn: z.ZodString;
+    ip: z.ZodString;
+    action: z.ZodEnum<{
+        added: "added";
+        removed: "removed";
+    }>;
+    created_at: z.ZodString;
+}, z.core.$strip>;
+type FailoverLogEntry = z.infer<typeof failoverLogSchema>;
 declare const groupSchema: z.ZodObject<{
     id: z.ZodNumber;
     name: z.ZodString;
@@ -643,6 +657,7 @@ declare const serviceDomainBindingSchema: z.ZodPipe<z.ZodObject<{
         skipped: "skipped";
     }>>;
     sync_status: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+    active_ips: z.ZodDefault<z.ZodArray<z.ZodString>>;
 }, z.core.$strip>, z.ZodTransform<{
     target_ips: string[];
     target_ip_weights: Record<string, number>;
@@ -668,6 +683,7 @@ declare const serviceDomainBindingSchema: z.ZodPipe<z.ZodObject<{
     health_check_aggregate: "any" | "all" | "majority";
     cert_monitoring: "auto" | "required" | "skipped";
     sync_status: string | null;
+    active_ips: string[];
     target_ip?: string | null | undefined;
 }, {
     binding_id: number;
@@ -690,6 +706,7 @@ declare const serviceDomainBindingSchema: z.ZodPipe<z.ZodObject<{
     health_check_aggregate: "any" | "all" | "majority";
     cert_monitoring: "auto" | "required" | "skipped";
     sync_status: string | null;
+    active_ips: string[];
     target_ips?: string[] | undefined;
     target_ip?: string | null | undefined;
     target_ip_weights?: Record<string, number> | undefined;
@@ -763,6 +780,7 @@ declare const serviceViewSchema: z.ZodObject<{
             skipped: "skipped";
         }>>;
         sync_status: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+        active_ips: z.ZodDefault<z.ZodArray<z.ZodString>>;
     }, z.core.$strip>, z.ZodTransform<{
         target_ips: string[];
         target_ip_weights: Record<string, number>;
@@ -788,6 +806,7 @@ declare const serviceViewSchema: z.ZodObject<{
         health_check_aggregate: "any" | "all" | "majority";
         cert_monitoring: "auto" | "required" | "skipped";
         sync_status: string | null;
+        active_ips: string[];
         target_ip?: string | null | undefined;
     }, {
         binding_id: number;
@@ -810,6 +829,7 @@ declare const serviceViewSchema: z.ZodObject<{
         health_check_aggregate: "any" | "all" | "majority";
         cert_monitoring: "auto" | "required" | "skipped";
         sync_status: string | null;
+        active_ips: string[];
         target_ips?: string[] | undefined;
         target_ip?: string | null | undefined;
         target_ip_weights?: Record<string, number> | undefined;
@@ -965,6 +985,7 @@ declare const serviceGroupViewSchema: z.ZodObject<{
                 skipped: "skipped";
             }>>;
             sync_status: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+            active_ips: z.ZodDefault<z.ZodArray<z.ZodString>>;
         }, z.core.$strip>, z.ZodTransform<{
             target_ips: string[];
             target_ip_weights: Record<string, number>;
@@ -990,6 +1011,7 @@ declare const serviceGroupViewSchema: z.ZodObject<{
             health_check_aggregate: "any" | "all" | "majority";
             cert_monitoring: "auto" | "required" | "skipped";
             sync_status: string | null;
+            active_ips: string[];
             target_ip?: string | null | undefined;
         }, {
             binding_id: number;
@@ -1012,6 +1034,7 @@ declare const serviceGroupViewSchema: z.ZodObject<{
             health_check_aggregate: "any" | "all" | "majority";
             cert_monitoring: "auto" | "required" | "skipped";
             sync_status: string | null;
+            active_ips: string[];
             target_ips?: string[] | undefined;
             target_ip?: string | null | undefined;
             target_ip_weights?: Record<string, number> | undefined;
@@ -1176,6 +1199,7 @@ declare const serviceGroupsResponseSchema: z.ZodObject<{
                     skipped: "skipped";
                 }>>;
                 sync_status: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+                active_ips: z.ZodDefault<z.ZodArray<z.ZodString>>;
             }, z.core.$strip>, z.ZodTransform<{
                 target_ips: string[];
                 target_ip_weights: Record<string, number>;
@@ -1201,6 +1225,7 @@ declare const serviceGroupsResponseSchema: z.ZodObject<{
                 health_check_aggregate: "any" | "all" | "majority";
                 cert_monitoring: "auto" | "required" | "skipped";
                 sync_status: string | null;
+                active_ips: string[];
                 target_ip?: string | null | undefined;
             }, {
                 binding_id: number;
@@ -1223,6 +1248,7 @@ declare const serviceGroupsResponseSchema: z.ZodObject<{
                 health_check_aggregate: "any" | "all" | "majority";
                 cert_monitoring: "auto" | "required" | "skipped";
                 sync_status: string | null;
+                active_ips: string[];
                 target_ips?: string[] | undefined;
                 target_ip?: string | null | undefined;
                 target_ip_weights?: Record<string, number> | undefined;
@@ -1338,6 +1364,7 @@ declare const serviceGroupsResponseSchema: z.ZodObject<{
                 skipped: "skipped";
             }>>;
             sync_status: z.ZodDefault<z.ZodNullable<z.ZodString>>;
+            active_ips: z.ZodDefault<z.ZodArray<z.ZodString>>;
         }, z.core.$strip>, z.ZodTransform<{
             target_ips: string[];
             target_ip_weights: Record<string, number>;
@@ -1363,6 +1390,7 @@ declare const serviceGroupsResponseSchema: z.ZodObject<{
             health_check_aggregate: "any" | "all" | "majority";
             cert_monitoring: "auto" | "required" | "skipped";
             sync_status: string | null;
+            active_ips: string[];
             target_ip?: string | null | undefined;
         }, {
             binding_id: number;
@@ -1385,6 +1413,7 @@ declare const serviceGroupsResponseSchema: z.ZodObject<{
             health_check_aggregate: "any" | "all" | "majority";
             cert_monitoring: "auto" | "required" | "skipped";
             sync_status: string | null;
+            active_ips: string[];
             target_ips?: string[] | undefined;
             target_ip?: string | null | undefined;
             target_ip_weights?: Record<string, number> | undefined;
@@ -2412,4 +2441,4 @@ declare const ingestAuditEventSchema: z.ZodObject<{
 }, z.core.$strip>;
 type IngestAuditEvent = z.infer<typeof ingestAuditEventSchema>;
 
-export { AUDIT_SEVERITIES, AUDIT_SOURCE_APPS, AUDIT_TARGET_TYPES, type AppSettingsPatch, type AppSwitcherConfig, type AppSwitcherEntry, type AuditListQuery, type AuditLogEntry, type AuditSeverity, type AuditSourceApp, type AuditTargetType, type BulkUpdateDomainsInput, CERT_ERROR, CERT_EXPIRED, CERT_MONITORING_VALUES, CERT_MONITOR_AUTO, CERT_MONITOR_REQUIRED, CERT_MONITOR_SKIPPED, CERT_OK, CERT_UNKNOWN, CERT_WARNING, type CertMonitoring, type Certificate, type CfDnsRecord, type CfHealthCheck, type CfZone, type CfdmBindingSyncItem, type ChangeDomainInput, type ChangeIpInput, type CreateDnsRecordInput, type CreateDnsRecordPayload, type CreateDomainInput, type CreateDomainMonitorInput, type CreateGroupInput, type CreateOriginHealthCheckInput, type CreateServiceBindingInput, type CreateServiceGroupInput, type CreateServiceInput, type CreateServiceNodeInput, type CreateServiceWithConfigInput, type CreateSubdomainInput, type DnsRecord, type Domain, type DomainEnvironment, type DomainListItem, type DomainMonitor, type DomainMonitorResult, type DomainMonitorType, type Group, type GroupWithStats, HEALTH_CHECK_AGGREGATES, HEALTH_CHECK_PROVIDERS, HEALTH_KV_CURSOR_KEY, HEALTH_KV_RESULTS_KEY, HEALTH_KV_TARGETS_KEY, HEALTH_PROBE_BATCH, HEALTH_PROBE_CONCURRENCY, HEALTH_PROBE_KV_TITLE, HEALTH_PROBE_SCRIPT_NAME, HEALTH_STATUS_PROVIDERS, type HealthCheckAggregate, type HealthCheckConfig, type HealthCheckProvider, type HealthCheckScope, type HealthCheckTarget, type HealthCheckType, type HealthProbeLog, type HealthProbeResultItem, type HealthProbeResultsDoc, type HealthProbeTargetItem, type HealthProbeTargetsDoc, type HealthStatusProvider, type HealthStatusQuery, type HealthWorkerStatus, type IngestAuditEvent, type IpHealthState, type IpHealthStatus, type JwtClaims, type LbMode, type LoginInput, type LoginRequest, type LoginResponse, type NodeHealthState, type NotificationLog, type OriginHealthCheck, type OriginHealthCheckRecord, type ParsedFqdn, type PatchDnsRecordPayload, type ReorderServicesInput, SYNC_CONFLICT, SYNC_ERROR, SYNC_PENDING_DELETE, SYNC_PENDING_PUSH, SYNC_SYNCED, type Service, type ServiceBinding, type ServiceBindingView, type ServiceCertificateRow, type ServiceDomainBinding, type ServiceDomainBindingView, type ServiceGroup, type ServiceGroupView, type ServiceGroupsResponse, type ServiceIpHealth, type ServiceNode, type ServiceNodeRecord, type ServiceOverview, type ServiceView, type Subdomain, type SubdomainRecord, type SyncJob, type ToggleEnabledInput, type ToggleServiceIpInput, type UpdateDomainInput, type UpdateServiceConfigInput, type UpdateServiceGroupInput, type UpdateServiceNodeInput, type UpdateSubdomainInput, ValidationError, type VpsTrackerEvent, aggregateHealthOk, appSettingsPatchSchema, appSwitcherConfigSchema, appSwitcherEntrySchema, appSwitcherIconSchema, auditListQuerySchema, auditLogEntrySchema, auditSeveritySchema, auditSourceAppSchema, auditTargetTypeSchema, bindingToFqdn, bulkUpdateDomainsSchema, certMonitoringSchema, certStatusFromExpiry, certificateSchema, cfdmBindingSyncItemSchema, cfdmSyncBindingsBodySchema, changeDomainSchema, changeIpSchema, clampGlobalpingLimit, createDnsRecordSchema, createDomainMonitorSchema, createDomainSchema, createGroupSchema, createOriginHealthCheckSchema, createServiceBindingSchema, createServiceGroupSchema, createServiceNodeSchema, createServiceSchema, createServiceWithConfigSchema, createSubdomainSchema, derivePrimaryProvider, dnsNameToSubdomainLabel, dnsRecordNamesMatch, dnsRecordSchema, domainEnvironmentSchema, domainListItemSchema, domainMonitorResultSchema, domainMonitorSchema, domainMonitorTypeSchema, domainSchema, fqdnToDisplay, groupSchema, groupWithStatsSchema, healthCheckAggregateSchema, healthCheckConfigSchema, healthCheckProviderSchema, healthCheckProvidersSchema, healthCheckScopeSchema, healthCheckTypeSchema, healthProbeLogSchema, healthStatusProviderSchema, healthStatusQuerySchema, ingestAuditEventSchema, ipHealthStateSchema, ipHealthStatusSchema, isIpLiteral, isValidIpv4, lbModeSchema, loginSchema, nodeHealthStateSchema, normalizeDnsRecordName, normalizeProbeProvider, normalizeStatusProvider, notificationLogSchema, originHealthCheckSchema, parseFqdn, parseGlobalpingLocations, parseHealthAggregate, parseHealthProviders, reorderServicesSchema, serializeHealthProviders, serviceBindingSchema, serviceCertificateRowSchema, serviceDomainBindingSchema, serviceGroupSchema, serviceGroupTypeSchema, serviceGroupViewSchema, serviceGroupsResponseSchema, serviceIpHealthSchema, serviceNodeSchema, serviceSchema, serviceViewSchema, shouldMonitorService, subdomainLabelToFqdn, subdomainSchema, targetHasProvider, targetProviders, toggleEnabledSchema, toggleServiceIpSchema, uniqueHealthProviders, updateDomainGroupSchema, updateDomainSchema, updateServiceConfigSchema, updateServiceGroupSchema, updateServiceNodeSchema, updateSubdomainSchema, validateDnsRecord, vpsTrackerEventSchema };
+export { AUDIT_SEVERITIES, AUDIT_SOURCE_APPS, AUDIT_TARGET_TYPES, type AppSettingsPatch, type AppSwitcherConfig, type AppSwitcherEntry, type AuditListQuery, type AuditLogEntry, type AuditSeverity, type AuditSourceApp, type AuditTargetType, type BulkUpdateDomainsInput, CERT_ERROR, CERT_EXPIRED, CERT_MONITORING_VALUES, CERT_MONITOR_AUTO, CERT_MONITOR_REQUIRED, CERT_MONITOR_SKIPPED, CERT_OK, CERT_UNKNOWN, CERT_WARNING, type CertMonitoring, type Certificate, type CfDnsRecord, type CfHealthCheck, type CfZone, type CfdmBindingSyncItem, type ChangeDomainInput, type ChangeIpInput, type CreateDnsRecordInput, type CreateDnsRecordPayload, type CreateDomainInput, type CreateDomainMonitorInput, type CreateGroupInput, type CreateOriginHealthCheckInput, type CreateServiceBindingInput, type CreateServiceGroupInput, type CreateServiceInput, type CreateServiceNodeInput, type CreateServiceWithConfigInput, type CreateSubdomainInput, type DnsRecord, type Domain, type DomainEnvironment, type DomainListItem, type DomainMonitor, type DomainMonitorResult, type DomainMonitorType, type FailoverLogEntry, type Group, type GroupWithStats, HEALTH_CHECK_AGGREGATES, HEALTH_CHECK_PROVIDERS, HEALTH_KV_CURSOR_KEY, HEALTH_KV_RESULTS_KEY, HEALTH_KV_TARGETS_KEY, HEALTH_PROBE_BATCH, HEALTH_PROBE_CONCURRENCY, HEALTH_PROBE_KV_TITLE, HEALTH_PROBE_SCRIPT_NAME, HEALTH_STATUS_PROVIDERS, type HealthCheckAggregate, type HealthCheckConfig, type HealthCheckProvider, type HealthCheckScope, type HealthCheckTarget, type HealthCheckType, type HealthProbeLog, type HealthProbeResultItem, type HealthProbeResultsDoc, type HealthProbeTargetItem, type HealthProbeTargetsDoc, type HealthStatusProvider, type HealthStatusQuery, type HealthWorkerStatus, type IngestAuditEvent, type IpHealthState, type IpHealthStatus, type JwtClaims, type LbMode, type LoginInput, type LoginRequest, type LoginResponse, type NodeHealthState, type NotificationLog, type OriginHealthCheck, type OriginHealthCheckRecord, type ParsedFqdn, type PatchDnsRecordPayload, type ReorderServicesInput, SYNC_CONFLICT, SYNC_ERROR, SYNC_PENDING_DELETE, SYNC_PENDING_PUSH, SYNC_SYNCED, type Service, type ServiceBinding, type ServiceBindingView, type ServiceCertificateRow, type ServiceDomainBinding, type ServiceDomainBindingView, type ServiceGroup, type ServiceGroupView, type ServiceGroupsResponse, type ServiceIpHealth, type ServiceNode, type ServiceNodeRecord, type ServiceOverview, type ServiceView, type Subdomain, type SubdomainRecord, type SyncJob, type ToggleEnabledInput, type ToggleServiceIpInput, type UpdateDomainInput, type UpdateServiceConfigInput, type UpdateServiceGroupInput, type UpdateServiceNodeInput, type UpdateSubdomainInput, ValidationError, type VpsTrackerEvent, aggregateHealthOk, appSettingsPatchSchema, appSwitcherConfigSchema, appSwitcherEntrySchema, appSwitcherIconSchema, auditListQuerySchema, auditLogEntrySchema, auditSeveritySchema, auditSourceAppSchema, auditTargetTypeSchema, bindingToFqdn, bulkUpdateDomainsSchema, certMonitoringSchema, certStatusFromExpiry, certificateSchema, cfdmBindingSyncItemSchema, cfdmSyncBindingsBodySchema, changeDomainSchema, changeIpSchema, clampGlobalpingLimit, createDnsRecordSchema, createDomainMonitorSchema, createDomainSchema, createGroupSchema, createOriginHealthCheckSchema, createServiceBindingSchema, createServiceGroupSchema, createServiceNodeSchema, createServiceSchema, createServiceWithConfigSchema, createSubdomainSchema, derivePrimaryProvider, dnsNameToSubdomainLabel, dnsRecordNamesMatch, dnsRecordSchema, domainEnvironmentSchema, domainListItemSchema, domainMonitorResultSchema, domainMonitorSchema, domainMonitorTypeSchema, domainSchema, failoverLogSchema, fqdnToDisplay, groupSchema, groupWithStatsSchema, healthCheckAggregateSchema, healthCheckConfigSchema, healthCheckProviderSchema, healthCheckProvidersSchema, healthCheckScopeSchema, healthCheckTypeSchema, healthProbeLogSchema, healthStatusProviderSchema, healthStatusQuerySchema, ingestAuditEventSchema, ipHealthStateSchema, ipHealthStatusSchema, isIpLiteral, isValidIpv4, lbModeSchema, loginSchema, nodeHealthStateSchema, normalizeDnsRecordName, normalizeProbeProvider, normalizeStatusProvider, notificationLogSchema, originHealthCheckSchema, parseFqdn, parseGlobalpingLocations, parseHealthAggregate, parseHealthProviders, reorderServicesSchema, serializeHealthProviders, serviceBindingSchema, serviceCertificateRowSchema, serviceDomainBindingSchema, serviceGroupSchema, serviceGroupTypeSchema, serviceGroupViewSchema, serviceGroupsResponseSchema, serviceIpHealthSchema, serviceNodeSchema, serviceSchema, serviceViewSchema, shouldMonitorService, subdomainLabelToFqdn, subdomainSchema, targetHasProvider, targetProviders, toggleEnabledSchema, toggleServiceIpSchema, uniqueHealthProviders, updateDomainGroupSchema, updateDomainSchema, updateServiceConfigSchema, updateServiceGroupSchema, updateServiceNodeSchema, updateSubdomainSchema, validateDnsRecord, vpsTrackerEventSchema };

packages/shared/dist/index.js

@@ -322,6 +322,15 @@ var healthProbeLogSchema = z.object({
   error: z.string().nullable(),
   checked_at: z.string()
 });
+var 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()
+});
 var groupSchema = z.object({
   id: z.number(),
   name: z.string(),
@@ -399,7 +408,8 @@ var serviceDomainBindingSchema = z.object({
   health_check_providers: healthCheckProvidersSchema.catch(["local"]),
   health_check_aggregate: healthCheckAggregateSchema.catch("majority"),
   cert_monitoring: certMonitoringSchema.default("auto"),
-  sync_status: z.string().nullable().default(null)
+  sync_status: z.string().nullable().default(null),
+  active_ips: z.array(z.string()).default([])
 }).transform((binding) => ({
   ...binding,
   target_ips: binding.target_ips && binding.target_ips.length > 0 ? binding.target_ips : binding.target_ip ? [binding.target_ip] : [],
@@ -1019,6 +1029,7 @@ export {
   domainMonitorSchema,
   domainMonitorTypeSchema,
   domainSchema,
+  failoverLogSchema,
   fqdnToDisplay,
   groupSchema,
   groupWithStatsSchema,

packages/shared/src/schemas.ts

@@ -100,6 +100,18 @@ export const healthProbeLogSchema = z.object({
 
 export type HealthProbeLog = z.infer<typeof healthProbeLogSchema>
 
+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 groupSchema = z.object({
   id: z.number(),
   name: z.string(),
@@ -184,6 +196,7 @@ export const serviceDomainBindingSchema = z
     health_check_aggregate: healthCheckAggregateSchema.catch('majority'),
     cert_monitoring: certMonitoringSchema.default('auto'),
     sync_status: z.string().nullable().default(null),
+    active_ips: z.array(z.string()).default([]),
   })
   .transform((binding) => ({
     ...binding,

packages/shared/src/types.ts

@@ -190,6 +190,7 @@ export interface ServiceDomainBindingView {
   fqdn: string;
   record_type: "A" | "CNAME";
   target_ips: string[];
+  active_ips: string[];
   target_ip_weights: Record<string, number>;
   target_ip_priorities: Record<string, number>;
   target_cname: string | null;