fix(services): не балансировать сервис с одним IP

KPI и карточка показывают живой OK, а не гистерезис unknown; DNS по-прежнему ждёт повторные успехи.

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

apps/api/src/app.ts

@@ -38,6 +38,10 @@ import {
   healthEngineFallbacksFromConfig,
   scheduleHealthCheckJob,
 } from "./services/health-check-scheduler.js";
+import {
+  createWeightedDnsTask,
+  scheduleWeightedDnsJob,
+} from "./services/weighted-dns-scheduler.js";
 import { fireEnsureHealthWorker } from "./services/health/health-worker-deploy.js";
 import { AsyncTask, CronJob } from "toad-scheduler";
 
@@ -133,6 +137,7 @@ export async function buildApp(opts: BuildAppOptions = {}) {
     app.decorate("reloadHealthCheckJob", () => {
       scheduleHealthCheckJob(app, config, healthTask);
     });
+    scheduleWeightedDnsJob(app, createWeightedDnsTask(app));
     if (config.cloudflareApiToken) {
       fireEnsureHealthWorker(
         app.db,

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/health-check-service.ts

@@ -386,7 +386,8 @@ function applyAggregatedStatus(
     { colo, provider: statusProvider },
   );
   const matchedNode = repos.findNodeByIp(db, target.ip);
-  if (matchedNode && matchedNode.enabled) {
+  // Binding-scope only: group apply must not clobber node with its own fetch failed.
+  if (matchedNode && matchedNode.enabled && target.scope === "binding") {
     repos.updateNode(db, matchedNode.id, {
       health_status: node,
       consecutive_failures: failures,

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

@@ -9,7 +9,10 @@ import type {
 import { AppError } from "../errors.js";
 import { isValidIpv4 } from "../lib/validators.js";
 import { getView } from "./service-config-service.js";
-import { selectActiveIpsByMode } from "./routing/index.js";
+import {
+  isSharedPool,
+  resolveDesiredAIps,
+} from "./routing/index.js";
 
 function assertAddress(address: string): void {
   if (!isValidIpv4(address)) {
@@ -89,8 +92,12 @@ export async function getOverview(
       : null;
 
   const active = new Set<string>();
+  const serviceIps = service.ips.filter(
+    (ip) => service.ip_enabled[ip] !== false,
+  );
   for (const binding of bindings) {
     const metas = repos.listBindingIpsWithMeta(db, binding.id);
+    const targetIps = metas.map((entry) => entry.ip);
     const rows = metas.map((entry) => {
       const status = repos.getIpHealthStatusRow(db, "binding", binding.id, entry.ip);
       return {
@@ -100,12 +107,15 @@ export async function getOverview(
         health: status ? status.status : ("unknown" as const),
       };
     });
-    for (const ip of selectActiveIpsByMode(
+    for (const ip of resolveDesiredAIps(
       {
         lb_mode: binding.lb_mode,
         health_check_enabled: binding.health_check_enabled,
       },
       rows,
+      targetIps,
+      Date.now(),
+      serviceIps,
     )) {
       active.add(ip);
     }
@@ -134,10 +144,13 @@ export function opsSummary(db: Db) {
     if (binding.lb_mode !== "failover" || !binding.health_check_enabled) {
       return false;
     }
-    return binding.target_ips.some((ip) => {
-      const row = repos.getIpHealthStatusRow(db, "binding", binding.id, ip);
-      return row?.status === "down";
-    });
+    return (
+      isSharedPool(binding.target_ips) &&
+      binding.target_ips.some((ip) => {
+        const row = repos.getIpHealthStatusRow(db, "binding", binding.id, ip);
+        return row?.status === "down";
+      })
+    );
   }).length;
   return {
     domains: domains.length,

apps/api/src/services/routing/index.ts

@@ -1,26 +1,55 @@
 import type { LbMode } from "@cfdm/shared";
 import { failoverDesired } from "./failover.js";
+import { canApplyLb, isSharedPool } from "./pool.js";
 import { roundRobinDesired } from "./round-robin.js";
 import type { LbIpRow, LbTargetConfig } from "./types.js";
+import { weightedDesired } from "./weighted.js";
 
 export type { LbIpRow, LbTargetConfig } from "./types.js";
 export { isHealthy } from "./health.js";
 export { withBindingLock } from "./binding-lock.js";
+export {
+  canApplyLb,
+  isSharedPool,
+  shouldRecordFailoverDnsDiff,
+  uniqueIpCount,
+} from "./pool.js";
+export { WEIGHTED_DNS_TTL, WEIGHTED_SLOT_MS, weightedDesired } from "./weighted.js";
 
 export function selectActiveIpsByMode(
   config: LbTargetConfig,
   rows: LbIpRow[],
+  nowMs = Date.now(),
 ): string[] {
   if (rows.length === 0) return [];
   if (config.lb_mode === "failover") {
     return failoverDesired(rows);
   }
-  // weighted = round_robin on DNS (one A per IP)
+  if (config.lb_mode === "weighted") {
+    return weightedDesired(rows, nowMs);
+  }
   return roundRobinDesired(rows);
 }
 
+export function resolveDesiredAIps(
+  config: LbTargetConfig,
+  rows: LbIpRow[],
+  fallbackIps: readonly string[],
+  nowMs = Date.now(),
+  serviceIps: readonly string[] = fallbackIps,
+): string[] {
+  const fallback = [...fallbackIps];
+  if (!canApplyLb(serviceIps, fallback)) return fallback;
+  if (!isSharedPool(rows.map((row) => row.ip))) return fallback;
+  if (config.lb_mode === "weighted" || config.health_check_enabled) {
+    const activeIps = selectActiveIpsByMode(config, rows, nowMs);
+    if (activeIps.length > 0) return activeIps;
+  }
+  return fallback;
+}
+
 export function strategyLabel(mode: LbMode): string {
   if (mode === "failover") return "Failover";
-  if (mode === "weighted") return "Round Robin (weighted alias)";
+  if (mode === "weighted") return "Weighted";
   return "Round Robin";
 }

apps/api/src/services/routing/pool.ts

@@ -0,0 +1,30 @@
+import type { LbMode } from "@cfdm/shared";
+
+export function uniqueIpCount(ips: readonly string[]): number {
+  return new Set(ips.filter(Boolean)).size;
+}
+
+/** Shared pool — two or more unique IPs. One IP (even duplicated) is not a pool. */
+export function isSharedPool(ips: readonly string[]): boolean {
+  return uniqueIpCount(ips) >= 2;
+}
+
+/** LB / drain only when the service itself has a pool AND this FQDN is shared. */
+export function canApplyLb(
+  serviceIps: readonly string[],
+  bindingIps: readonly string[],
+): boolean {
+  return isSharedPool(serviceIps) && isSharedPool(bindingIps);
+}
+
+export function shouldRecordFailoverDnsDiff(input: {
+  configuredIps: readonly string[];
+  lbMode: LbMode;
+  added: readonly string[];
+  removed: readonly string[];
+  downIps: ReadonlySet<string>;
+}): boolean {
+  if (!isSharedPool(input.configuredIps)) return false;
+  if (input.lbMode !== "weighted") return true;
+  return [...input.added, ...input.removed].some((ip) => input.downIps.has(ip));
+}

apps/api/src/services/routing/weighted.ts

@@ -0,0 +1,24 @@
+import type { LbIpRow } from "./types.js";
+import { isHealthy } from "./health.js";
+
+/** Slot length for time-sliced weighted DNS (one A at a time). */
+export const WEIGHTED_SLOT_MS = 60_000;
+
+/** Cloudflare DNS-only minimum TTL; Auto (1) is ~300s and would smear ratios. */
+export const WEIGHTED_DNS_TTL = 60;
+
+export function weightedDesired(rows: LbIpRow[], nowMs = Date.now()): string[] {
+  if (rows.length === 0) return [];
+  const healthy = rows.filter((r) => isHealthy(r.health));
+  const pool = healthy.length > 0 ? healthy : rows;
+  if (pool.length === 1) return [pool[0]!.ip];
+
+  const sorted = [...pool].sort((a, b) => a.ip.localeCompare(b.ip));
+  const cycle: string[] = [];
+  for (const row of sorted) {
+    const weight = Math.max(1, Math.round(row.weight));
+    for (let i = 0; i < weight; i++) cycle.push(row.ip);
+  }
+  const slot = Math.floor(nowMs / WEIGHTED_SLOT_MS) % cycle.length;
+  return [cycle[slot]!];
+}

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

@@ -29,15 +29,91 @@ import * as domainService from "./domain-service.js";
 import { syncServiceToVpsTracker } from "./vps-tracker-sync.js";
 import { fireEnsureHealthWorker, DEFAULT_HEALTH_FALLBACKS } from "./health/health-worker-deploy.js";
 import {
+  canApplyLb,
   isHealthy,
+  isSharedPool,
+  resolveDesiredAIps,
   selectActiveIpsByMode,
+  shouldRecordFailoverDnsDiff,
   withBindingLock,
+  WEIGHTED_DNS_TTL,
   type LbIpRow,
   type LbTargetConfig,
 } from "./routing/index.js";
 
 export type { LbIpRow, LbTargetConfig };
-export { selectActiveIpsByMode };
+export {
+  canApplyLb,
+  resolveDesiredAIps,
+  selectActiveIpsByMode,
+  shouldRecordFailoverDnsDiff,
+};
+
+const AUTO_DNS_TTL = 1;
+
+function ttlForBinding(mode: LbMode, ips: readonly string[]): number {
+  return mode === "weighted" && isSharedPool(ips) ? WEIGHTED_DNS_TTL : AUTO_DNS_TTL;
+}
+
+function enabledServiceIps(db: Db, serviceId: number): string[] {
+  return repos
+    .listServiceIpRows(db, serviceId)
+    .filter((row) => row.enabled)
+    .map((row) => row.ip);
+}
+
+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);
+  const configuredIps = repos.listBindingIps(db, bindingId);
+  const { config, rows } = getBindingLbState(db, bindingId);
+  const downIps = new Set(
+    rows.filter((row) => row.health === "down").map((row) => row.ip),
+  );
+  if (
+    !shouldRecordFailoverDnsDiff({
+      configuredIps,
+      lbMode: config.lb_mode,
+      added,
+      removed,
+      downIps,
+    })
+  ) {
+    return;
+  }
+  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;
@@ -227,16 +303,27 @@ function getGroupLbState(
   };
 }
 
-function computeActiveIps(
+function desiredAIps(
   db: Db,
   scope: HealthCheckScope,
   refId: number,
+  fallbackIps: string[],
 ): string[] {
   const state =
     scope === "binding"
       ? getBindingLbState(db, refId)
       : getGroupLbState(db, refId);
-  return selectActiveIpsByMode(state.config, state.rows);
+  const serviceIps =
+    scope === "binding"
+      ? enabledServiceIps(db, repos.getBinding(db, refId).service_id)
+      : fallbackIps;
+  return resolveDesiredAIps(
+    state.config,
+    state.rows,
+    fallbackIps,
+    Date.now(),
+    serviceIps,
+  );
 }
 
 async function collectKnownZones(
@@ -285,6 +372,11 @@ 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
+      ? []
+      : resolveDesiredAIps(config, rows, targetIps, Date.now(), ips);
+
     return {
       binding_id: binding.id,
       domain_id: binding.domain_id,
@@ -312,13 +404,13 @@ 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);
     }
   }
@@ -404,6 +496,21 @@ function fallbackCnameHealth(
   );
 }
 
+function overlayLiveHealth(
+  stored: IpHealthState | undefined,
+  live: IpHealthState | undefined,
+): IpHealthState {
+  if (live && live !== "unknown") return live;
+  return stored ?? live ?? "unknown";
+}
+
+function bestAliveDisplayStatus(statuses: readonly string[]): IpHealthState {
+  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";
+}
+
 function attachServiceHealth(
   db: Db,
   views: ServiceView[],
@@ -411,10 +518,13 @@ function attachServiceHealth(
   const ids = views.map((v) => v.id);
   const healthByService = repos.aggregateIpHealthByServiceIds(db, ids);
   const ipHealthByService = repos.listIpHealthByServiceIds(db, ids);
+  const liveByService = repos.listLatestLiveHealthByServiceIds(db, ids);
   return views.map((view) => {
     const health = healthByService.get(view.id);
     const rows = ipHealthByService.get(view.id) ?? [];
+    const liveRows = liveByService.get(view.id) ?? [];
     const byIp = new Map(rows.map((row) => [row.ip, row]));
+    const liveByIp = new Map(liveRows.map((row) => [row.ip, row]));
     const cnameFallback = fallbackCnameHealth(rows, view);
     const aRecordIps = new Set(
       (view.domains ?? []).flatMap((domain) =>
@@ -423,20 +533,33 @@ function attachServiceHealth(
     );
     const ip_health = (view.ips ?? []).map((ip) => {
       const row = byIp.get(ip) ?? (aRecordIps.has(ip) ? undefined : cnameFallback);
+      const live = liveByIp.get(ip);
+      const status = overlayLiveHealth(row?.status, live?.status);
+      const extras = live && live.status !== "unknown" ? live : row;
       return {
         ip,
-        status: row?.status ?? ("unknown" as const),
-        latency_ms: row?.latency_ms ?? null,
-        last_checked_at: row?.last_checked_at ?? null,
-        last_error: row?.last_error ?? null,
-        provider: row?.provider ?? "local",
-        colo: row?.colo ?? null,
+        status,
+        latency_ms: extras?.latency_ms ?? null,
+        last_checked_at: extras?.last_checked_at ?? null,
+        last_error:
+          live && live.status !== "unknown"
+            ? live.last_error
+            : (row?.last_error ?? null),
+        provider: extras?.provider ?? "local",
+        colo: extras?.colo ?? null,
       };
     });
+    const displayStatus = bestAliveDisplayStatus(ip_health.map((row) => row.status));
+    const latencyRow =
+      ip_health.find((row) => row.status === displayStatus && row.latency_ms != null) ??
+      ip_health.find((row) => row.latency_ms != null);
     return {
       ...view,
-      health_status: health?.health_status ?? "unknown",
-      health_latency_ms: health?.health_latency_ms ?? null,
+      health_status: overlayLiveHealth(health?.health_status, displayStatus),
+      health_latency_ms:
+        displayStatus !== "unknown"
+          ? (latencyRow?.latency_ms ?? null)
+          : (health?.health_latency_ms ?? null),
       ip_health,
     };
   });
@@ -571,7 +694,17 @@ async function syncBindingDns(
     return;
   }
 
-  await syncBindingADns(db, cf, bindingId, domainId, hostname, desiredIps);
+  const binding = repos.getBinding(db, bindingId);
+  const configuredIps = repos.listBindingIps(db, bindingId);
+  await syncBindingADns(
+    db,
+    cf,
+    bindingId,
+    domainId,
+    hostname,
+    desiredIps,
+    ttlForBinding(binding.lb_mode, configuredIps),
+  );
 }
 
 async function syncBindingCnameDns(
@@ -658,6 +791,7 @@ async function syncBindingADns(
   domainId: number,
   hostname: string,
   desiredIps: string[],
+  ttl: number,
 ): Promise<void> {
   const domain = repos.getDomain(db, domainId);
   const zoneName = domain.zone_name;
@@ -677,6 +811,14 @@ async function syncBindingADns(
 
   if (desiredIps.length === 0) {
     repos.setBindingDnsRecordId(db, bindingId, null);
+    recordFailoverDnsDiff(
+      db,
+      bindingId,
+      hostname,
+      zoneName,
+      existingRecords,
+      desiredIps,
+    );
     return;
   }
 
@@ -685,13 +827,15 @@ async function syncBindingADns(
 
   for (const ip of desiredIps) {
     const existing = refreshed.find((r) => r.content === ip);
+    const recordName = dnsNameForBinding(hostname, zoneName);
     let recordId: number;
     if (existing) {
-      if (!dnsRecordNamesMatch(existing.name, hostname, zoneName)) {
+      if (!dnsRecordNamesMatch(existing.name, hostname, zoneName) || existing.ttl !== ttl) {
         await dnsService.update(db, cf, domainId, existing.id, {
           record_type: "A",
-          name: dnsNameForBinding(hostname, zoneName),
+          name: recordName,
           content: ip,
+          ttl,
           proxied: false,
         });
       }
@@ -709,12 +853,24 @@ async function syncBindingADns(
       if (adopted) {
         repos.linkBindingRecord(db, bindingId, adopted.id);
         recordId = adopted.id;
+        if (
+          !dnsRecordNamesMatch(adopted.name, hostname, zoneName) ||
+          adopted.ttl !== ttl
+        ) {
+          await dnsService.update(db, cf, domainId, adopted.id, {
+            record_type: "A",
+            name: recordName,
+            content: ip,
+            ttl,
+            proxied: false,
+          });
+        }
       } else {
         const record = await dnsService.create(db, cf, domainId, {
           record_type: "A",
-          name: dnsNameForBinding(hostname, zoneName),
+          name: recordName,
           content: ip,
-          ttl: 1,
+          ttl,
           proxied: false,
         });
         repos.linkBindingRecord(db, bindingId, record.id);
@@ -725,6 +881,14 @@ async function syncBindingADns(
   }
 
   repos.setBindingDnsRecordId(db, bindingId, primaryId);
+  recordFailoverDnsDiff(
+    db,
+    bindingId,
+    hostname,
+    zoneName,
+    existingRecords,
+    desiredIps,
+  );
 }
 
 async function cleanupBindingDns(
@@ -943,12 +1107,7 @@ async function syncServiceBindingsToDns(
     }
     validateTargetIpsInPool(targetIps, ips);
 
-    if (binding.health_check_enabled) {
-      const activeIps = computeActiveIps(db, "binding", binding.id);
-      if (activeIps.length > 0) {
-        targetIps = activeIps;
-      }
-    }
+    const desiredIps = desiredAIps(db, "binding", binding.id, targetIps);
 
     await syncBindingDns(
       db,
@@ -956,7 +1115,7 @@ async function syncServiceBindingsToDns(
       binding.id,
       binding.domain_id,
       binding.hostname,
-      targetIps,
+      desiredIps,
       null,
     );
   }
@@ -988,6 +1147,7 @@ async function syncGroupDomainDnsRecords(
   domainId: number,
   hostname: string,
   desiredIps: string[],
+  ttl: number = AUTO_DNS_TTL,
 ): Promise<void> {
   const domain = repos.getDomain(db, domainId);
   const zoneName = domain.zone_name;
@@ -1003,14 +1163,16 @@ async function syncGroupDomainDnsRecords(
   if (desiredIps.length === 0) return;
 
   const refreshed = repos.listGroupDnsRecords(db, groupId);
+  const recordName = dnsNameForBinding(hostname, zoneName);
   for (const ip of desiredIps) {
     const existing = refreshed.find((r) => r.content === ip);
     if (existing) {
-      if (!dnsRecordNamesMatch(existing.name, hostname, zoneName)) {
+      if (!dnsRecordNamesMatch(existing.name, hostname, zoneName) || existing.ttl !== ttl) {
         await dnsService.update(db, cf, domainId, existing.id, {
           record_type: "A",
-          name: dnsNameForBinding(hostname, zoneName),
+          name: recordName,
           content: ip,
+          ttl,
           proxied: false,
         });
       }
@@ -1026,13 +1188,25 @@ async function syncGroupDomainDnsRecords(
     );
     if (adopted) {
       repos.linkGroupDnsRecord(db, groupId, adopted.id);
+      if (
+        !dnsRecordNamesMatch(adopted.name, hostname, zoneName) ||
+        adopted.ttl !== ttl
+      ) {
+        await dnsService.update(db, cf, domainId, adopted.id, {
+          record_type: "A",
+          name: recordName,
+          content: ip,
+          ttl,
+          proxied: false,
+        });
+      }
       continue;
     }
     const record = await dnsService.create(db, cf, domainId, {
       record_type: "A",
-      name: dnsNameForBinding(hostname, zoneName),
+      name: recordName,
       content: ip,
-      ttl: 1,
+      ttl,
       proxied: false,
     });
     repos.linkGroupDnsRecord(db, groupId, record.id);
@@ -1083,9 +1257,8 @@ async function syncGroupDomainDns(
   const knownZones = await collectKnownZones(db, cf);
   const { zoneName, hostname } = parseFqdn(domainValue, knownZones);
   const domainId = await resolveDomainId(db, cf, zoneName);
-  const desiredIps = group.health_check_enabled
-    ? computeActiveIps(db, "group", groupId)
-    : await collectGroupDnsIps(db, groupId);
+  const fallbackIps = await collectGroupDnsIps(db, groupId);
+  const desiredIps = desiredAIps(db, "group", groupId, fallbackIps);
   await syncGroupDomainDnsRecords(
     db,
     cf,
@@ -1093,6 +1266,7 @@ async function syncGroupDomainDns(
     domainId,
     hostname,
     desiredIps,
+    ttlForBinding(group.lb_mode, fallbackIps),
   );
 }
 
@@ -1273,14 +1447,7 @@ export async function updateConfig(
       }
 
       if (pushDns) {
-        let effectiveIps = targetIps;
-        const refreshedBinding = repos.getBinding(db, binding.id);
-        if (refreshedBinding.health_check_enabled) {
-          const activeIps = computeActiveIps(db, "binding", binding.id);
-          if (activeIps.length > 0) {
-            effectiveIps = activeIps;
-          }
-        }
+        const effectiveIps = desiredAIps(db, "binding", binding.id, targetIps);
         await syncBindingDns(
           db,
           cf,
@@ -1586,16 +1753,18 @@ export async function reconcileDnsForTarget(
   if (scope === "binding") {
     await withBindingLock(refId, async () => {
       const binding = repos.getBinding(db, refId);
-      if (!binding.health_check_enabled) return;
+      if (!binding.health_check_enabled && binding.lb_mode !== "weighted") return;
       const service = repos.getService(db, binding.service_id);
       if (!shouldPushDns(db, service)) return;
       const cnameTarget = binding.cname_target?.trim() || null;
       if (cnameTarget) return;
       const ips = repos.listServiceIps(db, service.id);
+      const poolIps = enabledServiceIps(db, service.id);
       const targetIps = repos.listBindingIps(db, binding.id);
       validateTargetIpsInPool(targetIps, ips);
-      const activeIps = computeActiveIps(db, "binding", refId);
-      const desiredIps = activeIps.length > 0 ? activeIps : targetIps;
+      const desiredIps = canApplyLb(poolIps, targetIps)
+        ? desiredAIps(db, "binding", refId, targetIps)
+        : targetIps;
       await syncBindingDns(
         db,
         cf,
@@ -1610,8 +1779,61 @@ export async function reconcileDnsForTarget(
   }
 
   const group = repos.getServiceGroup(db, refId);
-  if (!group.enabled || !group.domain?.trim() || !group.health_check_enabled) {
+  if (!group.enabled || !group.domain?.trim()) {
+    return;
+  }
+  if (!group.health_check_enabled && group.lb_mode !== "weighted") {
     return;
   }
   await syncGroupDomainDns(db, cf, refId);
 }
+
+export async function reconcileWeightedDns(
+  db: Db,
+  cf: CloudflareClient,
+): Promise<number> {
+  let n = 0;
+  for (const binding of repos.listAllBindings(db)) {
+    if (binding.lb_mode !== "weighted") continue;
+    if (binding.cname_target?.trim()) continue;
+    if (!isSharedPool(binding.target_ips ?? [])) continue;
+    try {
+      await withBindingLock(binding.id, async () => {
+        const latest = repos.getBinding(db, binding.id);
+        if (latest.lb_mode !== "weighted") return;
+        if (latest.cname_target?.trim()) return;
+        const service = repos.getService(db, latest.service_id);
+        if (!shouldPushDns(db, service)) return;
+        const targetIps = repos.listBindingIps(db, latest.id);
+        const poolIps = enabledServiceIps(db, service.id);
+        if (!canApplyLb(poolIps, targetIps)) return;
+        const ips = repos.listServiceIps(db, service.id);
+        validateTargetIpsInPool(targetIps, ips);
+        const desiredIps = desiredAIps(db, "binding", latest.id, targetIps);
+        await syncBindingDns(
+          db,
+          cf,
+          latest.id,
+          latest.domain_id,
+          latest.hostname,
+          desiredIps,
+          null,
+        );
+        n += 1;
+      });
+    } catch {
+      continue;
+    }
+  }
+  for (const group of repos.listServiceGroups(db)) {
+    if (group.lb_mode !== "weighted") continue;
+    if (!group.enabled || !group.domain?.trim()) continue;
+    try {
+      await syncGroupDomainDns(db, cf, group.id);
+      n += 1;
+    } catch {
+      continue;
+    }
+  }
+  return n;
+}

apps/api/src/services/weighted-dns-scheduler.ts

@@ -0,0 +1,39 @@
+import type { FastifyInstance } from "fastify";
+import { AsyncTask, SimpleIntervalJob } from "toad-scheduler";
+import * as serviceConfigService from "./service-config-service.js";
+import { WEIGHTED_SLOT_MS } from "./routing/weighted.js";
+
+export const WEIGHTED_DNS_JOB_ID = "weighted-dns";
+
+export function createWeightedDnsTask(app: FastifyInstance): AsyncTask {
+  return new AsyncTask(
+    WEIGHTED_DNS_JOB_ID,
+    async () => {
+      const n = await serviceConfigService.reconcileWeightedDns(app.db, app.cf);
+      if (n > 0) {
+        app.log.info({ reconciled: n }, "weighted dns rotated");
+      }
+    },
+    (err) => {
+      app.log.warn({ err }, "weighted dns rotate failed");
+    },
+  );
+}
+
+export function scheduleWeightedDnsJob(
+  app: FastifyInstance,
+  task: AsyncTask,
+): void {
+  const scheduler = app.scheduler;
+  if (!scheduler) return;
+  if (scheduler.existsById(WEIGHTED_DNS_JOB_ID)) {
+    scheduler.removeById(WEIGHTED_DNS_JOB_ID);
+  }
+  scheduler.addSimpleIntervalJob(
+    new SimpleIntervalJob(
+      { seconds: WEIGHTED_SLOT_MS / 1000, runImmediately: true },
+      task,
+      { id: WEIGHTED_DNS_JOB_ID, preventOverrun: true },
+    ),
+  );
+}

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

@@ -291,6 +291,77 @@ describe("health-check state derivation via runAllChecks", () => {
     expect(targets[0]?.ip).toBe("2.59.161.102");
     expect(targets[0]?.hostname).toBe("s.rkns.top");
   });
+
+  it("does not mark node unhealthy when binding majority is OK and group local fails", async () => {
+    const { createMemoryDb, repos, runMigrations } = await import("@cfdm/db");
+    const { db, sqlite } = createMemoryDb();
+    runMigrations(sqlite);
+
+    const tcp = await startTcpServer();
+    try {
+      const domain = repos.createDomain(db, null, "example.com", "zone-id");
+      const group = repos.createServiceGroup(
+        db,
+        "VPN",
+        "vpn",
+        null,
+        "vpn.example.com",
+        {
+          health_check_enabled: true,
+          health_check_type: "http",
+          health_check_port: 1,
+          health_check_timeout_ms: 200,
+          health_check_path: "/",
+        },
+      );
+      const service = repos.createService(db, "Svc", "svc");
+      repos.setServiceGroup(db, service.id, group.id);
+      repos.setServiceEnabled(db, service.id, true);
+      const binding = repos.insertBinding(db, domain.id, service.id, "@", null);
+      repos.updateBindingLbConfig(db, binding.id, {
+        health_check_enabled: true,
+        health_check_type: "tcp",
+        health_check_port: tcp.port,
+        health_check_timeout_ms: 500,
+      });
+      repos.replaceBindingIpsWithMeta(db, binding.id, [
+        { ip: "127.0.0.1", weight: 1, priority: 1 },
+      ]);
+      const node = repos.findNodeByIp(db, "127.0.0.1");
+      expect(node).not.toBeNull();
+
+      await healthCheckService.runAllChecks(db, {
+        probeGapMs: 0,
+        thresholds: {
+          degradedFailures: 1,
+          downFailures: 1,
+          latencyWarnMs: 1000,
+        },
+      });
+
+      const bindingHealth = repos.getIpHealthStatusRow(
+        db,
+        "binding",
+        binding.id,
+        "127.0.0.1",
+      );
+      const groupHealth = repos.getIpHealthStatusRow(
+        db,
+        "group",
+        group.id,
+        "127.0.0.1",
+      );
+      const after = repos.getNode(db, node!.id);
+
+      expect(bindingHealth?.status).toBe("up");
+      expect(groupHealth?.status).toBe("down");
+      expect(after.health_status).toBe("healthy");
+      expect(after.consecutive_failures).toBe(0);
+      expect(after.last_failure_reason).toBeNull();
+    } finally {
+      await new Promise<void>((resolve) => tcp.server.close(() => resolve()));
+    }
+  });
 });
 
 describe("CNAME health mapped onto service IPs", () => {
@@ -326,4 +397,85 @@ 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");
+  });
+
+  it("getView shows live OK over hysteresis unknown", 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, "RW Panel", "rw-panel");
+    repos.replaceServiceIps(db, service.id, ["2.59.161.102"]);
+    const binding = repos.insertBinding(db, domain.id, service.id, "c", null);
+    repos.replaceBindingIpsWithMeta(db, binding.id, [
+      { ip: "2.59.161.102", weight: 1, priority: 1 },
+    ]);
+    repos.upsertIpHealthStatus(
+      db,
+      "binding",
+      binding.id,
+      "2.59.161.102",
+      "unknown",
+      63,
+      0,
+      null,
+      1,
+    );
+    repos.insertHealthProbeLog(db, {
+      scope: "binding",
+      refId: binding.id,
+      ip: "2.59.161.102",
+      provider: "local",
+      status: "up",
+      ok: true,
+      latencyMs: 63,
+      colo: null,
+      error: null,
+    });
+
+    const view = await getView(db, service.id);
+    expect(view.health_status).toBe("up");
+    expect(view.ip_health[0]?.status).toBe("up");
+    expect(view.ip_health[0]?.latency_ms).toBe(63);
+  });
 });

apps/api/test/lb-reconcile.test.ts

@@ -1,9 +1,12 @@
 import { describe, expect, it } from "vitest";
 import {
+  resolveDesiredAIps,
   selectActiveIpsByMode,
+  shouldRecordFailoverDnsDiff,
   type LbIpRow,
   type LbTargetConfig,
 } from "../src/services/service-config-service.js";
+import { WEIGHTED_SLOT_MS } from "../src/services/routing/weighted.js";
 
 function row(
   ip: string,
@@ -17,6 +20,11 @@ function row(
   };
 }
 
+const weightedConfig: LbTargetConfig = {
+  lb_mode: "weighted",
+  health_check_enabled: true,
+};
+
 describe("selectActiveIpsByMode", () => {
   it("round_robin returns all healthy ips, falls back to all if none healthy", () => {
     const config: LbTargetConfig = {
@@ -75,22 +83,49 @@ describe("selectActiveIpsByMode", () => {
     expect(selectActiveIpsByMode(config, rows)).toEqual(["2.2.2.2"]);
   });
 
-  it("weighted returns all healthy ips (one A per ip; weights stored for display)", () => {
-    const config: LbTargetConfig = {
-      lb_mode: "weighted",
-      health_check_enabled: true,
-    };
+  it("weighted 1:3 picks the lighter ip on slot 0 and the heavier on slot 1", () => {
     const rows = [
-      row("1.1.1.1", { weight: 3, health: "up" }),
-      row("2.2.2.2", { weight: 1, health: "up" }),
-      row("3.3.3.3", { weight: 2, health: "down" }),
+      row("1.1.1.1", { weight: 1, health: "up" }),
+      row("2.2.2.2", { weight: 3, health: "up" }),
     ];
-    expect(selectActiveIpsByMode(config, rows).sort()).toEqual([
-      "1.1.1.1",
+    expect(selectActiveIpsByMode(weightedConfig, rows, 0)).toEqual(["1.1.1.1"]);
+    expect(selectActiveIpsByMode(weightedConfig, rows, WEIGHTED_SLOT_MS)).toEqual([
       "2.2.2.2",
     ]);
   });
 
+  it("weighted excludes down ips from the cycle", () => {
+    const rows = [
+      row("1.1.1.1", { weight: 1, health: "up" }),
+      row("2.2.2.2", { weight: 3, health: "down" }),
+    ];
+    expect(selectActiveIpsByMode(weightedConfig, rows, 0)).toEqual(["1.1.1.1"]);
+    expect(
+      selectActiveIpsByMode(weightedConfig, rows, WEIGHTED_SLOT_MS),
+    ).toEqual(["1.1.1.1"]);
+  });
+
+  it("weighted with one ip always returns that ip", () => {
+    expect(
+      selectActiveIpsByMode(weightedConfig, [row("1.1.1.1", { weight: 5 })], 0),
+    ).toEqual(["1.1.1.1"]);
+  });
+
+  it("weighted with all unknown rotates across every ip", () => {
+    const rows = [
+      row("1.1.1.1", { weight: 1, health: "unknown" }),
+      row("2.2.2.2", { weight: 3, health: "unknown" }),
+    ];
+    expect(selectActiveIpsByMode(weightedConfig, rows, 0)).toEqual(["1.1.1.1"]);
+    expect(selectActiveIpsByMode(weightedConfig, rows, WEIGHTED_SLOT_MS)).toEqual([
+      "2.2.2.2",
+    ]);
+  });
+
+  it("weighted returns empty array for no rows", () => {
+    expect(selectActiveIpsByMode(weightedConfig, [], 0)).toEqual([]);
+  });
+
   it("round_robin excludes unknown when another ip is up", () => {
     const config: LbTargetConfig = {
       lb_mode: "round_robin",
@@ -111,3 +146,115 @@ describe("selectActiveIpsByMode", () => {
     expect(selectActiveIpsByMode(config, [])).toEqual([]);
   });
 });
+
+describe("resolveDesiredAIps", () => {
+  it("keeps a dedicated single IP even when down", () => {
+    expect(
+      resolveDesiredAIps(
+        weightedConfig,
+        [row("1.1.1.1", { health: "down" })],
+        ["1.1.1.1"],
+      ),
+    ).toEqual(["1.1.1.1"]);
+  });
+
+  it("does not drain when the service has only one unique IP", () => {
+    expect(
+      resolveDesiredAIps(
+        weightedConfig,
+        [row("1.1.1.1", { health: "down" })],
+        ["1.1.1.1", "1.1.1.1"],
+        0,
+        ["1.1.1.1"],
+      ),
+    ).toEqual(["1.1.1.1", "1.1.1.1"]);
+  });
+
+  it("does not apply overlay when service pool is a single IP", () => {
+    const rows = [
+      row("1.1.1.1", { health: "up" }),
+      row("2.2.2.2", { health: "down" }),
+    ];
+    expect(
+      resolveDesiredAIps(
+        weightedConfig,
+        rows,
+        ["1.1.1.1", "2.2.2.2"],
+        0,
+        ["1.1.1.1"],
+      ),
+    ).toEqual(["1.1.1.1", "2.2.2.2"]);
+  });
+
+  it("applies weighted overlay on a shared pool", () => {
+    const rows = [
+      row("1.1.1.1", { weight: 1, health: "up" }),
+      row("2.2.2.2", { weight: 3, health: "up" }),
+    ];
+    expect(
+      resolveDesiredAIps(weightedConfig, rows, ["1.1.1.1", "2.2.2.2"], 0),
+    ).toEqual(selectActiveIpsByMode(weightedConfig, rows, 0));
+  });
+});
+
+describe("shouldRecordFailoverDnsDiff", () => {
+  it("skips duplicate listings of the same IP", () => {
+    expect(
+      shouldRecordFailoverDnsDiff({
+        configuredIps: ["1.1.1.1", "1.1.1.1"],
+        lbMode: "failover",
+        added: [],
+        removed: ["1.1.1.1"],
+        downIps: new Set(["1.1.1.1"]),
+      }),
+    ).toBe(false);
+  });
+
+  it("skips dedicated extra-FQDN", () => {
+    expect(
+      shouldRecordFailoverDnsDiff({
+        configuredIps: ["1.1.1.1"],
+        lbMode: "failover",
+        added: [],
+        removed: ["1.1.1.1"],
+        downIps: new Set(["1.1.1.1"]),
+      }),
+    ).toBe(false);
+  });
+
+  it("skips weighted live-to-live slot swap", () => {
+    expect(
+      shouldRecordFailoverDnsDiff({
+        configuredIps: ["1.1.1.1", "2.2.2.2"],
+        lbMode: "weighted",
+        added: ["2.2.2.2"],
+        removed: ["1.1.1.1"],
+        downIps: new Set(),
+      }),
+    ).toBe(false);
+  });
+
+  it("logs weighted swap when a down ip leaves the pool", () => {
+    expect(
+      shouldRecordFailoverDnsDiff({
+        configuredIps: ["1.1.1.1", "2.2.2.2"],
+        lbMode: "weighted",
+        added: ["2.2.2.2"],
+        removed: ["1.1.1.1"],
+        downIps: new Set(["1.1.1.1"]),
+      }),
+    ).toBe(true);
+  });
+
+  it("logs failover diffs on a shared pool", () => {
+    expect(
+      shouldRecordFailoverDnsDiff({
+        configuredIps: ["1.1.1.1", "2.2.2.2"],
+        lbMode: "failover",
+        added: ["2.2.2.2"],
+        removed: ["1.1.1.1"],
+        downIps: new Set(["1.1.1.1"]),
+      }),
+    ).toBe(true);
+  });
+});

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

@@ -1,40 +1,161 @@
+import { ShieldCheckIcon } from 'lucide-react'
+
 import {
   Timeline,
   TimelineContent,
+  TimelineDate,
   TimelineHeader,
   TimelineIndicator,
   TimelineItem,
   TimelineSeparator,
   TimelineTitle,
 } from '@/components/reui/timeline'
+import { HealthCheckBadge } from '@/components/health-check-badge'
+import { EmptyState } from '@/components/empty-state'
+import { formatDate, formatRelative, sqliteUtcToIso } from '@/lib/format'
+import {
+  failoverEventCopy,
+  type FailoverEvent,
+  type FailoverHistoryItem,
+} from '@/lib/failover-events'
+import { cn } from '@cfdm/ui/lib/utils'
+import type { ComponentProps } from 'react'
 
-export interface FailoverEvent {
-  id: string
-  title: string
-  detail: string
+type HealthBadgeStatus = ComponentProps<typeof HealthCheckBadge>['status']
+
+function failStreakLabel(count: number): string {
+  const mod10 = count % 10
+  const mod100 = count % 100
+  if (mod10 === 1 && mod100 !== 11) return `${count} ошибка подряд`
+  if (mod10 >= 2 && mod10 <= 4 && (mod100 < 12 || mod100 > 14)) {
+    return `${count} ошибки подряд`
+  }
+  return `${count} ошибок подряд`
 }
 
-export function FailoverTimeline({ events }: { events: FailoverEvent[] }) {
-  if (events.length === 0) {
+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(tone: 'down' | 'added' | 'removed'): string {
+  if (tone === 'added') return 'bg-success/25'
+  return 'bg-destructive/25'
+}
+
+/**
+ * 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,
+  history = [],
+}: {
+  events: FailoverEvent[]
+  history?: readonly FailoverHistoryItem[]
+}) {
+  if (events.length === 0 && history.length === 0) {
     return (
-      <p className="text-muted-foreground text-sm">
-        Событий failover пока нет.
-      </p>
+      <EmptyState
+        icon={ShieldCheckIcon}
+        title="Нет инцидентов"
+        description="Нет Down и нет выходов из общего пула"
+        stackedIcon={false}
+        centered={false}
+      />
     )
   }
 
   return (
-    <Timeline defaultValue={events.length} className="w-full">
-      {events.map((event, index) => (
-        <TimelineItem key={event.id} step={index + 1}>
-          <TimelineSeparator />
-          <TimelineIndicator />
-          <TimelineHeader>
-            <TimelineTitle>{event.title}</TimelineTitle>
-          </TimelineHeader>
-          <TimelineContent>{event.detail}</TimelineContent>
-        </TimelineItem>
-      ))}
-    </Timeline>
+    <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('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/health-check-config-fields.tsx

@@ -1,6 +1,7 @@
 import { FormFieldSimple } from '@/components/form-field'
 import { AppInput } from '@/components/app-input'
 import { SettingRow } from '@/components/setting-row'
+import { SelectField } from '@/components/select-field'
 import { Badge } from '@/components/reui/badge'
 import {
   NumberField,
@@ -9,15 +10,8 @@ import {
   NumberFieldIncrement,
   NumberFieldInput,
 } from '@/components/reui/number-field'
-import {
-  Select,
-  SelectContent,
-  SelectItem,
-  SelectTrigger,
-  SelectValue,
-} from '@cfdm/ui/components/select'
 import { Switch } from '@cfdm/ui/components/switch'
-import { FieldGroup } from '@cfdm/ui/components/field'
+import { Field, FieldGroup, FieldLabel } from '@cfdm/ui/components/field'
 import { Button } from '@cfdm/ui/components/button'
 import { ButtonGroup } from '@cfdm/ui/components/button-group'
 import { CableIcon, GlobeIcon } from 'lucide-react'
@@ -59,9 +53,14 @@ export interface LbAndHealthConfig extends HealthCheckConfig {
 const defaultLbModeOptions = [
   { value: 'round_robin', label: 'Round Robin' },
   { value: 'failover', label: 'Failover (приоритет)' },
-  { value: 'weighted', label: 'Weighted (веса)' },
+  { value: 'weighted', label: 'Веса (подмена IP)' },
 ]
 
+export interface LbPoolMetaChange {
+  weight?: number
+  priority?: number
+}
+
 function CompactNumberField({
   id,
   value,
@@ -95,6 +94,89 @@ function CompactNumberField({
   )
 }
 
+function PoolLbMetaFields({
+  idPrefix,
+  mode,
+  ips,
+  weights,
+  priorities,
+  onMetaChange,
+}: {
+  idPrefix: string
+  mode: Exclude<LbMode, 'round_robin'>
+  ips: readonly string[]
+  weights: Record<string, number>
+  priorities: Record<string, number>
+  onMetaChange?: (ip: string, meta: LbPoolMetaChange) => void
+}) {
+  const isWeighted = mode === 'weighted'
+  const minPriority =
+    ips.length === 0
+      ? 1
+      : Math.min(...ips.map((ip) => priorities[ip] ?? 1))
+
+  return (
+    <SettingRow
+      title={isWeighted ? 'Вес IP' : 'Приоритет IP'}
+      description={
+        isWeighted
+          ? 'Доля времени на общем FQDN: 1 и 3 = ¼ и ¾ цикла (слот 60 с)'
+          : '1 — основной, больше — запасной'
+      }
+      compact
+      stacked
+      className="gap-3 px-0 py-3"
+      contentClassName="min-w-0"
+    >
+      {ips.length === 0 ? (
+        <p className="text-muted-foreground text-sm">Сначала добавьте IP выше</p>
+      ) : (
+        <div className="flex w-full flex-col gap-2">
+          {ips.map((ip) => {
+            const isPrimary = (priorities[ip] ?? 1) === minPriority
+            const fieldId = isWeighted
+              ? `${idPrefix}-weight-${ip}`
+              : `${idPrefix}-priority-${ip}`
+            return (
+              <div key={ip} className="flex items-center gap-3">
+                <span className="min-w-0 flex-1 truncate font-mono text-sm">{ip}</span>
+                {!isWeighted ? (
+                  <Badge
+                    variant={isPrimary ? 'success-light' : 'outline'}
+                    size="xs"
+                    radius="full"
+                  >
+                    {isPrimary ? 'Основной' : 'Запасной'}
+                  </Badge>
+                ) : null}
+                <Field className="w-28 gap-0">
+                  <FieldLabel htmlFor={fieldId} className="sr-only">
+                    {isWeighted ? `Вес ${ip}` : `Приоритет ${ip}`}
+                  </FieldLabel>
+                  <CompactNumberField
+                    id={fieldId}
+                    value={isWeighted ? (weights[ip] ?? 1) : (priorities[ip] ?? 1)}
+                    min={1}
+                    max={100}
+                    onValueChange={(next) =>
+                      onMetaChange?.(
+                        ip,
+                        isWeighted
+                          ? { weight: next ?? 1 }
+                          : { priority: next ?? 1 },
+                      )
+                    }
+                  />
+                </Field>
+              </div>
+            )
+          })}
+        </div>
+      )}
+    </SettingRow>
+  )
+}
+
 export function HealthCheckConfigFields({
   value,
   onChange,
@@ -102,6 +184,10 @@ export function HealthCheckConfigFields({
   lbModeOptions = defaultLbModeOptions,
   idPrefix = 'health',
   showLbMode = true,
+  ips = [],
+  weights = {},
+  priorities = {},
+  onMetaChange,
   className,
 }: {
   value: LbAndHealthConfig
@@ -110,6 +196,10 @@ export function HealthCheckConfigFields({
   lbModeOptions?: { value: string; label: string }[]
   idPrefix?: string
   showLbMode?: boolean
+  ips?: readonly string[]
+  weights?: Record<string, number>
+  priorities?: Record<string, number>
+  onMetaChange?: (ip: string, meta: LbPoolMetaChange) => void
   className?: string
 }) {
   function patch(next: Partial<LbAndHealthConfig>) {
@@ -134,25 +224,28 @@ export function HealthCheckConfigFields({
           compact
           className={rowClass}
         >
-          <Select
+          <SelectField
             modal={false}
             value={value.lb_mode}
             onValueChange={(v) => patch({ lb_mode: (v ?? 'round_robin') as LbMode })}
-          >
-            <SelectTrigger id={`${idPrefix}-lb-mode`} className="w-full">
-              <SelectValue placeholder="Выберите режим" />
-            </SelectTrigger>
-            <SelectContent>
-              {lbModeOptions.map((item) => (
-                <SelectItem key={item.value} value={item.value}>
-                  {item.label}
-                </SelectItem>
-              ))}
-            </SelectContent>
-          </Select>
+            triggerId={`${idPrefix}-lb-mode`}
+            placeholder="Выберите режим"
+            options={lbModeOptions}
+          />
         </SettingRow>
       ) : null}
 
+      {showLbMode && value.lb_mode !== 'round_robin' ? (
+        <PoolLbMetaFields
+          idPrefix={idPrefix}
+          mode={value.lb_mode}
+          ips={ips}
+          weights={weights}
+          priorities={priorities}
+          onMetaChange={onMetaChange}
+        />
+      ) : null}
+
       <SettingRow
         title="Провайдер health-check"
         description="Кто пробирует цель. Можно выбрать несколько источников."

apps/web/src/components/reui-kit/index.ts

@@ -1,5 +1,6 @@
 export { UptimeChart, type UptimeProbe, type UptimePeriodKey, probeUptimePercent, lastProbeLatency } from './uptime-chart'
 export { ServiceHealthMonitor } from './service-health-monitor'
+export { ServiceFailoverPanel } from './service-failover-panel'
 export { applyFiltersToData, getActiveFilters, renderSingleSelectedLabel } from './filter-utils'
 export { CertStatusChart, GroupDomainsChart } from './dashboard-analytics'
 export { ResourcePage, type ResourcePageProps, type ResourcePageTab } from './resource-page'

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

@@ -0,0 +1,141 @@
+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, ServiceView } from '@/lib/schemas'
+import { Badge } from '@/components/reui/badge'
+import {
+  Frame,
+  FrameDescription,
+  FrameHeader,
+  FramePanel,
+  FrameTitle,
+} from '@/components/reui/frame'
+import {
+  Alert,
+  AlertDescription,
+  AlertTitle,
+} from '@/components/reui/alert'
+
+type LbMode = ServiceView['lb_mode']
+
+const PANEL_COPY: Record<LbMode, { title: string; description: string }> = {
+  failover: {
+    title: 'Failover (приоритет)',
+    description: 'Down снимается с общего FQDN. История: кто вышел из пула и кто вернулся',
+  },
+  weighted: {
+    title: 'Веса (подмена IP)',
+    description: 'На общем FQDN один IP по весам. Down выводится из цикла',
+  },
+  round_robin: {
+    title: 'Round Robin',
+    description: 'На общем FQDN все живые A. Down снимается с пула',
+  },
+}
+
+function failoverCountLabel(count: number): string {
+  const mod10 = count % 10
+  const mod100 = count % 100
+  if (mod10 === 1 && mod100 !== 11) return `${count} адрес Down`
+  if (mod10 >= 2 && mod10 <= 4 && (mod100 < 12 || mod100 > 14)) {
+    return `${count} адреса Down`
+  }
+  return `${count} адресов Down`
+}
+
+function alertDescription(events: { kind: string; fqdns: string[] }[]): string {
+  const removedCount = events.filter((event) => event.kind === 'removed').length
+  if (removedCount > 0) return 'Сняты с общего FQDN'
+  if (events.some((event) => event.fqdns.length > 0)) {
+    return 'Остались в A-записях общего FQDN как last-resort'
+  }
+  return 'Down: персональные FQDN не меняются'
+}
+
+/**
+ * Балансировка — текущие Down + кто вышел из общего пула.
+ * 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
+ * Docs: https://reui.io/docs/components/base/timeline
+ * Docs: https://reui.io/docs/components/base/badge
+ * Docs: https://reui.io/docs/components/base/alert
+ */
+export function ServiceFailoverPanel({
+  lbMode = 'round_robin',
+  ipHealth,
+  bindings,
+  history,
+  probes = [],
+}: {
+  lbMode?: LbMode
+  ipHealth: readonly FailoverHealthInput[]
+  bindings: readonly FailoverBindingPool[]
+  history: readonly FailoverLogEntry[]
+  probes?: readonly HealthLogProbe[]
+}) {
+  const copy = PANEL_COPY[lbMode] ?? PANEL_COPY.round_robin
+  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)
+
+  return (
+    <Frame stacked spacing="sm" className="min-w-0 w-full">
+      <FramePanel className="flex flex-col gap-3">
+        <FrameHeader className="gap-1 px-0 py-0">
+          <FrameTitle className="flex flex-wrap items-center gap-2">
+            {copy.title}
+            {events.length > 0 ? (
+              <Badge variant="destructive-light" size="xs" radius="full">
+                {events.length}
+              </Badge>
+            ) : (
+              <Badge variant="success-light" size="xs" radius="full">
+                OK
+              </Badge>
+            )}
+          </FrameTitle>
+          <FrameDescription>{copy.description}</FrameDescription>
+        </FrameHeader>
+
+        {events.length > 0 ? (
+          <Alert variant="destructive">
+            <UnplugIcon aria-hidden="true" />
+            <AlertTitle>{failoverCountLabel(events.length)}</AlertTitle>
+            <AlertDescription>{alertDescription(events)}</AlertDescription>
+          </Alert>
+        ) : null}
+
+        <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/service-edit-sheet.tsx

@@ -17,6 +17,7 @@ import {
   DEFAULT_BINDING_HEALTH,
   emptyAddressBlock,
   hydrateAddressBlock,
+  patchAddressIpMeta,
   toBindingDrafts,
   toDomainsPayload,
   type AddressBlockState,
@@ -282,6 +283,12 @@ export function ServiceEditSheet({
               idPrefix="service-health"
               value={primaryHealthValue}
               onChange={handlePrimaryHealthChange}
+              ips={address.nodes.map((node) => node.ip)}
+              weights={address.target_ip_weights}
+              priorities={address.target_ip_priorities}
+              onMetaChange={(ip, meta) =>
+                setAddress((current) => patchAddressIpMeta(current, ip, meta))
+              }
             />
           </section>
         </div>

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/components/services/service-unit-card.tsx

@@ -71,7 +71,7 @@ const LB_MODE_META: Record<
   weighted: {
     icon: ScaleIcon,
     className: 'text-info',
-    label: 'Weighted (веса)',
+    label: 'Веса (подмена IP)',
   },
 }
 

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

@@ -0,0 +1,336 @@
+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'>,
+): FailoverHealthInput {
+  return {
+    status: 'up',
+    consecutive_failures: 0,
+    last_error: null,
+    last_checked_at: null,
+    ...overrides,
+  }
+}
+
+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)
+    expect(isFailoverEventStatus('up')).toBe(false)
+    expect(isFailoverEventStatus('degraded')).toBe(false)
+    expect(isFailoverEventStatus('unknown')).toBe(false)
+    expect(isFailoverEventStatus('unhealthy')).toBe(false)
+  })
+
+  it('один IP у сервиса — не инцидент пула', () => {
+    const events = toFailoverEvents(
+      [
+        row({
+          ip: '10.0.0.1',
+          status: 'down',
+          consecutive_failures: 9,
+          last_error: 'fetch failed',
+        }),
+      ],
+      [
+        {
+          fqdn: 'solo.example.com',
+          configured: ['10.0.0.1'],
+          active: [],
+        },
+      ],
+    )
+    expect(events).toEqual([])
+  })
+
+  it('Down без shared pool не событие балансировки', () => {
+    const events = toFailoverEvents([
+      row({
+        ip: '130.49.213.153',
+        status: 'down',
+        consecutive_failures: 9,
+        last_error: 'fetch failed',
+      }),
+    ])
+    expect(events).toEqual([])
+  })
+
+  it('OK вне пула не инцидент (standby)', () => {
+    const events = toFailoverEvents(
+      [
+        row({ ip: '10.0.0.1', status: 'up' }),
+        row({ ip: '130.49.213.153', status: 'up' }),
+      ],
+      [
+        {
+          fqdn: 'pool.example.com',
+          configured: ['10.0.0.1', '130.49.213.153'],
+          active: ['10.0.0.1'],
+        },
+      ],
+    )
+    expect(events).toEqual([])
+  })
+
+  it('unknown и degraded вне пула не инцидент', () => {
+    const events = toFailoverEvents(
+      [
+        row({ ip: '10.0.0.1', status: 'up' }),
+        row({ ip: '10.0.0.2', status: 'unknown' }),
+        row({ ip: '10.0.0.3', status: 'degraded' }),
+      ],
+      [
+        {
+          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 на доп. FQDN last-resort, снятая с общего пула', () => {
+    const events = toFailoverEvents(
+      [
+        row({
+          ip: '130.49.213.153',
+          status: 'down',
+          consecutive_failures: 9,
+          last_error: 'fetch failed',
+          last_checked_at: '2026-08-20 07:00:00',
+        }),
+        row({ ip: '93.115.203.183', status: 'up' }),
+      ],
+      mskHip,
+    )
+    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')
+    expect(events[0]?.fqdns).not.toContain('nsgt.rkns.top')
+  })
+
+  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).toEqual([])
+  })
+
+  it('down без A-записи на configured FQDN — снятие', () => {
+    const events = toFailoverEvents(
+      [
+        row({
+          ip: '130.49.213.153',
+          status: 'down',
+          consecutive_failures: 9,
+          last_error: 'fetch failed',
+          last_checked_at: '2026-08-20 07:00:00',
+        }),
+      ],
+      [
+        {
+          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',
+      },
+    ])
+  })
+})
+
+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).toBe(
+      '130.49.213.153 вышла из пула (gt.rkns.top)',
+    )
+  })
+
+  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')
+  })
+
+  it('drops probe leave/return when the service has no shared pool', () => {
+    const history = mergeFailoverHistory(
+      [],
+      [
+        probe({ id: 1, status: 'up', checked_at: '2026-08-20T09:00:00Z' }),
+        probe({ id: 2, status: 'down', checked_at: '2026-08-20T09:10:00Z' }),
+      ],
+      [
+        {
+          fqdn: 'solo.example.com',
+          configured: ['130.49.213.153'],
+          active: ['130.49.213.153'],
+        },
+      ],
+    )
+    expect(history).toEqual([])
+  })
+
+  it('drops dedicated extra-FQDN DNS rows', () => {
+    const history = mergeFailoverHistory(
+      [
+        dns({
+          id: 11,
+          fqdn: 'nsgt.rkns.top',
+          action: 'added',
+          created_at: '2026-08-20 09:26:00',
+        }),
+      ],
+      [],
+      mskHip,
+    )
+    expect(history).toEqual([])
+  })
+})

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

@@ -0,0 +1,262 @@
+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
+}
+
+/** Binding IP health — тот же контур, что таблица активов. */
+export interface FailoverHealthInput {
+  ip: string
+  status: string
+  consecutive_failures?: number
+  last_error?: string | null
+  last_checked_at?: string | null
+}
+
+export interface FailoverBindingPool {
+  fqdn: string
+  configured: readonly string[]
+  active: readonly string[]
+}
+
+/** Unique A targets. Duplicates of the same IP are not a pool. */
+export function uniqueIpCount(ips: readonly string[]): number {
+  return new Set(ips.filter(Boolean)).size
+}
+
+/** Shared pool FQDN — two or more unique A targets. */
+export function isSharedPoolBinding(binding: FailoverBindingPool): boolean {
+  return uniqueIpCount(binding.configured) >= 2
+}
+
+export function hasSharedPool(
+  bindings: readonly FailoverBindingPool[],
+): boolean {
+  return bindings.some(isSharedPoolBinding)
+}
+
+/** 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'
+}
+
+/**
+ * Инциденты пула только если у сервиса есть shared FQDN (2+ уникальных IP).
+ * Один IP — не балансировка и не вывод из пула; Down смотрит health-монитор.
+ */
+export function toFailoverEvents(
+  ipHealth: readonly FailoverHealthInput[],
+  bindings: readonly FailoverBindingPool[] = [],
+): FailoverEvent[] {
+  if (!hasSharedPool(bindings)) return []
+  return ipHealth.filter((row) => isFailoverEventStatus(row.status)).map((row) => {
+    const removedFqdns: string[] = []
+    const lastResortFqdns: string[] = []
+    for (const binding of bindings) {
+      if (!isSharedPoolBinding(binding)) continue
+      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) => isSharedPoolBinding(binding) && binding.configured.includes(ip))
+    .map((binding) => binding.fqdn)
+}
+
+function isPoolFqdn(
+  fqdn: string,
+  bindings: readonly FailoverBindingPool[],
+): boolean {
+  const binding = bindings.find((item) => item.fqdn === fqdn)
+  if (!binding) return true
+  return isSharedPoolBinding(binding)
+}
+
+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
+      .filter((item) => isPoolFqdn(item.fqdn, bindings))
+      .map(dnsHistoryItem),
+    ...toIpAliveTransitions(probes)
+      .filter((transition) => fqdnsForIp(transition.ip, bindings).length > 0)
+      .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 +1,13 @@
 import { describe, expect, it } from 'vitest'
 
 import {
+  bestAliveHealthStatus,
   collapseStatusChanges,
   enabledHealthProviders,
+  latestHealthByIp,
   providerHealthStatuses,
+  resolveIpDisplayHealth,
+  resolveServiceDisplayHealth,
   worstHealthStatus,
   type HealthLogProbe,
 } from '@/lib/health-log'
@@ -73,7 +77,7 @@ 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 +88,7 @@ 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 +99,70 @@ 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')
+  })
+})
+
+describe('resolveServiceDisplayHealth', () => {
+  it('shows OK when live probes recovered and stored is still unknown', () => {
+    expect(
+      resolveServiceDisplayHealth(
+        'unknown',
+        [{ ip: '2.59.161.102', status: 'unknown' }],
+        [
+          probe({
+            id: 1,
+            ip: '2.59.161.102',
+            status: 'up',
+            checked_at: '2026-08-20T18:00:00Z',
+          }),
+        ],
+      ),
+    ).toBe('up')
+  })
+})

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 +131,85 @@ 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'
+}
+
+/** Service KPI / card: any-up of per-IP overlay (live probes beat hysteresis). */
+export function resolveServiceDisplayHealth(
+  stored: HealthLogStatus | undefined,
+  ipHealth: readonly { ip: string; status: string }[],
+  probes: readonly HealthLogProbe[] = [],
+): HealthLogStatus {
+  const liveByIp = latestHealthByIp(probes)
+  const statuses =
+    ipHealth.length > 0
+      ? ipHealth.map((row) =>
+          resolveIpDisplayHealth(
+            row.status as HealthLogStatus,
+            liveByIp.get(row.ip)?.status,
+          ),
+        )
+      : [...liveByIp.values()].map((row) => row.status)
+  return resolveIpDisplayHealth(stored, bestAliveHealthStatus(statuses))
+}

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/lib/service-address.test.ts

@@ -7,6 +7,7 @@ import {
   emptyAddressBlock,
   emptyBindingDraft,
   hydrateAddressBlock,
+  patchAddressIpMeta,
   removeAddressNode,
   toAddressBindings,
   toDomainsPayload,
@@ -82,6 +83,26 @@ describe('hydrateAddressBlock', () => {
     expect(state.preservedBindings).toHaveLength(1)
     expect(state.preservedBindings[0]?.fqdn).toBe('edge.example.com')
   })
+
+  it('поднимает веса и приоритеты с общего FQDN', () => {
+    const drafts = [
+      aRecord('gt.rkns.top', ['130.49.213.153', '93.115.203.183'], {
+        target_ip_weights: { '130.49.213.153': 3, '93.115.203.183': 1 },
+        target_ip_priorities: { '130.49.213.153': 2, '93.115.203.183': 1 },
+      }),
+      aRecord('nsgt.rkns.top', ['130.49.213.153']),
+    ]
+    const state = hydrateAddressBlock(drafts, ['130.49.213.153', '93.115.203.183'])
+    expect(state.target_ip_weights).toEqual({
+      '130.49.213.153': 3,
+      '93.115.203.183': 1,
+    })
+    expect(state.target_ip_priorities).toEqual({
+      '130.49.213.153': 2,
+      '93.115.203.183': 1,
+    })
+    expect(state.nodes[0]?.extraFqdn).toBe('nsgt.rkns.top')
+  })
 })
 
 describe('toDomainsPayload', () => {
@@ -161,6 +182,35 @@ describe('addAddressNode / addCommonFqdn', () => {
   })
 })
 
+describe('patchAddressIpMeta', () => {
+  it('меняет вес одного IP и не трогает extraFqdn', () => {
+    const state = {
+      ...addAddressNode(addAddressNode(emptyAddressBlock(), '1.1.1.1'), '2.2.2.2'),
+      nodes: [
+        { ip: '1.1.1.1', extraFqdn: 'msk.example.com' },
+        { ip: '2.2.2.2', extraFqdn: '' },
+      ],
+    }
+    const next = patchAddressIpMeta(state, '1.1.1.1', { weight: 7 })
+    expect(next.target_ip_weights).toEqual({ '1.1.1.1': 7, '2.2.2.2': 1 })
+    expect(next.target_ip_priorities).toEqual(state.target_ip_priorities)
+    expect(next.nodes).toEqual(state.nodes)
+  })
+
+  it('clamp веса и приоритета в 1–100', () => {
+    const state = addAddressNode(emptyAddressBlock(), '10.0.0.1')
+    expect(patchAddressIpMeta(state, '10.0.0.1', { weight: 0 }).target_ip_weights['10.0.0.1']).toBe(1)
+    expect(
+      patchAddressIpMeta(state, '10.0.0.1', { priority: 999 }).target_ip_priorities['10.0.0.1'],
+    ).toBe(100)
+  })
+
+  it('игнорирует IP вне пула', () => {
+    const state = addAddressNode(emptyAddressBlock(), '10.0.0.1')
+    expect(patchAddressIpMeta(state, '8.8.8.8', { weight: 5 })).toBe(state)
+  })
+})
+
 describe('CNAME / preservedBindings', () => {
   it('сохраняет CNAME в preserved при круге hydrate → payload', () => {
     const cname: ServiceBindingDraft = {

apps/web/src/lib/service-address.ts

@@ -243,6 +243,33 @@ export function addAddressNode(state: AddressBlockState, ip: string): AddressBlo
   }
 }
 
+const LB_META_MIN = 1
+const LB_META_MAX = 100
+
+function clampLbMeta(value: number): number {
+  if (!Number.isFinite(value)) return LB_META_MIN
+  return Math.min(LB_META_MAX, Math.max(LB_META_MIN, Math.round(value)))
+}
+
+export function patchAddressIpMeta(
+  state: AddressBlockState,
+  ip: string,
+  meta: { weight?: number; priority?: number },
+): AddressBlockState {
+  if (!state.nodes.some((node) => node.ip === ip)) return state
+  return {
+    ...state,
+    target_ip_weights:
+      meta.weight === undefined
+        ? state.target_ip_weights
+        : { ...state.target_ip_weights, [ip]: clampLbMeta(meta.weight) },
+    target_ip_priorities:
+      meta.priority === undefined
+        ? state.target_ip_priorities
+        : { ...state.target_ip_priorities, [ip]: clampLbMeta(meta.priority) },
+  }
+}
+
 function fqdnKey(value: string): string {
   return value.trim().toLowerCase()
 }

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,
@@ -35,6 +36,8 @@ export const serviceGroupsQueryOptions = () =>
       }
       return parsed.data
     },
+    refetchInterval: 10_000,
+    staleTime: 5_000,
   })
 
 export const servicesQueryOptions = () =>
@@ -89,6 +92,7 @@ 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,
 }
 
@@ -99,6 +103,8 @@ export const serviceViewQueryOptions = (id: number) =>
       const data = await api.get<unknown>(`/api/v1/services/${id}`)
       return serviceViewSchema.parse(data)
     },
+    refetchInterval: 10_000,
+    staleTime: 5_000,
   })
 
 export const serviceHealthLogQueryOptions = (id: number) =>
@@ -108,6 +114,17 @@ export const serviceHealthLogQueryOptions = (id: number) =>
       const data = await api.get<unknown>(`/api/v1/services/${id}/health-log`)
       return z.object({ items: z.array(healthProbeLogSchema) }).parse(data)
     },
+    refetchInterval: 10_000,
+    staleTime: 5_000,
+  })
+
+export const serviceFailoverLogQueryOptions = (id: number) =>
+  queryOptions({
+    queryKey: serviceDetailKeys.failoverLog(id),
+    queryFn: async () => {
+      const data = await api.get<unknown>(`/api/v1/services/${id}/failover-log`)
+      return z.object({ items: z.array(failoverLogSchema) }).parse(data)
+    },
   })
 
 export const serviceOverviewQueryOptions = (id: number) =>

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

@@ -14,7 +14,6 @@ import {
 import { ChangeDomainSheet } from '@/components/change-domain-sheet'
 import { ChangeIpSheet } from '@/components/change-ip-sheet'
 import { EmptyState } from '@/components/empty-state'
-import { FailoverTimeline } from '@/components/failover-timeline'
 import { FormFieldSimple } from '@/components/form-field'
 import { FormSheet } from '@/components/form-sheet'
 import { HealthCheckBadge } from '@/components/health-check-badge'
@@ -29,19 +28,15 @@ import {
 import { LbModeTile } from '@/components/services/service-unit-card'
 import {
   KpiStatGrid,
+  ServiceFailoverPanel,
   ServiceHealthMonitor,
 } from '@/components/reui-kit'
-import {
-  Frame,
-  FrameDescription,
-  FrameHeader,
-  FramePanel,
-  FrameTitle,
-} from '@/components/reui/frame'
 import { api } from '@/lib/api-client'
+import { hasSharedPool } from '@/lib/failover-events'
 import {
   enabledHealthProviders,
   providerHealthStatuses,
+  resolveServiceDisplayHealth,
 } from '@/lib/health-log'
 import type { ServiceView, UpdateServiceConfigInput } from '@/lib/schemas'
 import {
@@ -51,6 +46,7 @@ import {
   domainsListQueryOptions,
   serviceBindingKeys,
   serviceDetailKeys,
+  serviceFailoverLogQueryOptions,
   serviceGroupKeys,
   serviceGroupsQueryOptions,
   serviceHealthLogQueryOptions,
@@ -84,6 +80,7 @@ interface OverviewPayload {
     priority: number
     consecutive_failures: number
     last_failure_reason: string | null
+    last_check_at?: string | null
   }>
 }
 
@@ -96,6 +93,7 @@ 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())
@@ -106,6 +104,35 @@ 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 uniqueEnabledIps = useMemo(
+    () =>
+      (service?.ips ?? []).filter((ip) => service?.ip_enabled[ip] !== false),
+    [service?.ips, service?.ip_enabled],
+  )
+  const displayHealth = useMemo(
+    () =>
+      resolveServiceDisplayHealth(
+        service?.health_status,
+        service?.ip_health ?? [],
+        logItems,
+      ),
+    [service?.health_status, service?.ip_health, logItems],
+  )
+  const showPoolPanel =
+    uniqueEnabledIps.length >= 2 && hasSharedPool(failoverBindings)
   const nodes = (nodesQuery.data as OverviewPayload['nodes']) ?? []
 
   const [editOpen, setEditOpen] = useState(false)
@@ -135,6 +162,7 @@ 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) }),
     ])
   }
 
@@ -209,22 +237,6 @@ function ServiceDetailPage() {
   const isError = viewQuery.isError || overviewQuery.isError
   const error = viewQuery.error ?? overviewQuery.error
 
-  const failoverEvents =
-    (nodes.length > 0 ? nodes : (overview?.nodes ?? []))
-      .filter(
-        (node) =>
-          node.health_status === 'unhealthy' ||
-          node.health_status === 'down' ||
-          node.health_status === 'checking',
-      )
-      .map((node) => ({
-        id: node.address,
-        title: `${node.address}: ${node.health_status}`,
-        detail: node.last_failure_reason
-          ? `${node.last_failure_reason} · fail ${node.consecutive_failures}`
-          : `fail ${node.consecutive_failures}`,
-      }))
-
   const enabledProviders = useMemo(
     () => enabledHealthProviders(service?.domains ?? []),
     [service],
@@ -257,7 +269,7 @@ function ServiceDetailPage() {
             actions={
               <>
                 <LbModeTile mode={service.lb_mode} />
-                <HealthCheckBadge status={service.health_status} />
+                <HealthCheckBadge status={displayHealth} />
                 <Tooltip>
                   <TooltipTrigger
                     render={
@@ -284,20 +296,20 @@ function ServiceDetailPage() {
                 id: 'status',
                 icon: <ActivityIcon />,
                 label: 'Статус',
-                value: service.health_status === 'up' ? 'OK' : service.health_status,
+                value: displayHealth === 'up' ? 'OK' : displayHealth,
                 variant:
-                  service.health_status === 'down'
+                  displayHealth === 'down'
                     ? 'destructive'
-                    : service.health_status === 'degraded'
+                    : displayHealth === 'degraded'
                       ? 'warning'
                       : 'default',
                 iconClassName:
-                  service.health_status === 'down'
+                  displayHealth === 'down'
                     ? 'text-destructive'
-                    : service.health_status === 'degraded'
+                    : displayHealth === 'degraded'
                       ? 'text-warning'
                       : 'text-success',
-                hint: <HealthCheckBadge status={service.health_status} size="xs" />,
+                hint: <HealthCheckBadge status={displayHealth} size="xs" />,
               },
               {
                 id: 'fqdn',
@@ -311,21 +323,33 @@ function ServiceDetailPage() {
                 icon: <NetworkIcon />,
                 label: 'IP',
                 value: String(service.ips.length),
-                hint: `${service.active_ips.length} в пуле`,
+                hint: showPoolPanel
+                  ? `${service.active_ips.length} в пуле`
+                  : uniqueEnabledIps.length <= 1
+                    ? 'без балансировки'
+                    : service.ips.join(', ') || 'нет',
               },
               {
                 id: 'pool',
                 icon: <ServerIcon />,
-                label: 'Активный пул',
-                value: String((overview?.active_addresses ?? service.active_ips).length),
-                hint: (overview?.active_addresses ?? service.active_ips).join(', ') || 'нет',
+                label: showPoolPanel ? 'Активный пул' : 'Адреса',
+                value: String(
+                  (overview?.active_addresses ?? service.active_ips).length,
+                ),
+                hint:
+                  (overview?.active_addresses ?? service.active_ips).join(', ') ||
+                  'нет',
               },
             ]}
           />
 
           <section
             aria-label="Мониторинг"
-            className="grid min-w-0 items-start gap-2 @3xl:grid-cols-2"
+            className={
+              showPoolPanel
+                ? 'grid min-w-0 items-start gap-2 @3xl:grid-cols-2'
+                : 'grid min-w-0 items-start gap-2'
+            }
           >
             <ServiceHealthMonitor
               items={logItems}
@@ -333,17 +357,15 @@ function ServiceDetailPage() {
               statuses={providerStatuses}
               isLoading={logQuery.isLoading}
             />
-            <Frame dense spacing="sm" className="min-w-0 w-full">
-              <FrameHeader>
-                <FrameTitle>Failover</FrameTitle>
-                <FrameDescription>
-                  Нездоровые ноды и причины последней ошибки
-                </FrameDescription>
-              </FrameHeader>
-              <FramePanel>
-                <FailoverTimeline events={failoverEvents} />
-              </FramePanel>
-            </Frame>
+            {showPoolPanel ? (
+              <ServiceFailoverPanel
+                lbMode={service.lb_mode}
+                ipHealth={service.ip_health}
+                bindings={failoverBindings}
+                history={failoverHistory}
+                probes={logItems}
+              />
+            ) : null}
           </section>
 
           {service.ips.length === 0 && service.domains.length === 0 ? (
@@ -356,6 +378,7 @@ 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
@@ -2332,6 +2341,113 @@ export function listIpHealthByServiceIds(
   return result;
 }
 
+function parseIpHealthState(status: string | null): IpHealthState | undefined {
+  if (
+    status === "up" ||
+    status === "down" ||
+    status === "degraded" ||
+    status === "unknown"
+  ) {
+    return status;
+  }
+  return undefined;
+}
+
+function bestAliveIpState(statuses: readonly IpHealthState[]): IpHealthState {
+  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 service+IP+provider from health_probe_log, then any-up per IP.
+ * Display overlay — hysteresis in ip_health_status is unchanged (DNS).
+ */
+export function listLatestLiveHealthByServiceIds(
+  db: Db,
+  serviceIds: number[],
+): Map<number, ServiceIpHealthRow[]> {
+  const result = new Map<number, ServiceIpHealthRow[]>();
+  if (serviceIds.length === 0) return result;
+  const idList = sql.join(
+    serviceIds.map((id) => sql`${id}`),
+    sql`, `,
+  );
+  const rows = db.all<{
+    service_id: number;
+    ip: string;
+    provider: string | null;
+    status: string | null;
+    latency_ms: number | null;
+    last_error: string | null;
+    colo: string | null;
+    checked_at: string | null;
+  }>(sql`
+    SELECT sb.service_id AS service_id,
+      l.ip AS ip,
+      l.provider AS provider,
+      l.status AS status,
+      l.latency_ms AS latency_ms,
+      l.error AS last_error,
+      l.colo AS colo,
+      l.checked_at AS checked_at
+    FROM health_probe_log l
+    INNER JOIN service_bindings sb
+      ON l.scope = 'binding' AND l.ref_id = sb.id
+    INNER JOIN (
+      SELECT sb2.service_id AS service_id,
+        l2.ip AS ip,
+        l2.provider AS provider,
+        MAX(l2.id) AS max_id
+      FROM health_probe_log l2
+      INNER JOIN service_bindings sb2
+        ON l2.scope = 'binding' AND l2.ref_id = sb2.id
+      WHERE sb2.service_id IN (${idList})
+      GROUP BY sb2.service_id, l2.ip, l2.provider
+    ) latest
+      ON latest.max_id = l.id
+  `);
+
+  const byServiceIp = new Map<
+    string,
+    { serviceId: number; ip: string; probes: ServiceIpHealthRow[] }
+  >();
+  for (const row of rows) {
+    const status = parseIpHealthState(row.status);
+    if (!status) continue;
+    const key = `${row.service_id}\0${row.ip}`;
+    const probe: ServiceIpHealthRow = {
+      ip: row.ip,
+      status,
+      latency_ms: row.latency_ms,
+      last_checked_at: row.checked_at,
+      last_error: row.last_error,
+      provider: normalizeStatusProvider(row.provider),
+      colo: row.colo,
+    };
+    const bucket = byServiceIp.get(key);
+    if (bucket) bucket.probes.push(probe);
+    else {
+      byServiceIp.set(key, {
+        serviceId: row.service_id,
+        ip: row.ip,
+        probes: [probe],
+      });
+    }
+  }
+
+  for (const { serviceId, ip, probes } of byServiceIp.values()) {
+    const status = bestAliveIpState(probes.map((probe) => probe.status));
+    const preferred =
+      probes.find((probe) => probe.status === status) ?? probes[0]!;
+    const list = result.get(serviceId) ?? [];
+    list.push({ ...preferred, ip, status });
+    result.set(serviceId, list);
+  }
+  return result;
+}
+
 export function mergeHealthAggregates(
   parts: Array<HealthAggregate | undefined | null>,
 ): HealthAggregate {
@@ -3055,3 +3171,51 @@ 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;