fix(services): снимать Down только с общего FQDN

Персональные A не крутить и не трогать при падении. Панель называется по режиму балансировки. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-20 17:51:49 +07:00
parent 3d0ea33baf
commit 2b150fa3a6
8 changed files with 221 additions and 48 deletions
@@ -1,5 +1,6 @@
 import type { LbMode } from "@cfdm/shared";
 import { failoverDesired } from "./failover.js";
+import { isSharedPool } from "./pool.js";
 import { roundRobinDesired } from "./round-robin.js";
 import type { LbIpRow, LbTargetConfig } from "./types.js";
 import { weightedDesired } from "./weighted.js";
@@ -7,6 +8,7 @@ import { weightedDesired } from "./weighted.js";
 export type { LbIpRow, LbTargetConfig } from "./types.js";
 export { isHealthy } from "./health.js";
 export { withBindingLock } from "./binding-lock.js";
+export { isSharedPool, shouldRecordFailoverDnsDiff } from "./pool.js";
 export { WEIGHTED_DNS_TTL, WEIGHTED_SLOT_MS, weightedDesired } from "./weighted.js";

 export function selectActiveIpsByMode(
@@ -24,6 +26,21 @@ export function selectActiveIpsByMode(
  return roundRobinDesired(rows);
 }

+export function resolveDesiredAIps(
+  config: LbTargetConfig,
+  rows: LbIpRow[],
+  fallbackIps: readonly string[],
+  nowMs = Date.now(),
+): string[] {
+  const fallback = [...fallbackIps];
+  if (!isSharedPool(fallback)) 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 "Weighted";
@@ -0,0 +1,18 @@
+import type { LbMode } from "@cfdm/shared";
+
+/** Shared pool FQDN — two or more A targets. Dedicated extra-FQDN has one IP. */
+export function isSharedPool(ips: readonly string[]): boolean {
+  return ips.length >= 2;
+}
+
+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));
+}
@@ -30,7 +30,10 @@ import { syncServiceToVpsTracker } from "./vps-tracker-sync.js";
 import { fireEnsureHealthWorker, DEFAULT_HEALTH_FALLBACKS } from "./health/health-worker-deploy.js";
 import {
  isHealthy,
+  isSharedPool,
+  resolveDesiredAIps,
  selectActiveIpsByMode,
+  shouldRecordFailoverDnsDiff,
  withBindingLock,
  WEIGHTED_DNS_TTL,
  type LbIpRow,
@@ -38,12 +41,12 @@ import {
 } from "./routing/index.js";

 export type { LbIpRow, LbTargetConfig };
-export { selectActiveIpsByMode };
+export { resolveDesiredAIps, selectActiveIpsByMode, shouldRecordFailoverDnsDiff };

 const AUTO_DNS_TTL = 1;

-function ttlForLbMode(mode: LbMode): number {
-  return mode === "weighted" ? WEIGHTED_DNS_TTL : AUTO_DNS_TTL;
+function ttlForBinding(mode: LbMode, ipCount: number): number {
+  return mode === "weighted" && ipCount >= 2 ? WEIGHTED_DNS_TTL : AUTO_DNS_TTL;
 }

 export function failoverARecordDiff(
@@ -72,6 +75,22 @@ function recordFailoverDnsDiff(
  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,
@@ -271,33 +290,17 @@ function getGroupLbState(
  };
 }

-function computeActiveIps(
-  db: Db,
-  scope: HealthCheckScope,
-  refId: number,
-): string[] {
-  const state =
-    scope === "binding"
-      ? getBindingLbState(db, refId)
-      : getGroupLbState(db, refId);
-  return selectActiveIpsByMode(state.config, state.rows);
-}
-
 function desiredAIps(
  db: Db,
  scope: HealthCheckScope,
  refId: number,
  fallbackIps: string[],
 ): string[] {
-  const config =
+  const state =
    scope === "binding"
-      ? getBindingLbState(db, refId).config
-      : getGroupLbState(db, refId).config;
-  if (config.lb_mode === "weighted" || config.health_check_enabled) {
-    const activeIps = computeActiveIps(db, scope, refId);
-    if (activeIps.length > 0) return activeIps;
-  }
-  return fallbackIps;
+      ? getBindingLbState(db, refId)
+      : getGroupLbState(db, refId);
+  return resolveDesiredAIps(state.config, state.rows, fallbackIps);
 }

 async function collectKnownZones(
@@ -349,7 +352,7 @@ async function buildView(db: Db, serviceId: number): Promise<ServiceView> {
    const { config, rows } = getBindingLbState(db, binding.id);
    const bindingActiveIps = targetCname
      ? []
-      : selectActiveIpsByMode(config, rows);
+      : resolveDesiredAIps(config, rows, targetIps);

    return {
      binding_id: binding.id,
@@ -638,6 +641,7 @@ async function syncBindingDns(
  }

  const binding = repos.getBinding(db, bindingId);
+  const configuredIps = repos.listBindingIps(db, bindingId);
  await syncBindingADns(
    db,
    cf,
@@ -645,7 +649,7 @@ async function syncBindingDns(
    domainId,
    hostname,
    desiredIps,
-    ttlForLbMode(binding.lb_mode),
+    ttlForBinding(binding.lb_mode, configuredIps.length),
  );
 }

@@ -1208,7 +1212,7 @@ async function syncGroupDomainDns(
    domainId,
    hostname,
    desiredIps,
-    ttlForLbMode(group.lb_mode),
+    ttlForBinding(group.lb_mode, fallbackIps.length),
  );
 }

@@ -1702,6 +1706,7 @@ export async function reconcileDnsForTarget(
      if (cnameTarget) return;
      const ips = repos.listServiceIps(db, service.id);
      const targetIps = repos.listBindingIps(db, binding.id);
+      if (!isSharedPool(targetIps)) return;
      validateTargetIpsInPool(targetIps, ips);
      const desiredIps = desiredAIps(db, "binding", refId, targetIps);
      await syncBindingDns(
@@ -1735,6 +1740,7 @@ export async function reconcileWeightedDns(
  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);
@@ -1743,7 +1749,7 @@ export async function reconcileWeightedDns(
        const service = repos.getService(db, latest.service_id);
        if (!shouldPushDns(db, service)) return;
        const targetIps = repos.listBindingIps(db, latest.id);
-        if (targetIps.length === 0) return;
+        if (!isSharedPool(targetIps)) return;
        const ips = repos.listServiceIps(db, service.id);
        validateTargetIpsInPool(targetIps, ips);
        const desiredIps = desiredAIps(db, "binding", latest.id, targetIps);
@@ -1,6 +1,8 @@
 import { describe, expect, it } from "vitest";
 import {
+  resolveDesiredAIps,
  selectActiveIpsByMode,
+  shouldRecordFailoverDnsDiff,
  type LbIpRow,
  type LbTargetConfig,
 } from "../src/services/service-config-service.js";
@@ -144,3 +146,75 @@ 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("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 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);
+  });
+});
@@ -63,8 +63,8 @@ export function FailoverTimeline({
    return (
      <EmptyState
        icon={ShieldCheckIcon}
-        title="Нет инцидентов Failover"
-        description="Нет Down и нет выходов из пула"
+        title="Нет инцидентов"
+        description="Нет Down и нет выходов из общего пула"
        stackedIcon={false}
        centered={false}
      />
@@ -13,7 +13,7 @@ import {
  type HealthLogProbe,
  type HealthLogStatus,
 } from '@/lib/health-log'
-import type { FailoverLogEntry } from '@/lib/schemas'
+import type { FailoverLogEntry, ServiceView } from '@/lib/schemas'
 import { Badge } from '@/components/reui/badge'
 import {
  Frame,
@@ -28,6 +28,23 @@ import {
  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
@@ -38,8 +55,17 @@ function failoverCountLabel(count: number): string {
  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'
+}
+
 /**
- * Failover — текущие Down + кто вышел из пула и кто вернулся.
+ * 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
@@ -48,16 +74,19 @@ function failoverCountLabel(count: number): string {
 * 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)
@@ -77,14 +106,13 @@ export function ServiceFailoverPanel({
  })
  const events = toFailoverEvents(overlayHealth, bindings)
  const mergedHistory = mergeFailoverHistory(history, probes, bindings)
-  const removedCount = events.filter((event) => event.kind === 'removed').length

  return (
    <Frame stacked spacing="sm" className="min-w-0 w-full">
      <FramePanel className="flex flex-col gap-3">
        <FrameHeader className="gap-1 px-0 py-0">
          <FrameTitle className="flex flex-wrap items-center gap-2">
-            Failover
+            {copy.title}
            {events.length > 0 ? (
              <Badge variant="destructive-light" size="xs" radius="full">
                {events.length}
@@ -95,20 +123,14 @@ export function ServiceFailoverPanel({
              </Badge>
            )}
          </FrameTitle>
-          <FrameDescription>
-            Текущие Down и история: кто вышел из пула и кто вернулся
-          </FrameDescription>
+          <FrameDescription>{copy.description}</FrameDescription>
        </FrameHeader>

        {events.length > 0 ? (
          <Alert variant="destructive">
            <UnplugIcon aria-hidden="true" />
            <AlertTitle>{failoverCountLabel(events.length)}</AlertTitle>
-            <AlertDescription>
-              {removedCount > 0
-                ? 'Сняты с FQDN или остались last-resort'
-                : 'Остались в A-записях как last-resort'}
-            </AlertDescription>
+            <AlertDescription>{alertDescription(events)}</AlertDescription>
          </Alert>
        ) : null}

@@ -137,6 +137,7 @@ describe('toFailoverEvents', () => {
      },
    ])
    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', () => {
@@ -151,10 +152,8 @@ describe('toFailoverEvents', () => {
      ],
    )
    expect(events[0]?.kind).toBe('last-resort')
-    expect(events[0]?.fqdns).toEqual(['nsgt.rkns.top'])
-    expect(failoverEventCopy(events[0]!)).toBe(
-      'Down, в A-записях last-resort на nsgt.rkns.top',
-    )
+    expect(events[0]?.fqdns).toEqual([])
+    expect(failoverEventCopy(events[0]!)).toBe('Down, в A-записях last-resort')
  })

  it('down без A-записи на configured FQDN — снятие', () => {
@@ -279,7 +278,9 @@ describe('mergeFailoverHistory', () => {
      'removed:probe',
    ])
    expect(history[0]?.copy).toBe('130.49.213.153 добавлена на gt.rkns.top')
-    expect(history[1]?.copy).toContain('вышла из пула')
+    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', () => {
@@ -291,4 +292,20 @@ describe('mergeFailoverHistory', () => {
    expect(history).toHaveLength(1)
    expect(history[0]?.source).toBe('dns')
  })
+
+  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([])
+  })
 })
@@ -44,6 +44,11 @@ export interface FailoverBindingPool {
  active: readonly string[]
 }

+/** Shared pool FQDN — two or more A targets. Dedicated extra-FQDN has one IP. */
+export function isSharedPoolBinding(binding: FailoverBindingPool): boolean {
+  return binding.configured.length >= 2
+}
+
 /** Инцидент только при down. up / degraded / unknown — не вывод из пула. */
 export function isFailoverEventStatus(status: string): boolean {
  return status === 'down'
@@ -72,6 +77,7 @@ export function toFailoverEvents(
    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)
@@ -98,7 +104,18 @@ function fqdnsForIp(
  ip: string,
  bindings: readonly FailoverBindingPool[],
 ): string[] {
-  return bindings.filter((binding) => binding.configured.includes(ip)).map((binding) => binding.fqdn)
+  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 {
@@ -201,7 +218,9 @@ export function mergeFailoverHistory(
  bindings: readonly FailoverBindingPool[] = [],
 ): FailoverHistoryItem[] {
  const merged = [
-    ...dns.map(dnsHistoryItem),
+    ...dns
+      .filter((item) => isPoolFqdn(item.fqdn, bindings))
+      .map(dnsHistoryItem),
    ...toIpAliveTransitions(probes).map((transition) =>
      probeHistoryItem(transition, bindings),
    ),