fix(services): сразу возвращать IP в DNS после первой успешной пробы

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

apps/api/src/services/health-check-service.ts

@@ -290,7 +290,7 @@ export interface RunAllChecksOptions {
     target: HealthCheckTarget,
     prevState: IpHealthState | null,
     nextState: IpHealthState,
-  ) => void;
+  ) => void | Promise<void>;
 }
 
 function sleep(ms: number): Promise<void> {
@@ -326,7 +326,7 @@ function logSourceResult(
   });
 }
 
-function applyAggregatedStatus(
+async function applyAggregatedStatus(
   db: Db,
   target: HealthCheckTarget,
   sources: Array<{ provider: HealthCheckProvider; result: ProbeResult }>,
@@ -397,7 +397,7 @@ function applyAggregatedStatus(
     });
   }
   if (prevState !== state) {
-    options.onStatusChange?.(target, prevState, state);
+    await options.onStatusChange?.(target, prevState, state);
   }
 }
 
@@ -513,7 +513,7 @@ export async function runAllChecks(
       for (const source of sources) {
         logSourceResult(db, target, source.provider, source.result);
       }
-      applyAggregatedStatus(db, target, sources, options);
+      await applyAggregatedStatus(db, target, sources, options);
     }
   }
 

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

@@ -1,16 +1,16 @@
 import type { LbIpRow } from "./types.js";
-import { isHealthy } from "./health.js";
+import { isPoolMember } from "./health.js";
 
 export function failoverDesired(rows: LbIpRow[]): string[] {
   if (rows.length === 0) return [];
-  const healthy = rows.filter((r) => isHealthy(r.health));
-  const pool = healthy.length > 0 ? healthy : rows;
+  const live = rows.filter((r) => isPoolMember(r.health));
+  const pool = live.length > 0 ? live : rows;
   const sorted = [...pool].sort(
     (a, b) => a.priority - b.priority || a.weight - b.weight,
   );
   const minPriority = sorted[0]!.priority;
   const primaries = sorted.filter((r) => r.priority === minPriority);
-  if (healthy.length > 0) {
+  if (live.length > 0) {
     return primaries.map((r) => r.ip);
   }
   return [sorted[0]!.ip];

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

@@ -3,3 +3,12 @@ import type { IpHealthState, NodeHealthState } from "@cfdm/shared";
 export function isHealthy(state: IpHealthState | NodeHealthState | string): boolean {
   return state === "up" || state === "healthy";
 }
+
+export function isDown(state: IpHealthState | NodeHealthState | string): boolean {
+  return state === "down" || state === "unhealthy";
+}
+
+/** A-pool membership: only Down is drained. Recovering (unknown/checking) and Slow return immediately. */
+export function isPoolMember(state: IpHealthState | NodeHealthState | string): boolean {
+  return !isDown(state);
+}

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

@@ -6,7 +6,7 @@ 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 { isDown, isHealthy, isPoolMember } from "./health.js";
 export { withBindingLock } from "./binding-lock.js";
 export {
   canApplyLb,

apps/api/src/services/routing/round-robin.ts

@@ -1,8 +1,8 @@
 import type { LbIpRow } from "./types.js";
-import { isHealthy } from "./health.js";
+import { isPoolMember } from "./health.js";
 
 export function roundRobinDesired(rows: LbIpRow[]): string[] {
-  const healthy = rows.filter((r) => isHealthy(r.health));
-  const pool = healthy.length > 0 ? healthy : rows;
+  const live = rows.filter((r) => isPoolMember(r.health));
+  const pool = live.length > 0 ? live : rows;
   return pool.map((r) => r.ip);
 }

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

@@ -1,5 +1,5 @@
 import type { LbIpRow } from "./types.js";
-import { isHealthy } from "./health.js";
+import { isPoolMember } from "./health.js";
 
 /** Slot length for time-sliced weighted DNS (one A at a time). */
 export const WEIGHTED_SLOT_MS = 60_000;
@@ -9,8 +9,8 @@ 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;
+  const live = rows.filter((r) => isPoolMember(r.health));
+  const pool = live.length > 0 ? live : rows;
   if (pool.length === 1) return [pool[0]!.ip];
 
   const sorted = [...pool].sort((a, b) => a.ip.localeCompare(b.ip));

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

@@ -30,7 +30,7 @@ import { syncServiceToVpsTracker } from "./vps-tracker-sync.js";
 import { fireEnsureHealthWorker, DEFAULT_HEALTH_FALLBACKS } from "./health/health-worker-deploy.js";
 import {
   canApplyLb,
-  isHealthy,
+  isPoolMember,
   isSharedPool,
   resolveDesiredAIps,
   selectActiveIpsByMode,
@@ -287,7 +287,7 @@ function getGroupLbState(
         } else {
           existing.weight += weight;
           existing.priority = Math.min(existing.priority, priority);
-          if (isHealthy(existing.health) && status && !isHealthy(status.status as IpHealthState)) {
+          if (isPoolMember(existing.health) && status && !isPoolMember(status.status as IpHealthState)) {
             existing.health = status.status as IpHealthState;
           }
         }

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

@@ -70,6 +70,18 @@ describe("selectActiveIpsByMode", () => {
     expect(selectActiveIpsByMode(config, rows)).toEqual(["1.1.1.1"]);
   });
 
+  it("failover returns recovering unknown primary immediately", () => {
+    const config: LbTargetConfig = {
+      lb_mode: "failover",
+      health_check_enabled: true,
+    };
+    const rows = [
+      row("1.1.1.1", { priority: 1, health: "unknown" }),
+      row("2.2.2.2", { priority: 2, health: "up" }),
+    ];
+    expect(selectActiveIpsByMode(config, rows)).toEqual(["1.1.1.1"]);
+  });
+
   it("failover falls back to min-priority ip among all when none healthy", () => {
     const config: LbTargetConfig = {
       lb_mode: "failover",
@@ -105,6 +117,17 @@ describe("selectActiveIpsByMode", () => {
     ).toEqual(["1.1.1.1"]);
   });
 
+  it("weighted includes recovering unknown in the cycle", () => {
+    const rows = [
+      row("1.1.1.1", { weight: 1, health: "up" }),
+      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 with one ip always returns that ip", () => {
     expect(
       selectActiveIpsByMode(weightedConfig, [row("1.1.1.1", { weight: 5 })], 0),
@@ -126,7 +149,7 @@ describe("selectActiveIpsByMode", () => {
     expect(selectActiveIpsByMode(weightedConfig, [], 0)).toEqual([]);
   });
 
-  it("round_robin excludes unknown when another ip is up", () => {
+  it("round_robin puts recovering unknown back with live ips", () => {
     const config: LbTargetConfig = {
       lb_mode: "round_robin",
       health_check_enabled: true,
@@ -135,7 +158,25 @@ describe("selectActiveIpsByMode", () => {
       row("1.1.1.1", { health: "up" }),
       row("2.2.2.2", { health: "unknown" }),
     ];
-    expect(selectActiveIpsByMode(config, rows)).toEqual(["1.1.1.1"]);
+    expect(selectActiveIpsByMode(config, rows).sort()).toEqual([
+      "1.1.1.1",
+      "2.2.2.2",
+    ]);
+  });
+
+  it("round_robin keeps degraded in the pool with live ips", () => {
+    const config: LbTargetConfig = {
+      lb_mode: "round_robin",
+      health_check_enabled: true,
+    };
+    const rows = [
+      row("1.1.1.1", { health: "up" }),
+      row("2.2.2.2", { health: "degraded" }),
+    ];
+    expect(selectActiveIpsByMode(config, rows).sort()).toEqual([
+      "1.1.1.1",
+      "2.2.2.2",
+    ]);
   });
 
   it("returns empty array for no rows", () => {

docs/Home.md

@@ -44,9 +44,10 @@ health-check работают на двух уровнях:
 - **Change Domain** — перенос привязок между зонами `POST /api/v1/services/:id/change-domain`.
 
 Режимы LB: `round_robin`, `failover`, `weighted`. В Cloudflare free `weighted`
-работает как `round_robin` (одна A на IP). `unknown` **не** считается healthy и
-не попадает в пул, пока нет успешных проб; восстановление — `UNHEALTHY → CHECKING → HEALTHY`
-после `HEALTH_SUCCESS_RECOVERIES` (default 2). Пороги и cron движка задаются в
+работает как `round_robin` (одна A на IP). В A-пул попадает всё, кроме **Down**
+(`unknown` / checking / Slow возвращаются в DNS на первой успешной пробе).
+Бейдж Healthy — `UNHEALTHY → CHECKING → HEALTHY` после `HEALTH_SUCCESS_RECOVERIES`
+(default 2). Пороги и cron движка задаются в
 **Настройки → Health-check** (env — fallback, пока значения не сохранены в UI).
 
 ### Источники проб: Local, Cloudflare Worker, Globalping