fix(resilience): honor model lockout max cooldown

open-sse/services/accountFallback.ts

@@ -18,6 +18,7 @@ import {
   DEFAULT_RESILIENCE_SETTINGS,
   resolveResilienceSettings,
 } from "../../src/lib/resilience/settings";
+import { resolveModelLockoutSettings } from "../../src/lib/resilience/modelLockoutSettings";
 import {
   getAllCircuitBreakerStatuses,
   getCircuitBreaker,
@@ -37,6 +38,7 @@ export type ProviderProfile = {
   useUpstreamRetryHints: boolean;
   /** Issue #2100 follow-up. Stored override; undefined → per-provider default. */
   useUpstream429BreakerHints?: boolean;
+  maxCooldownMs: number;
   maxBackoffSteps: number;
   failureThreshold: number;
   resetTimeoutMs: number;
@@ -326,6 +328,7 @@ function buildProviderProfile(
     // Issue #2100 follow-up: propagate stored override (boolean | undefined)
     // so the runtime resolver picks user setting first, then per-provider default.
     useUpstream429BreakerHints: connectionCooldown.useUpstream429BreakerHints,
+    maxCooldownMs: resolveModelLockoutSettings(settings).maxCooldownMs,
     maxBackoffSteps: connectionCooldown.maxBackoffSteps,
     failureThreshold: providerBreaker.failureThreshold,
     resetTimeoutMs: providerBreaker.resetTimeoutMs,

src/sse/services/auth.ts

@@ -1916,6 +1916,7 @@ export async function markAccountUnavailable(
     const disableCooling = connProviderSpecificData.disableCooling === true;
 
     const isPerModelQuotaProvider = hasPerModelQuota(provider, model, connectionPassthroughModels);
+    const modelLockoutOptions = { maxCooldownMs: effectiveProviderProfile?.maxCooldownMs };
     if (
       isPerModelQuotaProvider &&
       provider &&
@@ -1942,6 +1943,7 @@ export async function markAccountUnavailable(
           : (fallbackResult.baseCooldownMs ?? effectiveProviderProfile?.baseCooldownMs ?? 0),
         effectiveProviderProfile,
         {
+          ...modelLockoutOptions,
           exactCooldownMs:
             fallbackResult.usedUpstreamRetryHint === true ? fallbackResult.cooldownMs : null,
           maxCooldownMs: mlSettings.maxCooldownMs,
@@ -2019,6 +2021,7 @@ export async function markAccountUnavailable(
           COOLDOWN_MS.serviceUnavailable,
         effectiveProviderProfile,
         {
+          ...modelLockoutOptions,
           exactCooldownMs:
             fallbackResult.usedUpstreamRetryHint === true ? fallbackResult.cooldownMs : null,
           maxCooldownMs: mlSettings.maxCooldownMs,