maxpool 1.5.2 → 1.5.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "maxpool",
-  "version": "1.5.2",
+  "version": "1.5.3",
   "description": "Multi-account Claude Code proxy with adaptive, rate-aware load balancing across Claude accounts",
   "type": "module",
   "main": "src/index.js",

package/src/account-manager.js

@@ -48,6 +48,10 @@ const DEFAULT_SCHEDULER = {
   safetyMaxGlobalActive: 150,
   cooldownMs: 30_000,
   maxCooldownMs: 15 * 60_000,
+  // Fixed cooldown for NETWORK-class failures (lost connectivity / token-refresh
+  // fetch-failed). Short + non-escalating so the fleet auto-recovers seconds after
+  // connectivity returns, instead of the exponential maxCooldownMs bench.
+  networkCooldownMs: 5_000,
   weeklySoftThreshold: 0.65,
   weeklyReserveThreshold: 0.85,
   weeklyCriticalThreshold: 0.95,
@@ -1698,9 +1702,28 @@ export class AccountManager {
     console.log(`[Maxpool] Account "${account.name}" disabled after HTTP ${status} (${reason})`);
   }
 
-  markTransientFailure(accountIndex, reason = 'transient_error') {
+  markTransientFailure(accountIndex, reason = 'transient_error', { network = false } = {}) {
     const account = this.accounts[accountIndex];
     if (!account) return;
+
+    if (network) {
+      // A network-class failure (lost connectivity / token-refresh `fetch failed`)
+      // is a FLEET-WIDE condition, not this account's fault — every account fails at
+      // once. Use a SHORT FIXED cooldown so the whole fleet retries within seconds of
+      // connectivity returning and recovers AUTOMATICALLY, never the exponential
+      // 15-min bench that stranded the fleet long after a multi-hour outage and forced
+      // a manual restart (2026-06-29 hotel-network nightly cutoff). Deliberately does
+      // NOT bump consecutiveFailures: a network blip must not poison the scoring
+      // penalty (_scoreAccount) or prime the next REAL per-account failure for the max
+      // cooldown — so the counter stays a pure request-health signal and needs no reset.
+      account.failedRequests++;
+      account.lastError = reason;
+      account.lastErrorAt = Date.now();
+      account.cooldownUntil = Date.now() + this.scheduler.networkCooldownMs;
+      console.log(`[Maxpool] Account "${account.name}" cooling down for ${Math.ceil(this.scheduler.networkCooldownMs / 1000)}s after ${reason} (network — short fixed, auto-recovers)`);
+      return;
+    }
+
     const failures = Math.max(1, account.consecutiveFailures + 1);
     const cooldown = Math.min(
       this.scheduler.maxCooldownMs,
@@ -1812,7 +1835,11 @@ export class AccountManager {
         // a failed proactive refresh shouldn't kill a still-valid token
         if (!account.expiresAt || Date.now() >= account.expiresAt) {
           if (err.retryable) {
-            this.markTransientFailure(accountIndex, `token_refresh_${err.status || 'network'}`);
+            // A retryable refresh failure with NO HTTP status is a network/connectivity
+            // failure (fetch failed / timeout) → short fixed cooldown so it auto-recovers
+            // when the network returns. A retryable HTTP status (429 / 5xx from the OAuth
+            // endpoint) is server-side → keep the exponential backoff.
+            this.markTransientFailure(accountIndex, `token_refresh_${err.status || 'network'}`, { network: !err.status });
           } else {
             this.markAuthFailed(accountIndex, err.status || 401, 'token_refresh_failed');
           }

package/src/config.js

@@ -59,6 +59,10 @@ export function createDefaultConfig() {
       safetyMaxGlobalActive: 150,
       cooldownMs: 30_000,
       maxCooldownMs: 15 * 60_000,
+      // Fixed cooldown for network-class failures (lost connectivity / token-refresh
+      // fetch-failed) — short + non-escalating so the fleet auto-recovers seconds after
+      // connectivity returns, never the exponential maxCooldownMs bench.
+      networkCooldownMs: 5_000,
       // Weekly (7d) quota tiers — how aggressively to de-prioritise an account
       // as its weekly usage climbs. Each is a fraction (0..1) of the weekly
       // limit. Below soft = full speed; soft..reserve = mild penalty;

package/src/oauth.js

@@ -17,6 +17,11 @@ const DEFAULT_CLIENT_ID = '9d1c250a-e61b-44d9-88ed-5944d1962f5e';
 export async function refreshAccessToken(refreshToken, endpoint = DEFAULT_TOKEN_ENDPOINT) {
   const maxRetries = 2;
   const baseDelayMs = 500;
+  // Bound each refresh POST so a hung connect during an outage can't pin the
+  // single-flight _refreshPromise indefinitely (which would stall that account's
+  // recovery). Comfortably above a normal refresh latency; a timeout is classified
+  // as a network error below and retried / short-cooled.
+  const perAttemptTimeoutMs = 10_000;
 
   for (let attempt = 0; attempt <= maxRetries; attempt++) {
     try {
@@ -37,6 +42,7 @@ export async function refreshAccessToken(refreshToken, endpoint = DEFAULT_TOKEN_
           refresh_token: refreshToken,
           client_id: DEFAULT_CLIENT_ID,
         }),
+        signal: AbortSignal.timeout(perAttemptTimeoutMs),
       });
 
       if (!res.ok) {
@@ -60,6 +66,7 @@ export async function refreshAccessToken(refreshToken, endpoint = DEFAULT_TOKEN_
     } catch (err) {
       const isNetworkError = err instanceof Error &&
         (err.message.includes('fetch failed') ||
+          err.name === 'TimeoutError' || err.name === 'AbortError' ||
           (err.code === 'ECONNRESET' || err.code === 'ECONNREFUSED' ||
            err.code === 'ETIMEDOUT' || err.code === 'UND_ERR_CONNECT_TIMEOUT'));
 

package/src/server.js

@@ -855,7 +855,10 @@ async function forwardRequest(
         err.message.includes('terminated'));
 
     if (isTransient) {
-      accountManager.markTransientFailure(account.index, err.code || err.message || 'network_error');
+      // Network-class (ECONNRESET / fetch failed / timeout / terminated): short fixed
+      // cooldown, no exponential escalation — the fleet auto-recovers seconds after
+      // connectivity returns instead of being benched for up to 15 min.
+      accountManager.markTransientFailure(account.index, err.code || err.message || 'network_error', { network: true });
       accountManager.releaseAccount(lease);
       excludedIndexes.add(account.index);
       if (canRetryBufferedBody && retryCount + 1 < maxAttempts && !res.headersSent) {