maxpool 1.0.1 → 1.0.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "maxpool",
-  "version": "1.0.1",
+  "version": "1.0.3",
   "description": "Multi-account Claude Code proxy with adaptive, rate-aware load balancing across Claude accounts",
   "type": "module",
   "main": "src/index.js",

package/src/config.js

@@ -36,6 +36,12 @@ export function createDefaultConfig() {
       apiKey: 'mp-' + randomBytes(24).toString('base64url'),
     },
     upstream: 'https://api.anthropic.com',
+    // Per-account "stop using this account" gate, applied to BOTH the 5h
+    // session window and the 7d weekly window (whichever utilization is
+    // higher). 0.90 = stop routing to an account once it crosses 90% of a
+    // window, leaving a 10% safety margin so it is never hard-limited (429).
+    // Raise toward 0.97 to squeeze more out of accounts before rotating
+    // (less margin, slightly higher 429 risk); lower to rotate more eagerly.
     switchThreshold: 0.90,
     quotaProbeSeconds: 0, // background quota probe; 0 = off (opt-in)
     routing: {
@@ -48,11 +54,28 @@ export function createDefaultConfig() {
       safetyMaxGlobalActive: 150,
       cooldownMs: 30_000,
       maxCooldownMs: 15 * 60_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;
+      // reserve..critical = heavy penalty; above exhausted = effectively
+      // parked until the weekly window resets. Tune to trade burst capacity
+      // against weekly-limit safety.
+      weeklySoftThreshold: 0.65,
+      weeklyReserveThreshold: 0.85,
+      weeklyCriticalThreshold: 0.95,
+      weeklyExhaustedThreshold: 0.985,
     },
     retry: {
       maxAttemptsPerRequest: 0,
       maxRetryBufferBytes: 10 * 1024 * 1024,
     },
+    // On quit (q / Ctrl-C / SIGTERM): stop accepting new requests, give
+    // in-flight requests up to drainTimeoutMs to finish, then force-exit.
+    // A second signal forces an immediate exit. Kept short so quit works
+    // under a continuous request flood instead of hanging.
+    shutdown: {
+      drainTimeoutMs: 15_000,
+    },
     queue: {
       enabled: true,
       maxWaitMs: 24 * 60 * 60 * 1000,

package/src/index.js

@@ -193,7 +193,11 @@ async function serverWorkerCommand() {
   let syncTimer = null;
   let draining = false;
   let restartController = null;
-  const drainTimeoutMs = Math.max(1000, Number(config.shutdown?.drainTimeoutMs) || 10 * 60_000);
+  // Quit drains in-flight requests, then force-exits. Kept short so a single
+  // 'q' / Ctrl-C / SIGTERM actually quits under a continuous request flood
+  // (where there are always active requests) instead of waiting indefinitely.
+  // A second signal forces an immediate exit. Override via config.shutdown.drainTimeoutMs.
+  const drainTimeoutMs = Math.max(1000, Number(config.shutdown?.drainTimeoutMs) || 15_000);
   const hooks = {
     onRequestStart: (id, info) => {
       const accepted = restartController.requestStarted(id);
@@ -242,7 +246,7 @@ async function serverWorkerCommand() {
     if (tui?.running) tui.stop();
 
     console.log(`\n[Maxpool] Draining shutdown (${reason}).`);
-    console.log(`[Maxpool] Stopped accepting new requests; waiting for ${restartController.activeRequests.size} active request(s). Press Ctrl-C again to force.`);
+    console.log(`[Maxpool] Stopped accepting new requests; waiting up to ${Math.ceil(drainTimeoutMs / 1000)}s for ${restartController.activeRequests.size} active request(s), then forcing exit. Press Ctrl-C again to force now.`);
 
     let done = false;
     let reportTimer = null;

package/src/server.js

@@ -846,7 +846,7 @@ function unavailableMessage(accountManager, requestInfo = {}, retryAfter, willRe
   return `All ${n} accounts exhausted. Retry in ${retryAfter}s.`;
 }
 
-export const __serverTest = { unavailableMessage, isRetriableUpstreamStatus };
+export const __serverTest = { unavailableMessage, isRetriableUpstreamStatus, headerValue, getMaxpoolProfile };
 
 async function readErrorBody(upstreamRes, limitBytes = 64 * 1024) {
   if (!upstreamRes.body) return '';
@@ -1227,7 +1227,8 @@ function containsThinkingBlock(value) {
 }
 
 function getMaxpoolProfile(headers) {
-  const profile = String(headers['x-maxpool-profile'] || 'claude').trim().toLowerCase();
+  // headerValue() handles the x-teamclaude-* legacy fallback.
+  const profile = String(headerValue(headers, 'x-maxpool-profile') || 'claude').trim().toLowerCase();
   return profile || 'claude';
 }
 
@@ -1272,7 +1273,15 @@ function prepareRuntimeProviders(accountManager, headers) {
 }
 
 function headerValue(headers, name) {
-  const value = headers[name.toLowerCase()];
+  const lname = name.toLowerCase();
+  let value = headers[lname];
+  // Backward compatibility: sessions launched before the teamclaude→maxpool
+  // rename send x-teamclaude-* headers (a process's ANTHROPIC_CUSTOM_HEADERS is
+  // fixed at launch). Fall back to the legacy name so already-running sessions
+  // keep full routing/fallback without needing a restart.
+  if ((value == null || value === '') && lname.startsWith('x-maxpool-')) {
+    value = headers['x-teamclaude-' + lname.slice('x-maxpool-'.length)];
+  }
   if (Array.isArray(value)) return value[0];
   return value ? String(value).trim() : '';
 }