Session 7b: Fix pipeline - body parser, Redis queueing, poller visibility, auto-start

src/app.js

@@ -51,7 +51,11 @@ app.use(missionHeader);
 // Stripe webhook needs raw body — must be before express.json()
 app.use('/api/stripe/webhook', express.raw({ type: 'application/json' }));
 
-app.use(express.json());
+// Body parser limit raised to 10MB to accommodate full-slate poller
+// payloads. The default 100KB rejected real WNBA slates with 413.
+// Per-route limits below can tighten or loosen this for specific paths.
+app.use(express.json({ limit: '10mb' }));
+app.use(express.urlencoded({ extended: true, limit: '10mb' }));
 
 // Health check — public minimal status (Coolify, uptime monitors). Detailed
 // adapter + Python service status only with X-VYNDR-Internal-Key.
@@ -119,9 +123,11 @@ app.use('/api/waitlist', waitlistRoutes);
 app.use('/api/pipeline', pipelineRoutes);
 app.use('/api/share-card', shareCardRoutes);
 app.use('/api/push', pushRoutes);
-// Resolution payloads carry full ESPN box scores (50-100KB). Scope a larger
-// limit to /api/grading only so the other routes keep the safer 100KB default.
-app.use('/api/grading', express.json({ limit: '2mb' }), gradingRoutes);
+// Resolution payloads carry full ESPN box scores plus per-game prop
+// arrays. Full-slate WNBA / MLB resolves exceed 2MB in practice — keep
+// /api/grading aligned with the global 10MB ceiling. Correction sweep
+// stays small (just a window-hours integer + flags).
+app.use('/api/grading', express.json({ limit: '10mb' }), gradingRoutes);
 app.use('/api/grading', express.json({ limit: '256kb' }), correctionRoutes);
 const widgetRoutes = require('./routes/widget');
 app.use('/api/widget', widgetRoutes);

src/utils/redis.js

@@ -11,10 +11,13 @@ let degraded = false;
 function getRedisClient() {
   if (!client) {
     client = new Redis(process.env.REDIS_URL || 'redis://127.0.0.1:6379', {
-      // Don't block the app starting up if Redis is down at boot. Failed
-      // commands fall through to the cache-miss path immediately.
+      // Eager connect — but allow commands to QUEUE while the TCP/auth
+      // handshake completes. Without offline queueing, pollers booting
+      // alongside the API throw "Stream isn't writeable" before Redis
+      // is ready. With queueing, ioredis flushes the backlog the moment
+      // the connection enters READY state.
       lazyConnect: false,
-      enableOfflineQueue: false,
+      enableOfflineQueue: true,
       maxRetriesPerRequest: 1,
       retryStrategy(times) {
         // Exponential backoff up to 30s. ioredis keeps trying forever on its
@@ -30,7 +33,11 @@ function getRedisClient() {
       }
     });
     client.on('ready', () => {
+      // Surface the ready transition in PM2/container logs so operators
+      // can confirm the connection actually established. Distinct from
+      // reconnect-after-outage which the line below logs at info.
       if (degraded) console.info('[redis] reconnected, leaving degraded mode');
+      else console.log('[redis] connected and ready');
       degraded = false;
     });
   }

src/utils/supabase.js

@@ -15,9 +15,14 @@ function getSupabaseClient() {
 
 function getSupabaseServiceClient() {
   if (!serviceClient) {
+    // SUPABASE_SERVICE_ROLE_KEY is the canonical name across .env.example
+    // and the Next.js side. The fallback to SUPABASE_SERVICE_KEY supports
+    // any deployment still set with the legacy variable so the rename
+    // doesn't take down a running instance — remove the fallback once all
+    // Coolify envs are updated.
     serviceClient = createClient(
       process.env.SUPABASE_URL,
-      process.env.SUPABASE_SERVICE_KEY
+      process.env.SUPABASE_SERVICE_ROLE_KEY || process.env.SUPABASE_SERVICE_KEY
     );
   }
   return serviceClient;