fix: cap job drain under Vercel 300s timeout

Limit process route to 2 rounds / ~8 jobs with a 240s wall so Hobby
functions exit cleanly; stale reclaim handles interrupted work.

Author: thribhuvan003

app/api/v1/internal/jobs/process/route.ts

@@ -2,21 +2,29 @@ import { assertCronAuth } from '@/lib/api/cron-auth';
 import { processAgentJobs } from '@/lib/jobs/process';
 import { runErasureBatch } from '@/lib/data-rights/erasure';
 
-/** One cron hit should clear backlog, not just 10 jobs every 5 minutes. */
-const DRAIN_ROUNDS = 4;
+/**
+ * Hobby functions die at ~300s. Keep headroom: GHA curl max-time is 280s.
+ * More backlog drains on the next 5m cron + reclaimStaleRunningJobs.
+ */
+const DRAIN_ROUNDS = 2;
+const WALL_MS = 240_000;
 
 async function processJobs(limit?: number) {
+  const started = Date.now();
   const erasures = await runErasureBatch();
   const totals = { processed: 0, succeeded: 0, failed: 0, reclaimed: 0 };
   for (let i = 0; i < DRAIN_ROUNDS; i += 1) {
-    const result = await processAgentJobs({ limit });
+    if (Date.now() - started > WALL_MS) break;
+    const result = await processAgentJobs({
+      limit: limit ?? 8,
+    });
     totals.processed += result.processed;
     totals.succeeded += result.succeeded;
     totals.failed += result.failed;
     totals.reclaimed += result.reclaimed;
     if (result.processed === 0) break;
   }
-  return Response.json({ ...totals, erasures });
+  return Response.json({ ...totals, erasures, elapsed_ms: Date.now() - started });
 }
 
 export async function GET(request: Request) {

lib/jobs/process.ts

@@ -6,7 +6,7 @@ import type { Database, Json } from "@/supabase/database.types";
 
 type AgentJobRow = Database["public"]["Tables"]["agent_jobs"]["Row"];
 
-const DEFAULT_BATCH_SIZE = 15;
+const DEFAULT_BATCH_SIZE = 8;
 const MAX_ATTEMPTS = 3;
 const RETRY_BASE_DELAY_MS = 60_000;
 const RETRY_MAX_DELAY_MS = 30 * 60_000;