Loop Breaker Skill

Loop Breaker Skill

Scaffold protective middleware to prevent infinite recursion, self-calling Workers, and runaway loops that cause "denial-of-wallet" attacks on your Cloudflare bill.

Why This Matters

In serverless, a loop isn't just a frozen browser tab—it's a direct billing multiplier:

Recursion Depth Middleware

The most common and costly loop occurs when a Worker calls itself (or calls a second Worker that calls the first) via fetch(). This creates an infinite chain of HTTP requests.

Pattern: X-Recursion-Depth Header

Core Implementation:

// src/​middleware/​recursion-guard.ts
import { createMiddleware } from 'hono/​factory';
import type { Bindings } from '../​types';

const RECURSION_HEADER = 'X-Recursion-Depth';
const DEFAULT_DEPTH_LIMIT = 3;

interface RecursionGuardOptions {
  maxDepth?: number;
  onLimitExceeded?: (depth: number) => Response;
}

export const recursionGuard = (options: RecursionGuardOptions = {}) => {
  const maxDepth = options.maxDepth ?? DEFAULT_DEPTH_LIMIT;

  return createMiddleware<{ Bindings: Bindings }>(async (c, next) => {
    const currentDepth = parseInt(c.req.header(RECURSION_HEADER) || '0', 10);

    if (currentDepth > maxDepth) {
      console.error(`Recursion limit exceeded: depth=${currentDepth}, limit=${maxDepth}`);

      if (options.onLimitExceeded) {
        return options.onLimitExceeded(currentDepth);
      }

      // HTTP 508 Loop Detected (RFC 5842)
      return c.json({
        error: 'Loop Detected',
        message: `Recursion depth ${currentDepth} exceeds limit ${maxDepth}`,
        code: 'RECURSION_LIMIT_EXCEEDED'
      }, 508);
    }

    // Store for use in outgoing requests
    c.set('recursionDepth', currentDepth);

    await next();
  });
};

/​**
 * Create headers for outgoing fetch requests that preserve recursion tracking
 */
export function createRecursionHeaders(
  currentHeaders: Headers,
  currentDepth: number
): Headers {
  const newHeaders = new Headers(currentHeaders);
  newHeaders.set(RECURSION_HEADER, (currentDepth + 1).toString());
  return newHeaders;
}

Usage in Hono App

// src/​index.ts
import { Hono } from 'hono';
import { recursionGuard, createRecursionHeaders } from './​middleware/​recursion-guard';
import type { Bindings } from './​types';

// Extend Hono context to include recursionDepth
type Variables = {
  recursionDepth: number;
};

const app = new Hono<{ Bindings: Bindings; Variables: Variables }>();

// Apply recursion guard to all routes
app.use('*', recursionGuard({ maxDepth: 3 }));

// Example: Worker that calls another Worker
app.post('/​webhook', async (c) => {
  const currentDepth = c.get('recursionDepth');

  // When making outgoing requests, pass the incremented depth
  const response = await fetch('https://other-worker.example.com/​process', {
    method: 'POST',
    headers: createRecursionHeaders(c.req.raw.headers, currentDepth),
    body: await c.req.text(),
  });

  return c.json({ processed: true, depth: currentDepth });
});

export default app;

Types Extension

Add to src/​types.ts:

// Add to Hono Variables
declare module 'hono' {
  interface ContextVariableMap {
    recursionDepth: number;
  }
}

Service Binding Recursion Guard

When using Service Bindings (RPC), you still need recursion protection but headers don't apply. Use a context-passing pattern:

// src/​services/​with-recursion-context.ts

interface RecursionContext {
  depth: number;
  maxDepth: number;
  callChain: string[];
}

export function createRecursionContext(
  workerName: string,
  existingContext?: RecursionContext
): RecursionContext {
  const ctx = existingContext ?? { depth: 0, maxDepth: 5, callChain: [] };

  return {
    depth: ctx.depth + 1,
    maxDepth: ctx.maxDepth,
    callChain: [...ctx.callChain, workerName],
  };
}

export function checkRecursionLimit(ctx: RecursionContext): void {
  if (ctx.depth > ctx.maxDepth) {
    throw new RecursionLimitError(ctx);
  }

  // Detect cycles in call chain
  const seen = new Set<string>();
  for (const worker of ctx.callChain) {
    if (seen.has(worker)) {
      throw new RecursionCycleError(worker, ctx.callChain);
    }
    seen.add(worker);
  }
}

class RecursionLimitError extends Error {
  constructor(ctx: RecursionContext) {
    super(`Recursion limit exceeded: ${ctx.depth} > ${ctx.maxDepth}`);
    this.name = 'RecursionLimitError';
  }
}

class RecursionCycleError extends Error {
  constructor(worker: string, chain: string[]) {
    super(`Recursion cycle detected: ${worker} in chain [${chain.join(' -> ')}]`);
    this.name = 'RecursionCycleError';
  }
}

Service Binding Usage

// When calling another Worker via Service Binding
const ctx = createRecursionContext('api-gateway', incomingContext);
checkRecursionLimit(ctx);

const result = await c.env.DATA_SERVICE.fetchUser(userId, ctx);

Queue Idempotency Guard

Cloudflare Queues can create "Retry Loops" when a Worker fails to process a message due to a persistent bug. This burns CPU and D1 writes on every retry.

// src/​middleware/​idempotency-guard.ts

interface IdempotencyOptions {
  kv: KVNamespace;
  ttlSeconds?: number;
  keyPrefix?: string;
}

export async function ensureIdempotent(
  messageId: string,
  options: IdempotencyOptions
): Promise<{ alreadyProcessed: boolean; markProcessed: () => Promise<void> }> {
  const { kv, ttlSeconds = 86400, keyPrefix = 'processed:' } = options;
  const key = `${keyPrefix}${messageId}`;

  const existing = await kv.get(key);
  if (existing) {
    return {
      alreadyProcessed: true,
      markProcessed: async () => {},
    };
  }

  return {
    alreadyProcessed: false,
    markProcessed: async () => {
      await kv.put(key, Date.now().toString(), { expirationTtl: ttlSeconds });
    },
  };
}

// Queue consumer usage
export default {
  async queue(batch: MessageBatch<QueueMessage>, env: Bindings): Promise<void> {
    for (const msg of batch.messages) {
      const { alreadyProcessed, markProcessed } = await ensureIdempotent(
        msg.id,
        { kv: env.IDEMPOTENCY_KV }
      );

      if (alreadyProcessed) {
        console.log(`Skipping duplicate message: ${msg.id}`);
        msg.ack();
        continue;
      }

      try {
        await processMessage(msg.body, env);
        await markProcessed();
        msg.ack();
      } catch (error) {
        console.error(`Failed to process message ${msg.id}:`, error);
        msg.retry(); // Will go to DLQ after max_retries
      }
    }
  },
};

Durable Object Loop Protection

Durable Objects billing includes duration (wall time). A setInterval without clear termination keeps the DO active and billing.

Pattern: Hibernation-Aware Intervals

// DO with safe interval handling
export class SafeTimerDO implements DurableObject {
  private intervalId: ReturnType<typeof setInterval> | null = null;
  private state: DurableObjectState;

  constructor(state: DurableObjectState) {
    this.state = state;
  }

  // BAD: This keeps DO alive forever
  // startBadTimer() {
  //   setInterval(() => this.doWork(), 1000);
  // }

  // GOOD: Alarm-based timing (hibernation compatible)
  async startSafeTimer(intervalMs: number): Promise<void> {
    await this.state.storage.setAlarm(Date.now() + intervalMs);
  }

  async alarm(): Promise<void> {
    await this.doWork();
    // Reschedule only if still needed
    const shouldContinue = await this.checkShouldContinue();
    if (shouldContinue) {
      await this.state.storage.setAlarm(Date.now() + 1000);
    }
    // Otherwise, DO hibernates and stops billing
  }

  private async checkShouldContinue(): Promise<boolean> {
    // Check if there's a reason to keep running
    const activeConnections = await this.state.storage.get('activeConnections');
    return (activeConnections as number) > 0;
  }

  private async doWork(): Promise<void> {
    // Periodic work here
  }

  async fetch(request: Request): Promise<Response> {
    const url = new URL(request.url);

    if (url.pathname === '/​start-timer') {
      await this.startSafeTimer(1000);
      return new Response('Timer started with hibernation support');
    }

    return new Response('OK');
  }
}

D1 Loop Protection

Detect and prevent N+1 query patterns that multiply costs.

// src/​utils/​query-batcher.ts

/​**
 * Prevents N+1 queries by batching related lookups
 */
export class QueryBatcher<K, V> {
  private pending: Map<K, { resolve: (v: V | null) => void; reject: (e: Error) => void }[]> = new Map();
  private scheduled = false;

  constructor(
    private batchFn: (keys: K[]) => Promise<Map<K, V>>,
    private maxBatchSize = 100,
    private delayMs = 0
  ) {}

  async load(key: K): Promise<V | null> {
    return new Promise((resolve, reject) => {
      const callbacks = this.pending.get(key) ?? [];
      callbacks.push({ resolve, reject });
      this.pending.set(key, callbacks);

      if (!this.scheduled) {
        this.scheduled = true;
        if (this.delayMs > 0) {
          setTimeout(() => this.dispatch(), this.delayMs);
        } else {
          queueMicrotask(() => this.dispatch());
        }
      }
    });
  }

  private async dispatch(): Promise<void> {
    const batch = new Map(this.pending);
    this.pending.clear();
    this.scheduled = false;

    const keys = Array.from(batch.keys()).slice(0, this.maxBatchSize);

    try {
      const results = await this.batchFn(keys);

      for (const [key, callbacks] of batch) {
        const value = results.get(key) ?? null;
        for (const { resolve } of callbacks) {
          resolve(value);
        }
      }
    } catch (error) {
      for (const [, callbacks] of batch) {
        for (const { reject } of callbacks) {
          reject(error as Error);
        }
      }
    }
  }
}

// Usage example - prevents N+1 when fetching user data
const userBatcher = new QueryBatcher<string, User>(async (userIds) => {
  const placeholders = userIds.map(() => '?').join(',');
  const results = await db
    .prepare(`SELECT * FROM users WHERE id IN (${placeholders})`)
    .bind(...userIds)
    .all<User>();

  return new Map(results.results.map(u => [u.id, u]));
});

// Instead of: for (const id of userIds) { await getUser(id); } // N queries
// Use: await Promise.all(userIds.map(id => userBatcher.load(id))); // 1 query

Configuration Templates

wrangler.jsonc with Loop Protection

{
  "name": "loop-protected-worker",
  "main": "src/​index.ts",
  "compatibility_date": "2025-01-01",

  // Billing Safety: Kill runaway loops fast
  "limits": {
    "cpu_ms": 100
  },

  // Use Service Bindings instead of fetch() chains
  "services": [
    { "binding": "AUTH_SERVICE", "service": "auth-worker" }
  ],

  // Queue with DLQ to break retry loops
  "queues": {
    "consumers": [{
      "queue": "events",
      "max_retries": 1,
      "dead_letter_queue": "events-dlq",
      "max_concurrency": 10
    }]
  },

  // KV for idempotency tracking
  "kv_namespaces": [
    { "binding": "IDEMPOTENCY_KV", "id": "..." }
  ]
}

Detection Checklist

Use this checklist when reviewing code for loop vulnerabilities:

Static Analysis

• Search for fetch( calls that might hit the same Worker
• Check for while(true) or for(;;) without break conditions
• Look for recursive function calls without depth limits
• Verify setInterval in Durable Objects has termination logic
• Check D1 queries inside iteration blocks
• Verify R2 writes are not inside high-frequency loops

Runtime Protection

• limits.cpu_ms set in wrangler config
• Recursion depth header middleware applied
• Queue consumers have idempotency checks
• DLQ configured for all queues
• Alarm API used instead of setInterval in DOs

Anti-Patterns

Related Skills

• architect: Billing Safety Limits section
• guardian: Loop-Sensitive Resource Auditing
• implement: Queue Consumer with Idempotency
• patterns/​circuit-breaker: External API loop protection