BREAKING: Cloudflare MINOR 🟡 workarounds inside [19g1m7tsvncw]

Cloudflare is down: Some Replicate models are failing to start. Immediate workarounds for indie hackers.

BREAKING: Cloudflare Incident - Replicate Models Failing to Start

Status: INVESTIGATING | Severity: MINOR | Impact: Partial Disruption

Last Updated: NOW | Incident ID: 19g1m7tsvncw

---

What's Down & Who's Affected

Cloudflare is reporting issues with some Replicate models failing to start on their infrastructure. This affects:

Indie projects using Replicate API calls routed through Cloudflare Workers

Applications with image generation, video processing, or LLM inference workflows

Sites using Cloudflare Pages Functions that call Replicate endpoints

Teams relying on Replicate-to-Cloudflare integrations for serverless pipelines

Not affected: Direct Replicate API calls (without Cloudflare routing). Cloudflare's DNS, DDoS protection, and CDN services are operating normally.

---

Immediate Workarounds (Do This NOW)

1. Bypass Cloudflare for Replicate Calls

If you're routing Replicate requests through Cloudflare Workers: ```javascript // Instead of: cloudflare-worker → replicate // Use: client → replicate directly

const response = await fetch('https://api.replicate.com/v1/predictions', { method: 'POST', headers: { 'Authorization': Token ${REPLICATE_API_TOKEN} }, body: JSON.stringify({ version, input }) }); ```

2. Implement Request Retry Logic

Add exponential backoff for failed model starts: ```javascript const maxRetries = 5; for (let i = 0; i < maxRetries; i++) { try { const prediction = await replicate.run(model, { input }); break; } catch (err) { await new Promise(r => setTimeout(r, 1000 * Math.pow(2, i))); } } ```

3. Queue-Based Architecture

Don't wait for model responses synchronously. Use job queues (Bull, AWS SQS) to decouple requests from processing.

---

How to Check If Your Project Is Affected

1. Check Replicate Status: Visit [replicate.com/status](https://replicate.com/status) and [Cloudflare status](https://www.cloudflarestatus.com) 2. Test Your Integration: Run a simple model prediction on your affected endpoint 3. Monitor Logs: Check for 503/504 errors from Replicate with model startup timeouts 4. API Response Code: Watch for "status": "failed" with message containing "failed to start"

---

Alternative Tools to Consider

If you need redundancy:

Together AI - Similar inference API, different infrastructure

Hugging Face Inference API - Text/vision models, enterprise-grade

AWS SageMaker - Enterprise option, higher cost but isolated infrastructure

Modal Labs - GPU-optimized serverless alternative to Cloudflare Workers

Banana.dev - Replicate competitor with similar pricing

---

How to Monitor Recovery

1. Follow real-time updates: [Cloudflare Status](https://www.cloudflarestatus.com) + [Replicate Status](https://replicate.com/status) 2. Subscribe to alerts: Enable Slack/Email notifications on both platforms 3. Test predictions: Use this test script every 5 minutes: ```bash curl -X POST https://api.replicate.com/v1/predictions \ -H "Authorization: Token $REPLICATE_API_TOKEN" \ -d '{"version": "test", "input": {}}' ``` 4. Check incident updates: Refresh this page for status changes

---

Key Takeaway

This is MINOR and partial—direct Replicate calls are fine. Avoid routing through Cloudflare Workers until Cloudflare posts all-clear. Use this time to implement proper queue-based architecture for production resilience.

Questions? Post in #infra on the community Slack. We're tracking this live.

---

*Last checked: NOW UTC | Next update: 15 minutes*