Written from an Azure-to-Cloudflare migration perspective. For evaluating migration of existing Azure Functions + Blob Storage architecture.

1. Sign-Up — 3 Minutes

Step 1. Create Account

  1. Go to https://dash.cloudflare.com/sign-up
  2. Enter email + password (or GitHub SSO)
  3. Verify email (check spam folder too)

No credit card required. Workers, Pages, R2, KV free tiers all available immediately without a credit card.

Step 2. Add Domain (Optional)

If you have an existing domain:

  1. Dashboard -> Add a Site
  2. Enter domain -> Select Free plan
  3. Change nameservers (at your domain registrar to Cloudflare's ns1/ns2)

You can skip this if you don't have a domain. Use *.workers.dev or *.pages.dev subdomains instead.

Step 3. Enable 2FA (Strongly Recommended)

  1. My Profile -> Authentication
  2. Two-factor Authentication -> Enable
  3. Register with Google Authenticator or Authy

A Cloudflare account hijack affects all services. 2FA is not optional — it's mandatory.

2. Free Tier Limits — From an App Perspective

Workers (Serverless Functions)

Item Free Limit Estimated Usage (DAU 1,000)
Requests 100,000/day DAU 1,000 x 100 req = 100K (near limit)
CPU time per request 10ms LLM calls are wait time (not CPU time), safe
Subrequests 50/invocation Sufficient
Script size 1 MB Sufficient

Free up to DAU 1,000. For DAU 5,000+, switch to Workers Paid ($5/mo).

Workers KV (Key-Value Cache)

Item Free Limit App Usage
Reads 100,000/day 80% cache hit x DAU 1,000 x 100 req = 80K (safe)
Writes 1,000/day New LLM response caching
Deletes 1,000/day Sufficient
List 1,000/day Rarely used
Storage 1 GB Sufficient for text-only cache
Value size limit 25 MB/key LLM response is a few KB, sufficient

With 80% cache hit rate, fully within free limits.

R2 (Object Storage, Azure Blob Replacement)

Item Free Limit App Usage
Storage 10 GB Current Azure Blob usage estimated under 1 GB
Class A operations (writes) 1,000,000/month Sufficient
Class B operations (reads) 10,000,000/month More than sufficient
Egress (bandwidth) Unlimited free Decisive differentiator

Free egress is R2's biggest advantage. Azure Blob egress costs eliminated.

Pages (Static Sites)

Item Free Limit
Bandwidth Unlimited
Builds 500/month
Concurrent builds 1
Custom domains 100
Sites Unlimited

Fully sufficient for frontend hosting. Equivalent to Azure Static Web Apps.

Workers AI (Optional)

Item Free Limit
Neurons 10,000/day (~5,000-10,000 requests)
Models Llama 3.x, DeepSeek, etc.

Unnecessary if using DeepSeek API directly. But useful as backup LLM.

Reset Timing

All free limits reset: Daily at UTC 00:00 (KST 09:00)

3. Recommended Migration Phases

Phase 1. Static Assets Only (Low Risk, 2-3 hours)

Azure Static Web Apps -> Cloudflare Pages
- Deploy frontend (HTML/CSS/JS) to Pages
- Keep LLM calls on Azure Functions
- Only change: update API endpoint URL in frontend
  • Risk: Very low (frontend only)
  • Effect: 50-100ms faster first page load for Korean users

Phase 2. Migrate Blob to R2 (Recommended post-launch)

Azure Blob Storage -> Cloudflare R2
- Auto-migrate via Super Slurper or Sippy
- S3-compatible API minimizes code changes
- Large egress cost savings
  • Risk: Medium (URL structure changes)
  • Effect: Free egress + enhanced edge cache

Phase 3. Move Functions to Workers (Review 14 days after launch)

Azure Functions (Python) -> Cloudflare Workers (JS/TS or Python WASM)
- Rewrite DeepSeek API calls using fetch()
- Redesign 4-tier cache with Workers KV + Cache API
  • Risk: High (different runtime)
  • Effect: Cold start eliminated, further edge latency reduction
  • Recommended timing: After launch + 14 days burn-in data

4. First Setup After Sign-Up — 5 Minutes

Install Wrangler CLI

# Requires Node.js 18+
npm install -g wrangler

# Login (browser auth)
wrangler login

# Verify
wrangler whoami

First Pages Deploy (Static Site Test)

# Git integration
# Dashboard -> Pages -> Connect to Git
# Connect GitHub repo -> auto-build on main branch

# Or direct deploy
wrangler pages deploy ./build --project-name=my-test

After deploy: https://my-test.pages.dev

First R2 Bucket

# Create bucket within free limits
wrangler r2 bucket create my-blob

# Test file upload
wrangler r2 object put my-blob/test.txt --file ./test.txt

5. 4 Key Caveats

Caveat 1. Workers' Python support is limited

Supported: Pyodide-based WASM (limited stdlib)
Not supported: native C dependency packages (some numpy, asyncio)

Check which Python packages your app depends on. Pure Python + fetch can be ported; complex dependencies may require JavaScript/TypeScript rewrite.

Caveat 2. KV consistency is eventual

KV write -> up to 60 seconds before readable from other edge locations
Unsuitable for data requiring immediate consistency (e.g., user sessions)

For strong consistency (user sessions), use Durable Objects (requires Workers Paid $5/mo).

Caveat 3. Daily reset risk on free limits

Workers requests 100K/day = avg 4,166 req/hour
If peak hour (e.g., US market close KST 22:30) hits 10K req/hour
-> 100K limit exhausted by afternoon
-> Service down until next UTC 00:00 (KST 09:00)

Peak hour monitoring is essential. Switch to Workers Paid ($5/mo) when reaching 1,000 users.

Caveat 4. DeepSeek API call latency is the bottleneck

Cloudflare edge: 10-50ms
DeepSeek API: 2,000-8,000ms (LLM inference)
User-perceived latency: 95%+ is LLM call

Cloudflare migration's perceptible effect is concentrated on first-page-load. LLM response time itself won't change. Be honest in migration ROI calculations.

References