The Internet Is CloudFlare | ✉️ #83

Illustration for MKDEV Dispatch #83. Features a smiling person holding a cat with paper plane sketches in the background. Text reads, "The Internet is Cloudflare." The image background is a mix of orange and gray, complementing the graphic design.

Last updated: 13/12/25 | Published: 13/12/25

Hey! 👋

We’ve learned recently, that The Internet is not only us-east-1, but it’s also CloudFlare. Actually, even more so, because even the companies that proudly left the cloud still ended up being unavailable to the customers during the recent CloudFlare outage.

I get the argument of moving a regular 3-tier applications from the cloud to another compute provider that doesn’t call themselves cloud. Replace ALB with Nginx, EC2 with KVM or just bare metal machines, and run your PostgreSQL yourself - the open source software we have today is incredible and you really won’t sacrifice much. You can automate this with tinkerbell and alternatives, maybe with some configuration management tools on top very nicely - practically as nice as you would duck tape AWS services with Terraform.

That would handle a single location pretty well. It will also handle multiple locations/DCs very well - we’ve done it for many of our customers who just can’t go to the cloud for various reason.

And then comes the CDN. CloudFlare boasts to cover 330 cities in 125+ countries, with 13,000 networks connecting directly to their system. CloudFront, which sits in front of mkdev.me, has 700+ Points of Presence (PoPs). Most of the static requests to mkdev.me never even reach AWS Frankfurt region - the content, including HTML, is served directly from the closest edge location. Regardless if you access our website from Singapore, Los Angeles or Munich, latency is well below 100ms, in most cases it’s under 50ms.

Configuring CDN like this is pretty trivial, and the cost of CDN providers is pretty low - for smaller businesses, it’s even free. The cost of running your own globally distributed network of edge locations is… more?

So I get why every second website out there puts CloudFlare in front of it, including my primary email provider hey.com and our primary project management software, basecamp.com - both are famously cloud-free, both entirely inaccessible when CloudFlare is down. The Internet depends on certain services like CloudFlare not because we were forced to, but because of the convenience and price. They just offer a great service.

CloudFlare is also unique in how it transitions from being a CDN to being a fully-featured cloud - with the core difference to hyperscalers being that most of the features of this cloud run on edge locations, as close to your customers as possible. You don’t need to duck tape dozens of regions to make your application run as close to your customers as possible - this behaviour is just the nature of how CloudFlare and similar providers work.

Do we have a none-cloud, self-hosted option today? Is it worth building one? And as the blast radius of outages of existing providers seems to get bigger, are we going to see alternatives emerging? As much as I love CloudFlare and the cloud, we can’t stop the world every time there is a faulty deployment happening inside a handful of companies.

What We've Shared

DevOps Accents, episode 66: Is Kubernetes an Engineering Choice or a Must. In this episode of DevOps Accents, Kirill previews for Leo his speech for DevOps Pro Europe 2026. He examines the rise of Kubernetes as a near-requirement in tech—for both organizations and professionals—and why its dominance has surprisingly little to do with scaling, microservices, or capacity concerns.
Bits & Pretzels 2025 Impressions: In this extended clip from episode 65 of DevOps Accents, Kirill shares his unfiltered impressions from Bits & Pretzels 2025 in Munich — from deflated AI hype and actually useful startups to what really matters at conferences: the networking, the random conversations, and the weird security incident that partly shut down Oktoberfest.
Explaining AI Explainability: The Current Reality for Businesses. In the second article of his explainable AI series, Paul Larsen looks at what today’s XAI tools really deliver for different stakeholders—from users to regulators—and where they still fall short for trust, liability and high-risk decisions.

What We've Discovered

Collaboration sucks: Adopting practices from this post might be one of the biggest productivity "hacks" you can do in the short term.
Improve API discoverability with the new Amazon API Gateway Portal: This is going to replace a lot of home-grown swagger-to-html solutions - with a better security out of the box too.
Inside Husky’s query engine: Real-time access to 100 trillion events. New deep dive into DataDog's Husky - the storage system powering most of their features, including logs.
Zero downtime database migrations: Lessons from moving a live production database. From building a robust migration framework down to particular tools and configurations, this article about Tines migrating a database from one AWS region to another covers it all.
Stop Reactive Network Troubleshooting: Monitor These 5 Metrics to Prevent Downtime. A good list that is worth translation to your alarms configuration.

The 84th mkdev dispatch will arrive on Friday, December 26th. See you next time!