Blog platform - write once, cache everywhere

Why this design

A blog platform (think Medium, Substack, Ghost, a company engineering blog) is a classic first system design because the ratio is so extreme: tens of thousands of posts written a day, tens of millions of page views. That ratio says the job is serving the same HTML to many people cheaply, which is caching. And caching moves the hard part to the one moment something changes: publishing has to find and invalidate every page that shows the post.

What you will learn

• Why 2,500 reads per write makes the CDN, not the database, the serving tier.
• How a 95 percent cache hit ratio turns 3,000 page views a second at peak into 150 requests at the origin.
• Why publishing is an invalidation problem: one post changes five or more URLs.
• Why images go to object storage, and why they dominate storage by 100x.
• Where a single primary database is fine for years, and why.

How to read this board

1. Top: PUBLISH. Author, editor, the posts API hex, and the publish event that starts the fan-out.
2. Right: STORES. The posts database (one primary), a read replica, and the media bucket.
3. Bottom: READ. Reader, CDN edge, the render service that builds HTML on a miss.
4. Bottom right: AFTER PUBLISH. The red purge worker and the search indexer.
5. Press play, then read pages 1 to 6, then the tables. The conversation is the red box: who decides which pages are stale?

Functional

• Authors write drafts in a rich editor, with autosave, and upload images.
• Publish a post: it gets a permanent URL /@author/slug and appears on the home page, the author page, its tag pages and the RSS feed.
• Edit a published post; readers see the change within a minute.
• Readers read posts without logging in. Search posts by title and text.
• Unpublish or delete a post; it disappears everywhere within a minute.

Non-functional

• Time to first byte under 100 ms for a cached page anywhere in the world; under 500 ms on a miss.
• Read availability 99.95 percent - pages keep serving from the CDN even if the origin is down.
• Write availability 99.9 percent; autosave must never lose more than a few seconds of typing.
• Survive one post going viral (front page of a news aggregator) without the origin noticing.
• Search may lag publishing by a minute.

Out of scope

• Comments, likes and paywalls (each personalises a page and needs its own design).
• Email newsletters to subscribers.
• Custom domains per blog (a CDN configuration problem, not a data one).