Hey, author of the post here.

The whole app lives in Firebase. Once a user hits a profile page, a cloud function runs, get the data from the DB, constructs the page, and caches it at the CDN for 24h.

If a user add/delete something on their profile, the cache is removed. Otherwise, the content is served from cache only with no need for a lot of computation.

With this, you can still get your way through Firebase's generous free tier https://firebase.google.com/pricing

Firebase free tier is only 125k invocation per month. How are covering +1M invocations per month on free tier?

Caches hit on the CDN are not trigerring an invocation.

So it's not covering 1M+ invocations per month on the free tier.

1 million hits in a month is only one request every couple seconds. It doesn't take much to host that.

> 1 million hits in a month is only one request every couple seconds.

That's a really naive view of how hits are distributed over time.

Even if rush hour has 10 times more traffic than off hours, that's still only 4 requests per second. Any unoptimized Postgres on a 2/2 node should be able to handle these loads just fine.

That's the average.

Of course, hits are not homogeneously distributed. But an average of less than one hit a second leaves more than enough leeway for any reasonable clustering you may come with. Any small computer can handle thousands of times more than that.

There is some questions about bandwidth costs, that can vary wildly.