Embed Notice

But actually shoveling data over the network to make new replicas requires making actual copies of the data.

So many options to choose from, but no matter what you do and how well it scales:

Databases cannot be scaled with reactive autoscaling.

We need automated load testing, load prediction and predictive autoscaling, so that the capacity is there and the caches are prewarmed when the load arrives.

This will take another 5 years to automate. For the moment, we test manually in prod.