Not sure about webapps but in analytics, this is a common practice. Real data has too many exceptions and edge cases. If you dont develop and test with real data, you will end up effectively developing or testing on production.

I have always made clones of production data on my dev machine for testing -- it has the added benefit of also regularly testing my DB backup restoration process. How else do you test and optimize SQL queries that are only slow with production-size data?

If the data is too big to fit on my machine, I might clone to a nearby colocated server. Testing your DB backup and restoration mechanism becomes even MORE important if you have huge amounts of data.

You can clone the "depersonalized" data for your developer teammates

in a team: 10 developers working on 32 issues: --> need at least 32 separated environments.

I generate "backtests" of a stock trading model, IE I see how it would have done historically. Each backtest reads lots of vendor data from Postgres, and writes some output to Postgres. I can't run a backtest on a "dev" version of the vendor data, I need all of it to make my results correct. The vendor data is too big to easily have two copies.

To make this more complex, it's not a simple process where I read vendor data then generate analysis. My model is generated in several steps, which involves reading vendor data, generating output, then combining the output I generated in a previous step with more vendor data to generate the next step of output.

Backtests are "prod-like" in that they must use real vendor data and must generate correct results that drive business decisions. But they are "dev-like" since I don't want my backtest to interfere with my production system, which generates the version of the model that I currently trade. For a backtest I might want to make a new database table, or change the behavior of an existing process that generates data.

I've tried 2 solutions to this: One is have classes to handle all data reads and writes, and configuration that causes functions to read or write from a dev or production DB as needed. This is a pain to set up, but works well.

The other solution is DB clones. One advantage of DB clones is that it lets me write SQL that joins data I've generated on vendor data. I don't love having business logic in SQL for the obvious reasons, but it can be very performant and easy to maintain. Using classes for data access means that I can't easily do a SQL join between vendor data (which is always on the production DB) and data I generate (which might be on the dev DB.)

I think it will work well for verifying DB changes and optimizing SQL on real data. This seems like a great tool to eliminate the risk of database-related downtime.