Checkout the slides

This talk will be a deep dive into the new incremental backup feature which has been committed for (hopefully) inclusion in PostgreSQL 17.

First, I'll explain how the feature works from a user perspective, and briefly discuss some possible use cases.

Second, I'll discuss the architecture of the feature, both how it actually works and alternative designs which I considered and ultimately rejected.

Third, I'll discuss some consequences of the design, both in terms of some interesting corner cases that are allowed and also, importantly, how we're able to reject cases where the user has done something that is not valid.

Finally, I'll discuss possible future improvements -- things that don't work now, but potentially could with more work.