Speaker
Description
Failure is not an exception in large‑scale data systems. It is an expected and recurring condition. Disks fail, nodes crash, networks partition, and systems must recover correctly without losing or corrupting data. In scientific computing environments, recovery is not just about uptime, but about preserving correctness, reproducibility, and trust in results.
This lecture focuses on how database recovery works in practice, using write‑ahead logging (WAL) as central concepts. We explain how WAL enables atomicity and durability, how it is generated and stored, and how databases distinguish between committed and uncommitted changes after a crash. The session walks through typical recovery scenarios, illustrating how logs are replayed. We will also cover how to run an on-line backup of a constantly changing database.