The book covers reading files and databases, cleaning and transforming, and assembling all of it into a process that runs on its own, with version control, monitoring and deployment. It then moves to streaming with the tools of that moment. The extract covers the first part, which is the part I wanted.
What I took from it
The list of what makes a pipeline production-grade: it can be run again without duplicating anything, it validates what it receives instead of trusting it, it records what it did, and it raises an alarm when it fails. Those four things separate a process that works from one you can also leave alone. In mid-sized companies, almost every integration I find half-satisfies the first and none of the other three.
Rerunning without duplicating is the idea that prevents the most argument. A process that runs twice because of a retry must not create two journal entries, two payments or two invoices. That is solved in the design, with a key that identifies the event, and not with a promise that nobody will press the button twice.
And validation at the boundary: check the data as it arrives, not when it explodes three steps later. A bank file with a shifted column is caught at the door or chased for a week through the accounts.
What I left out
The tools, which is most of the rest: the screenshots and menus of 2020 no longer match anything, and that is the problem with books written around a product. I also leave out the streaming section, which for a company closing its books on a daily cycle solves a problem it does not have. It is an entry-level book, and the extract is enough.