It's mostly a rather dense engineering textbook, but it contains lots of things I found insightful. I most particularly remember a segment like "We make things more reliable by adding more layers of Swiss cheese, on the assumption that failure modes are uncorrelated and it's only when all the failures take place that the system breaks. But this doesn't work when the system is being attacked by an intelligence, because an intelligence will explicitly correlate failures."

The book is very much designed for Google-scale systems, though: everything is assumed to be microservices, for example.