Comment by jwatte

Let's say a typical base service (network attached RAM or whatever) has 99.99% reliability. If you have a dependency on 100 of those, you're suddenly closer to 99% reliability. So you switch to higher-level dependencies, and only have 10 dependencies, for a 99.9% reliability. But! It turns out, those dependencies each have dependencies, so they're really already more like 99.9% at best, and you're back at 99% reliability.

"good enough" is, indeed, just good enough to make it not worthwhile to rip out all the upstreams and roll your own everything from scratch, because the cost of the occasional outages is much lower than the cost of reinventing every single wheel, nut, bolt, axle, bearing, and grease formulation.