[ryan_lane]

You have to have enough people to "be that asshole" that tells people in their design docs that they are creating a dependency loop and need to design things to avoid it. I did that for quite a while at Lyft, and I'm hoping that people continued doing that after I stopped. My guess, though, is that I and others have missed things and that even somewhere that was careful to avoid it, there's still some places that could make cold boots difficult.

The best way to avoid this problem is to have a secondary datacenter that is hot, and to never let both fail. Of course, this is what twitter is doing. Designing a DR plan where you can lose multiple primary/secondary datacenters is hard enough that basically no one does it.

This is something that really doesn't belong in the whistleblower complaints because it wasn't under his responsibilities, twitter wasn't lying to anyone about having properly implemented DR plans, and twitter's DR plan isn't out of the ordinary across the industry.