A first-principles walk through Lustre, the parallel file system behind most of the world's supercomputers, and why its internals look so much like a distributed database.

A quick framework for deciding whether a change is reliable, scalable, and maintainable before it is shipped.

A follow-up on SELECT FOR UPDATE — what happens once locking works, and the new problems it quietly introduces.

A lesson in database transactions, race conditions, isolation levels, and why SELECT FOR UPDATE exists.