"The Mess We're In" by Joe Armstrong

Caption notes

In this 2014 Strange Loop talk, Joe Armstrong starts with a string of tooling failures while preparing slides. He uses state space to explain why two seemingly identical failures can occur on machines in different states, making an online fix hard to reproduce. He also points out that the reliability benefit of replication depends on whether replica failures are independent. Later he proposes using content fingerprints to identify identical files and changes in compressed size to look for similar material. He notes that the heuristic has limits as data grows or when code is renamed. He closes by advocating systems composed of small units that can be validated.