Most comments here seem to be about "what Shadow should have done", making it obvious, given the distribution of HN users, that Shadow really should have used a

Most comments here seem to be about "what Shadow should have done", making it obvious, given the distribution of HN users, that Shadow really should have used a different systems approach. But I'd like to focus on a different aspect: they used the systems approach that they did (given budget and deadline), and the outcome was poor. Is it difficult, using their approach, to accomplish what they accomplished? Seems like a single function call is the likely culprit. What exactly caused this... A race condition, database sharding? What's the cause of this issue, at a technical level?