What steps do you take to debug a distributed system failure that only occurs under high load at Tensorwave?
