ReferenceIncidentsEra 3 · Series ALatency regressionA deploy put one more synchronous call on every request: each answer is 150 ms slower.What happens#A deploy put one more synchronous call on every request: each answer is 150 ms slower.How to prevent it#Percentiles and tracing on the request path, and a canary that watches p99.The postmortem opens the first of these that you do not own yet:PercentilesDistributed tracing