Quantiles at Scale: Choosing the Right Estimation Algorithms for Observability - Mike Shi

This talk dives deep into quantile algorithms (p50, p90, p99) used in observability systems, comparing approaches like t-digest and sketches at extreme scale (hundreds of petabytes). Mike Shi shares practical lessons from ClickHouse operations, explaining when to use approximate algorithms vs exact computation and the critical design decision between query-time computation and pre-aggregation. Ideal for engineers building or optimizing high-scale observability platforms.