The metrics product we built worked — But we killed it and started over anyway
Blog post from Sentry
Sentry initially developed a metrics product that pre-aggregated time-series data, but during internal testing, the limitations of this approach became evident, especially regarding cardinality and the ability to provide actionable context for debugging. This led to the decision to abandon the original product just two weeks before its launch and instead develop a new system centered around trace-connected telemetry using raw-event storage and on-demand aggregation, facilitated by the Event Analytics Platform built on ClickHouse. This shift solved previous issues by allowing dynamic slicing of metrics and linking them to traces without prohibitive costs, ultimately enhancing the debugging process by providing richer, more connected context. This reimagined approach aligns with Sentry's commitment to offer robust application-level insights, and it also supports their AI agent, Seer, in offering effective debugging solutions. Despite the challenges and the decision to halt the initial product, the new metrics system is anticipated to better serve developers by integrating seamlessly with other telemetry data, offering a more insightful and efficient debugging experience.