December 2016 Summaries

Pre-aggregated metrics, while offering quick and cost-effective storage solutions, impose significant limitations on data exploration and problem-solving due to their reliance on a fixed set of well-known metrics and a storage strategy that involves breaking down events into discrete counters. This approach reduces data granularity and complicates the differentiation between signal and noise. The addition of new attributes or high-cardinality attributes, such as user IDs or OS versions, can exponentially increase the number of unique metrics and the storage required, leading to difficult choices between managing storage costs and losing valuable data insights. For example, identifying problematic queries at Parse would have been challenging with pre-aggregated metrics due to the need to prioritize certain users over others. Honeycomb offers a solution that allows more flexible data segmentation and the ability to track detailed breakdowns, enabling users to better understand and respond to diverse traffic patterns without being constrained by their tools.

Pre-aggregated metrics, although efficient and easy to understand, often fall short in addressing new and unpredictable questions that arise as systems grow more complex. Initially, using default metrics, like those provided by a Cassandra cluster, may seem adequate, but they can limit insights, as they rely on predefined data, potentially obscuring root causes of issues. This approach necessitates constant updates to metrics, which can become overwhelming. Honeycomb offers a solution by preserving data for query-time aggregation, allowing for flexible analysis by grouping or filtering data based on various attributes, thus providing deeper insights and addressing the limitations of pre-aggregated metrics.