How Time Series Databases Work–and Where They Don't
The article explains how time series databases (TSDBs) are optimized for handling timestamped data but struggle with high-cardinality data, complex joins, and event-level analysis, where alternative tools like Honeycomb's columnar store can offer better performance and flexibility.
Background
- Time series databases (TSDBs) are specialized databases optimized for storing and querying timestamped data—metrics like CPU usage, API request latencies, or sensor readings. Popular TSDBs include InfluxDB, Prometheus, TimescaleDB, and Graphite.
- Honeycomb is a "observability" platform (founded 2013 by Charity Majors) that competes with traditional monitoring tools. Its product is built on a columnar, event-based store, not a pure TSDB.
- This blog post argues that TSDBs break down under high-cardinality data—e.g., when a metric has millions of unique tag combinations (like per-user or per-request-id metrics). TSDBs pre-define schemas and pre-aggregate, which works poorly for exploratory debugging of rare or complex events.
- The post is part of a long-running industry debate: traditional time-series monitoring (promoted by TSDB vendors) vs. "observability" (the Honeycomb/Datadog approach), which stores raw, high-dimensional events for ad-hoc querying. The distinction matters for engineers choosing infrastructure for debugging modern distributed systems.