Time-Series Databases (TimescaleDB, InfluxDB, Prometheus)

Time-series data consists of continuous sequences of measurements associated with explicit timestamps (e.g., CPU utilization, financial stock ticks, IoT temperature sensors, GPS coordinates).

When attempting to store millions of datapoints per second in standard PostgreSQL or MySQL B-Tree tables:

1. Index Thrashing & Random I/O: Every incoming insert modifies multiple secondary B-Tree indexes. Once the index size exceeds available RAM buffer cache, every write turns into random disk I/O, causing write throughput to collapse.
2. Vacuum & Deletion Overhead: Deleting expired historical data (DELETE FROM metrics WHERE created_at < NOW() - INTERVAL '30 days') causes massive table bloat, locks rows, and triggers expensive database vacuum processes.
3. Storage Inefficiency: Storing uncompressed floating-point timestamps and metrics in standard row format consumes terabytes of expensive disk space.

Time-Series Databases (TSDBs) are engineered around a fundamental invariant: Data is strictly append-only, ordered by time, and rarely if ever updated after insertion.