Skip to main content

🗓️ 12072025 1023

TIME SERIES DATABASE
  • Optimized for appending: TSDBs assume write-once data; use delta encoding and compression for efficient storage
  • Multi-dimensional model: Data = metric name + timestamp + labels (e.g., region="ap-sg", instance="app1")

Labels allow slicing and dicing.

  • Cardinality: Total number of unique time-series (label combinations). High cardinality increases memory, storage, and query latency
  • Scaling limits: Single Prometheus typically handles ~1–2 million series. Exceeding this requires strategies like sharding, federation, or adopting specialized TSDBs

References​