Time-Series Database
Subject: Computer and Technology noun /ˈtaɪm ˌsɪr.iːz ˈdeɪ.təˌbeɪs/ 1 definition
- database-systems
- storage-engines
- data-management
- system-monitoring
-
Definition (noun)
Entry #158 · UpdatedA Time-Series Database (TSDB) is a specialized database system purpose-built to ingest, compress, store, and query chronological sequences of measurements indexed by timestamp. Designed for append-heavy workloads such as infrastructure metrics, IoT sensor readings, application telemetry, and financial market ticks, it partitions data into time-bounded blocks and applies domain-specific encodings—such as delta-of-delta timestamps and Gorilla XOR compression—to sustain high write throughput while minimizing storage footprints. Unlike general-purpose relational or document databases optimized for normalized schemas, arbitrary cross-entity joins, and frequent in-place row updates, a TSDB prioritizes sequential appends, fast time-window aggregations, and automated data lifecycle policies such as downsampling and retention expiration.
Common Types
Purpose-Built Standalone TSDBs: Custom storage engines built from the ground up around time-partitioned segments, inverted tag indexes, and columnar compression to deliver high metric ingestion rates and rapid range scans (such as InfluxDB, Prometheus, and VictoriaMetrics).
Relational TSDB Extensions: Add-on engines that layer automatic time-based partitioning, specialized compression, and continuous aggregations onto an existing relational database, preserving full SQL queries and joins against relational metadata (such as TimescaleDB on PostgreSQL).
Distributed Wide-Column TSDBs: Scalable architectures layered atop distributed key-value or wide-column storage backends that shard series across cluster nodes using composite metric-and-time-bucket keys (such as OpenTSDB on Apache HBase).
Round-Robin and Fixed-Window Stores: Compact storage formats that write incoming samples into fixed-size circular buffers and automatically downsample older intervals into coarser resolutions, guaranteeing bounded disk usage over indefinite timeframes (such as RRDtool).
Example
A platform engineering team configures Prometheus to scrape CPU, memory, and request-latency metrics every 15 seconds across 1,200 Kubernetes pods, organizing timestamped samples into time-partitioned compressed storage blocks indexed by
namespaceandservicelabels. When an alert fires, an on-call engineer queries the database with PromQL to compute the 99th-percentile latency across 5-minute rolling windows over the past six hours while older raw blocks automatically age out under a 30-day retention policy.