Time-Series Database (TSDB)

Time-Series Databases (TSDB) sont des SGBD spécialisés pour stocker et requêter efficacement des données indexées par temps — metrics infrastructure, IoT sensor data, financial market data, application telemetry. Optimisées pour ingestion massive (millions points/sec), compression élevée, et requêtes range temporal.

Caractéristiques uniques TSDB vs RDBMS classique :
(1) **Append-only workload** — ingestion massive, updates rares.
(2) **Time as primary index** — natif partitioning par time chunks.
(3) **Columnar storage** typique — meilleur compression sur similar values temporal series.
(4) **Downsampling automatique** — retention policies réduisant résolution older data (1s recent → 1min after 1 day → 5min after 7 days → 1h after 30 days).
(5) **Optimized functions** — derivatives, moving averages, rate, percentiles built-in.
(6) **Aggregations temporal** — group by time bucket.
(7) **High cardinality handling** challenge (labels/tags explosion).

Leaders 2024 :
(1) **InfluxDB** — TSDB open source historique, v3 written Rust + DataFusion (massive perf improvement). Cloud Serverless option.
(2) **TimescaleDB** — Postgres extension, hypertables auto-partitioning. Full SQL standard + spatial + JSON. Standard PostgreSQL ecosystem benefit.
(3) **Prometheus** — pull-based metrics, used Kubernetes monitoring standard, PromQL language.
(4) **VictoriaMetrics** — Prometheus-compatible, plus performant, single binary deploy.
(5) **Grafana Mimir, Thanos, Cortex** — long-term Prometheus storage scaling.
(6) **QuestDB** — high-performance, SQL.
(7) **TDengine** — IoT-focused.
(8) **ClickHouse** — columnar OLAP, also great for time-series.
(9) **Apache Pinot** — real-time analytics.
(10) **AWS Timestream, Azure Data Explorer (Kusto/ADX), GCP BigQuery** — cloud managed.

Use cases :
(1) **Infrastructure monitoring** — Prometheus pour Kubernetes metrics, Datadog ingestion.
(2) **Application performance monitoring** (APM) — traces, latency histograms.
(3) **IoT** — millions de sensors envoyant readings continus.
(4) **Financial markets** — tick data, OHLC candlesticks.
(5) **DevOps observability** — logs aggregations, business metrics.
(6) **Energy/utilities** — smart meters, grid monitoring.
(7) **Web analytics** — page views, conversions over time.

Query languages : PromQL (Prometheus), Flux/InfluxQL (InfluxDB), SQL extended (TimescaleDB, QuestDB), KQL (ADX). Compétences DEA-C01, DP-203, DP-300.