ThinkerQAQ

NOTE

1.1 Time-Series Database

Time-series databases, time-series data, data models, and storage implementation.

ObservabilityUpdated 2 min readhistorical

This is a historical learning note and may contain outdated or incomplete understanding.

1. What Is a Time-Series Database

A time-series database is a database optimized for timestamped or time-series data.

Time-series databases are optimized for continuous timestamped writes, compression, aggregation, and time-range queries. Relational databases can also store historical data; the difference is mainly the data model and the optimizations made for time-series workloads.

2. What Is Time-Series Data

Time-series data is a sequence of observations associated with time.

3. Why Use a Time-Series Database

3.1. Recording with a Relational Model

  • If time-series data is stored in an ordinary table:
    • Common challenges at large scale include:
      • high write volume;
      • expensive time-based aggregation;
      • storage cost.

3.2. Time-Series Model

The exact terminology varies by product. In an InfluxDB-style model:

  • metric: the metric name. It can be loosely compared with a table in a relational database.
  • data point: one observation, loosely comparable with a row.
  • timestamp: when the data point was produced.
  • field: a value that commonly changes over time, such as wind direction or speed.
  • tag: indexed metadata that commonly identifies a series, such as sensor or city.

In the example:

  • the metric is Wind, and every data point has a timestamp;
  • fields are direction and speed;
  • tags are sensor and city.

4. Time-Series Database Implementation

4.1. LSM Tree

LSM trees are commonly used in write-heavy storage systems.

4.2. Distributed Storage

When time-series volume exceeds the capacity or throughput of one machine, data can be distributed across multiple nodes. Small deployments do not inherently require distributed storage.

4.2.1. Distribution Algorithm

A distributed TSDB needs a partitioning strategy that determines where series are stored.

4.2.2. Shard Key

A common idea is to partition by a combination such as metric + tags so that points from the same logical series can remain colocated for time-range access. Actual designs vary by database.

5. References