Time-Series Data

Time-series data is a sequence of timestamped key-value pairs published by a device or computed by the platform. It represents continuously changing measurements — sensor readings, counters, state changes — where every data point matters, not just the latest.

Compared to attributes:

Each time-series entry consists of:

• Key — the measurement name (e.g., temperature)
• Value — string, boolean, integer, double, or JSON
• Timestamp — Unix epoch in milliseconds; auto-assigned by the server if not provided by the device

Simple payload:

{ "temperature": 26.5, "humidity": 78, "pressure": 1013.2 }

With explicit timestamp:

{ "ts": 1745000000000, "values": { "temperature": 26.5, "humidity": 78 } }

Devices publish time-series data over MQTT, HTTP, CoAP, LwM2M, and other protocols. Each protocol may use different payload format that is mapped to the same internal JSON format; only the transport mechanism differs. See the Connectivity guide for protocol-specific details.

Many protocols support multiple readings with different timestamps in a single message — useful for batching offline data or high-frequency sensors:

[
  { "ts": 1745000000000, "values": { "temperature": 25.1 } },
  { "ts": 1745000060000, "values": { "temperature": 25.4 } },
  { "ts": 1745000120000, "values": { "temperature": 26.0 } }
]

ThingsBoard stores time-series data in PostgreSQL by default. For high-volume deployments, Cassandra is supported as an alternative backend optimized for write-heavy workloads and long retention periods. A separate latest values table is maintained for fast access to the current readings.

To configure automatic data expiration (TTL), see Configure Time-Series Data Retention.