Series
A series stores rows ordered by a required key column - a timestamp or an integer. It is the shape for measurements, audit trails, and historical records: data that arrives in order, is queried by range, and grows without an upper bound.
The key defines the order
Every series declares its ordering column with with { key: column }. The key can be an integer - useful for sequence numbers, block heights, or logical clocks. Reads return the newest entries first. Run the snippets on this page in order:
Time-keyed series and precision
With a datetime key, the series stores timestamps at a declared precision: second, millisecond (the default), microsecond, or nanosecond:
If a write omits the key, the engine assigns the current time - so an append-only event log needs nothing beyond the payload:
Range queries follow the key
Filters on the key column select by the series' native order, which is what time-window and "since X" queries compile to:
History can be corrected
A series is ordered, not immutable. Late corrections - a recalibrated sensor, an amended audit record - are ordinary updates selected by key:
Classifying entries with tags
A series can attach a tag type - a named set of variants, optionally with payload fields - to classify where each entry came from or what kind it is, without widening the schema:
Series or something else?
- --Current state keyed by identity (one row per user, per order): a table.
- --Only the last N entries matter: a ring buffer.
- --Derived history - the output of a pipeline recorded over time: a series-backed view.