Tables
Tables hold authoritative, mutable state - the facts your application asserts directly. They are the primary storage shape: rows are inserted, updated, and deleted transactionally, and everything derived (views, subscriptions) ultimately sources from them or from the other storage shapes.
Creating a table
A table is a named set of typed columns inside a namespace. Column types include integers (int1 through int16, uint1 through uint16), floats (float4, float8), bool, strings (utf8, with text as an alias), blob, and temporal types (date, time, datetime, duration). See Data Types for the full list. Run the snippets on this page in order:
Reads return the newest rows first by default; add sort for an explicit order.
Optional columns
Columns are non-nullable unless declared Option(type). An optional column that was never set holds none - ReifyDB's explicit absent value, which queries can test for and which renders distinctly in results. See Working with none.
Auto-increment columns
An integer column declared with { auto_increment } is assigned the next value of a per-column sequence on insert, so writers never need to coordinate IDs. The sequence itself can be inspected and repositioned - see Sequences.
Primary keys
A primary key is declared as a separate statement after the table exists, and can span multiple columns: create primary key on ns::table { col1, col2 }.
Updating and deleting
update sets the listed fields and leaves every other column untouched; delete removes whatever matches the filter. Both take a filter to select rows, and both accept returning to hand back the affected rows in the same statement - useful when the write itself computes something you need, like a generated ID or the post-update value:
Without returning, mutations report what happened as a count:
System columns
Every stored row carries engine-maintained columns prefixed with #. They are not returned by default; project them explicitly when you need them. #rownum is the row's stable number within its table:
Evolving the schema
alter table adds and drops columns in place. Add new columns as Option(type) when existing rows have no value for them:
When a table is not the right shape
- --Bounded recent history with automatic eviction: use a ring buffer
- --Time-ordered measurements and audit records: use a series
- --Repeated low-cardinality strings: intern them with a dictionary
- --State computed from other state: never write it by hand - derive it with a view