Flux data model

Flux employs a basic data model built from basic data types. The data model consists of tables, records, columns and streams.

Record

A record is a tuple of named values and is represented using a record type.

Column

A column has a label and a data type. The available data types for a column are:

Table

A table is set of records with a common set of columns and a group key.

The group key is a list of columns. A table’s group key denotes which subset of the entire dataset is assigned to the table. All records within a table will have the same values for each column that is part of the group key. These common values are referred to as the “group key value” and can be represented as a set of key value pairs.

A tables schema consists of its group key and its columns’ labels and types.

Stream of tables

A stream represents a potentially unbounded set of tables. A stream is grouped into individual tables using their respective group keys. Tables within a stream each have a unique group key value.

A stream is represented using the stream type stream[A] where A: Record. The group key is not explicitly modeled in the Flux type system.

Missing values (null)

null is a predeclared identifier representing a missing or unknown value. null is the only value comprising the null type. Any non-boolean operator that operates on basic types returns null when at least one of its operands is null.

Think of null as an unknown value. The following table explains how null values behave in expressions:

Operating on something unknown produces something that is still unknown. The only place where this is not the case is in boolean logic. Because boolean types are nullable, Flux implements ternary logic as a way of handling boolean operators with null operands. By interpreting a null operand as an unknown value, we have the following definitions:

• not null = null
• null or false = null
• null or true = true
• null or null = null
• null and false = false
• null and true = null
• null and null = null

Because records are represented using record types, attempting to access a column whose value is unknown or missing from a record will also return null.

Transformations

Transformations define a change to a stream. Transformations may consume an input stream and always produce a new output stream. The output stream group keys have a stable output order based on the input stream. Specific ordering may change between releases, but is not considered a breaking change.

Most transformations output one table for every table they receive from the input stream. Transformations that modify group keys or values regroup the tables in the output stream. A transformation produces side effects when constructed from a function that produces side effects.

Transformations are represented using function types.

Match parameter names

Some transformations (for example, map and filter) are represented using higher-order functions (functions that accept other functions). Each argument passed into a function must match the parameter name defined for the function.

For example, filter accepts fn, which takes one argument named r:

from(bucket: "db")
   |> filter(fn: (r) => ...)

If a parameter is renamed from r to v, the script fails:

from(bucket: "db")
    |> filter(fn: (v) => ...)
 // FAILS!: 'v' != 'r'.

Because Flux does not support positional arguments, parameter names matter. The transformation (in this case, filter) must know r is the parameter name to successfully invoke the function.