JSON models

Calcite models can be represented as JSON files. This page describes the structure of those files.

Models can also be built programmatically using the Schema SPI.

Elements

Root

{
  version: '1.0',
  defaultSchema: 'mongo',
  schemas: [ Schema... ]
}

version (required string) must have value 1.0.

defaultSchema (optional string). If specified, it is the name (case-sensitive) of a schema defined in this model, and will become the default schema for connections to Calcite that use this model.

schemas (optional list of Schema elements).

Schema

Occurs within root.schemas.

{
  name: 'foodmart',
  path: ['lib'],
  cache: true,
  materializations: [ Materialization... ]
}

name (required string) is the name of the schema.

type (optional string, default map) indicates sub-type. Values are:

path (optional list) is the SQL path that is used to resolve functions used in this schema. If specified it must be a list, and each element of the list must be either a string or a list of strings. For example,

  path: [ ['usr', 'lib'], 'lib' ]

declares a path with two elements: the schema ‘/usr/lib’ and the schema ‘/lib’. Most schemas are at the top level, and for these you can use a string.

materializations (optional list of Materialization) defines the tables in this schema that are materializations of queries.

cache (optional boolean, default true) tells Calcite whether to cache metadata (tables, functions and sub-schemas) generated by this schema.

  • If false, Calcite will go back to the schema each time it needs metadata, for example, each time it needs a list of tables in order to validate a query against the schema.

  • If true, Calcite will cache the metadata the first time it reads it. This can lead to better performance, especially if name-matching is case-insensitive.

However, it also leads to the problem of cache staleness. A particular schema implementation can override the Schema.contentsHaveChangedSince method to tell Calcite when it should consider its cache to be out of date.

Tables, functions and sub-schemas explicitly created in a schema are not affected by this caching mechanism. They always appear in the schema immediately, and are never flushed.

Map Schema

Like base class Schema, occurs within root.schemas.

{
  name: 'foodmart',
  type: 'map',
  tables: [ Table... ],
  functions: [ Function... ]
}

name, type, path, cache, materializations inherited from Schema.

tables (optional list of Table elements) defines the tables in this schema.

functions (optional list of Function elements) defines the functions in this schema.

Custom Schema

Like base class Schema, occurs within root.schemas.

{
  name: 'mongo',
  type: 'custom',
  factory: 'org.apache.calcite.adapter.mongodb.MongoSchemaFactory',
  operand: {
    host: 'localhost',
    database: 'test'
  }
}

name, type, path, cache, materializations inherited from Schema.

factory (required string) is the name of the factory class for this schema. Must implement interface org.apache.calcite.schema.SchemaFactory and have a public default constructor.

operand (optional map) contains attributes to be passed to the factory.

JDBC Schema

Like base class Schema, occurs within root.schemas.

{
  name: 'foodmart',
  type: 'jdbc',
  jdbcDriver: TODO,
  jdbcUrl: TODO,
  jdbcUser: TODO,
  jdbcPassword: TODO,
  jdbcCatalog: TODO,
  jdbcSchema: TODO
}

name, type, path, cache, materializations inherited from Schema.

jdbcDriver (optional string) is the name of the JDBC driver class. If not specified, uses whichever class the JDBC DriverManager chooses.

jdbcUrl (optional string) is the JDBC connect string, for example “jdbc:mysql://localhost/foodmart”.

jdbcUser (optional string) is the JDBC user name.

jdbcPassword (optional string) is the JDBC password.

jdbcCatalog (optional string) is the name of the initial catalog in the JDBC data source.

jdbcSchema (optional string) is the name of the initial schema in the JDBC data source.

Materialization

Occurs within root.schemas.materializations.

{
  view: 'V',
  table: 'T',
  sql: 'select deptno, count(*) as c, sum(sal) as s from emp group by deptno'
}

view (optional string) is the name of the view; null means that the table already exists and is populated with the correct data.

table (required string) is the name of the table that materializes the data in the query. If view is not null, the table might not exist, and if it does not, Calcite will create and populate an in-memory table.

sql (optional string, or list of strings that will be concatenated as a multi-line string) is the SQL definition of the materialization.

Table

Occurs within root.schemas.tables.

{
  name: 'sales_fact',
  columns: [ Column... ]
}

name (required string) is the name of this table. Must be unique within the schema.

type (optional string, default custom) indicates sub-type. Values are:

columns (list of Column elements, required for some kinds of table, optional for others such as View)

View

Like base class Table, occurs within root.schemas.tables.

{
  name: 'female_emps',
  type: 'view',
  sql: "select * from emps where gender = 'F'",
  modifiable: true
}

name, type, columns inherited from Table.

sql (required string, or list of strings that will be concatenated as a multi-line string) is the SQL definition of the view.

path (optional list) is the SQL path to resolve the query. If not specified, defaults to the current schema.

modifiable (optional boolean) is whether the view is modifiable. If null or not specified, Calcite deduces whether the view is modifiable.

A view is modifiable if contains only SELECT, FROM, WHERE (no JOIN, aggregation or sub-queries) and every column:

  • is specified once in the SELECT clause; or
  • occurs in the WHERE clause with a column = literal predicate; or
  • is nullable.

The second clause allows Calcite to automatically provide the correct value for hidden columns. It is useful in multi-tenant environments, where the tenantId column is hidden, mandatory (NOT NULL), and has a constant value for a particular view.

Errors regarding modifiable views:

  • If a view is marked modifiable: true and is not modifiable, Calcite throws an error while reading the schema.
  • If you submit an INSERT, UPDATE or UPSERT command to a non-modifiable view, Calcite throws an error when validating the statement.
  • If a DML statement creates a row that would not appear in the view (for example, a row in female_emps, above, with gender = 'M'), Calcite throws an error when executing the statement.

Custom Table

Like base class Table, occurs within root.schemas.tables.

{
  name: 'female_emps',
  type: 'custom',
  factory: 'TODO',
  operand: {
    todo: 'TODO'
  }
}

name, type, columns inherited from Table.

factory (required string) is the name of the factory class for this table. Must implement interface org.apache.calcite.schema.TableFactory and have a public default constructor.

operand (optional map) contains attributes to be passed to the factory.

Stream

Information about whether a table allows streaming.

Occurs within root.schemas.tables.stream.

{
  stream: true,
  history: false
}

stream (optional; default true) is whether the table allows streaming.

history (optional; default false) is whether the history of the stream is available.

Column

Occurs within root.schemas.tables.columns.

{
  name: 'empno'
}

name (required string) is the name of this column.

Function

Occurs within root.schemas.functions.

{
  name: 'MY_PLUS',
  className: 'com.example.functions.MyPlusFunction',
  methodName: 'apply',
  path: []
}

name (required string) is the name of this function.

className (required string) is the name of the class that implements this function.

methodName (optional string) is the name of the method that implements this function.

If methodName is specified, the method must exist (case-sensitive) and Calcite will create a scalar function. The method may be static or non-static, but if non-static, the class must have a public constructor with no parameters.

If methodName is “*”, Calcite creates a function for every method in the class.

If methodName is not specified, Calcite looks for a method called “eval”, and if found, creates a a table macro or scalar function. It also looks for methods “init”, “add”, “merge”, “result”, and if found, creates an aggregate function.

path (optional list of string) is the path for resolving this function.

Lattice

Occurs within root.schemas.lattices.

{
  name: 'star',
  sql: [
    'select 1 from "foodmart"."sales_fact_1997" as "s"',
    'join "foodmart"."product" as "p" using ("product_id")',
    'join "foodmart"."time_by_day" as "t" using ("time_id")',
    'join "foodmart"."product_class" as "pc" on "p"."product_class_id" = "pc"."product_class_id"'
  ],
  auto: false,
  algorithm: true,
  algorithmMaxMillis: 10000,
  rowCountEstimate: 86837,
  defaultMeasures: [ {
    agg: 'count'
  } ],
  tiles: [ {
    dimensions: [ 'the_year', ['t', 'quarter'] ],
    measures: [ {
      agg: 'sum',
      args: 'unit_sales'
    }, {
      agg: 'sum',
      args: 'store_sales'
    }, {
      agg: 'count'
    } ]
  } ]
}

name (required string) is the name of this lattice.

sql (required string, or list of strings that will be concatenated as a multi-line string) is the SQL statement that defines the fact table, dimension tables, and join paths for this lattice.

auto (optional boolean, default true) is whether to materialize tiles on need as queries are executed.

algorithm (optional boolean, default false) is whether to use an optimization algorithm to suggest and populate an initial set of tiles.

algorithmMaxMillis (optional long, default -1, meaning no limit) is the maximum number of milliseconds for which to run the algorithm. After this point, takes the best result the algorithm has come up with so far.

rowCountEstimate (optional double, default 1000.0) estimated number of rows in the lattice

tiles (optional list of Tile elements) is a list of materialized aggregates to create up front.

defaultMeasures (optional list of Measure elements) is a list of measures that a tile should have by default. Any tile defined in tiles can still define its own measures, including measures not on this list. If not specified, the default list of measures is just ‘count(*)’:

[ { name: 'count' } ]

statisticProvider (optional name of a class that implements org.apache.calcite.materialize.LatticeStatisticProvider) provides estimates of the number of distinct values in each column.

You can use a class name, or a class plus a static field. Example:

  "statisticProvider": "org.apache.calcite.materialize.Lattices#CACHING_SQL_STATISTIC_PROVIDER"

If not set, Calcite will generate and execute a SQL query to find the real value, and cache the results.

See also: Lattices.

Tile

Occurs within root.schemas.lattices.tiles.

{
  dimensions: [ 'the_year', ['t', 'quarter'] ],
  measures: [ {
    agg: 'sum',
    args: 'unit_sales'
  }, {
    agg: 'sum',
    args: 'store_sales'
  }, {
    agg: 'count'
  } ]
}

dimensions (list of strings or string lists, required, but may be empty) defines the dimensionality of this tile. Each dimension is a column from the lattice, like a GROUP BY clause. Each element can be either a string (the unique label of the column within the lattice) or a string list (a pair consisting of a table alias and a column name).

measures (optional list of Measure elements) is a list of aggregate functions applied to arguments. If not specified, uses the lattice’s default measure list.

Measure

Occurs within root.schemas.lattices.defaultMeasures and root.schemas.lattices.tiles.measures.

{
  agg: 'sum',
  args: [ 'unit_sales' ]
}

agg is the name of an aggregate function (usually ‘count’, ‘sum’, ‘min’, ‘max’).

args (optional) is a column label (string), or list of zero or more column labels

Valid values are:

  • Not specified: no arguments
  • null: no arguments
  • Empty list: no arguments
  • String: single argument, the name of a lattice column
  • List: multiple arguments, each a column label

Unlike lattice dimensions, measures can not be specified in qualified format, {@code [“table”, “column”]}. When you define a lattice, make sure that each column you intend to use as a measure has a unique label within the lattice (using “{@code AS label}” if necessary), and use that label when you want to pass the column as a measure argument.