Skip to content
Go to file

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Use cases  |  Setup  |  Status  |  API  |  Contact

Build Status

Effortless JSON storage for Tarantool

You may use this module to receive and store structured data you get from external world. It has a few important strengths:

  • You are not forced to define any kind of schema for your documents
  • Still, they are stored with very little redundancy
  • You can build indices on arbitrary fields (even nested)
  • There are convenient high-level functions for data manipulation
  • The module works transparently for local spaces, remote spaces and even sharded spaces!
  • You can do "eventually consistent" selects and joins across sharded spaces!

Use cases

This module is suitable for projects where having a strict schema is not desirable. And especially for small codebases, where you don't want to write lots of boilerplate.


This module has no outside dependencies, so you can just drop document.lua into the root of your project.

Alternatively, you can use Tarantool package manager:

tarantoolctl rocks install document



doc = require('document')
json = require('json')


box.schema.create_space('test', {if_not_exists = true})
doc.create_index(, 'primary',
                 {parts={'id', 'unsigned'}, if_not_exists=true})

Actual data manipulation

doc.insert(, {id=1, foo="foo", bar={baz=3}})
doc.insert(, {id=2, foo="bar", bar={baz=0}})

print('All tuples')
for _, r in do
    print('tuple:', json.encode(r))

print('Tuples where bar.baz > 0')
for _, r in, {{'$bar.baz', '>', 0}}) do
    print('tuple:', json.encode(r))

print('Deleting a tuple where primary key == 2')
doc.delete(, {{"$id", "==", 2}})

How it works

A naive implementation would have just stored JSON documents as strings inside a tuple, and extracted indices into separate fields of the tuple.

A more optimized approach is what mongodb or postgresql are doing: instead of storing JSON documents as text, invent a compact binary format and store it inside a tuple.

But we decided to take another approach, and dynamically figure out document schema. We walk through the incoming document and put each leaf element into a separate tuple field, essentially "flattening" it. If we already saw such field previously, then schema already contains a mapping between path in the document and a position inside the tuple. If not, then we extend the schema and add a new field, assigning a new rightmost column in the tuple to store its data.

When data is selected back, we reconstruct the original object using document schema.

Our experiments show that most documents can achieve 5x to 10x compression due to the method, because the schema is stored only once per space.


Queries are written using Lua tables, and are just lists of conditions of the following form:

{left, op, right}

Where left and right parts of the condition are either regular values or references to field name, and op is a comparison operator.

Example values for left and right:

  • 1
  • nil
  • "foo"
  • "$id"

Here, the "$id" is a special form that references tuple field by name. You can put a "path" there, separated with ".", like "$".

Example values for op:

  • ">"
  • ">="
  • "=="
  • "<="
  • "<"

Query examples:

  • {{"$id", ">", 10}}
  • {{"$id", ">", 10}, {"$id", "<", 100}}
  • {{"$", "==", "foo"}, {"$qty", "==", 0}}


  • The functionality for dealing with regular spaces is feature-complete
  • Serialization/deserialization should be reasonably fast for most use-cases (though, there are no benchmarks at the moment)
  • Selects/joins across sharded spaces may have bugs. There is no automated test coverage for this case.


doc.insert(space, tbl)

Insert document tbl into space.

doc.delete(space, query)

Insert table tbl into space.

Delete documents from space, that match query (see Queries above), query, options)

Select documents from space that match query (see Queries above) and return an iterator to the result set.

options is a table with the following optional keys:

  • limit: maximum number of results to return
  • offset: the offset from the beginning of the result set

doc.join(space1, space2, query, options)

Perform an inner join of spaces space1 and space2, where both items satisfy query (see Queries above).

options is a table with the following optional keys:

  • limit: maximum number of results to return
  • offset: the offset from the beginning of the result set

Low level API

doc.flatten(space, tbl)

Converts document tbl to flat array, updating schema for space space as necessary.

doc.unflatten(space, tbl)

Converts flat array tbl to a nested document, according to schema for space space.

create_index(index_name, options)

Behaves similar to, but allows to specify string field names in addition to numeric in parts.

field_key(space, field_name)

Returns integer key for field named field_name in a flattened document. If you need a key for nested documents, use dot notation, like: "".


This module was initialy written by Konstantin Nazarov.

You can reach out to him at


Effortless JSON storage for Tarantool




No packages published


You can’t perform that action at this time.