Schemaless Serialization

The binary schemaless serialization is an attempt to define a serialization format that can serialize a document containing all the information about the structure and the data with support for partial serialization/deserialization.

The types described here are different from the types used in the binary protocol.

A serialized record has the following shape:

  1. (serialization-version:byte)(class-name:string)(header:byte[])(data:byte[])

Version

1 byte that contains the version of the current serialization (in order to allow progressive serialization upgrades).

Class Name

A string containing the name of the class of the record. If the record has no class, class-name will be just an empty string.

Header

The header contains a list of fields of the serialized records, along with their position in the data field. The header has the following shape:

  1. (field:field-definition)+

Where field-definition has this shape:

  1. (field-type:varint)(field-contents:byte[])

The field contents depend on the value of the field-type varint. Once decoded to an integer, its value can be:

  • 0 - signals the end of the header
  • a positive number i - signals that what follows is a named field (and it’s the length of the field name, see below)
  • a negative number i - signals that what follows is a property (and it encodes the property id, see below)

Named fields

Named fields are encoded as:

  1. (field-name:string)(pointer-to-data-structure:int32)(data-type:byte)
  • field-name - the name of the field. The field-type varint is included in field-name (as per the string encoding) as mentioned above.
  • pointer-to-data-structure - a pointer to the data structure for the current field in the data segment. It’s 0 if the field is null.
  • data-type - the type id of the type of the value for the current field. The supported types (with their ids) are defined in this section.

Properties

Properties are encoded as:

  1. (property-id:varint)(pointer-to-data-structure:int32)
  • property-id - is the field-type described above. It’s a negative value that encodes a property id. Decoding of this value into the corresponding property id can be done using this formula: (property-id * -1) - 1. The property identified by this id will be found in the schema (with its name and its type), stored in the globalProperties field at the root of the document that represents the schema definition.
  • pointer-to-data-structure - a pointer to the data structure for the current field in the data segment. It’s 0 if the field is null.

Data

The data segment is where the values of the fields are stored. It’s an array of data structures (each with a type described by the corresponding field).

  1. (data:data-structure[])

Each type is serialized differently, as described below.

Type serialization

SHORT, INTEGER, LONG

Integers are encoded using the varint (with ZigZag) algorithm used by Google’s ProtoBuf (and specified here).

BYTE

Bytes is stored raw, as single bytes.

BOOLEAN

Booleans are serialized using a single byte, 0 meaning false and 1 meaning true.

FLOAT

This is stored as flat byte array copying the memory from the float memory.

  1. (float:byte[4])

DOUBLE

This is stored as flat byte array copying the memory from the double memory.

  1. (double:byte[8])

DATETIME

The date is converted to milliseconds (from the Unix epoch) and stored as the type LONG.

DATE

The date is converted to seconds (from the Unix epoch), moved to midnight UTC+0, divided by 86400 (the number of seconds in a day) and stored as the type LONG.

STRING

Strings are stored as binary structures (encoded as UTF-8). Strings are preceded by their size (in bytes).

  1. (size:varint)(string:byte[])
  • size - the number of the bytes in the string stored as a varint
  • string - the bytes of the string encoded as UTF-8

BINARY

Byte arrays are stored like STRINGs.

  1. (size:varint)(bytes:byte[])
  • size - the number of the bytes to store
  • bytes - the raw bytes

EMBEDDED

Embedded documents are serialized using the protocol described in this document (recursively). The serialized document is just a byte array.

  1. (serialized-document:bytes[])

EMBEDDEDLIST, EMBEDDEDSET

The embedded collections are stored as an array of bytes that contain the serialized document in the embedded mode.

  1. (size:varint)(collection-type:byte)(items:item[])
  • size - the number of items in the list/set
  • collection-type - the type of the elements in the list or ANY if the type is unknown.
  • items an array of values serialized by type or if the type of the collection is ANY the item will have it’s own structure.

The item data structure has the following shape:

  1. (data-type:byte)(data:byte)
  • data-type - the type id of the data in the item
  • data - the data in the item serialized according to the data-type

EMBEDDEDMAP

Maps can have keys with the following types:

  • STRING
  • SHORT
  • INTEGER
  • LONG
  • BYTE
  • DATE
  • DATETIME
  • DECIMAL
  • FLOAT
  • DOUBLE

As of now though, all keys are converted to STRINGs.

An EMBEDDEDMAP is serialized as an header and a list of values.

  1. (size:varint)(header:header-structure)(values:byte[][])
  • size - the number of key-value pairs in the map
  • header - serialized as (key-type:byte)(key-value:byte[])(pointer-to-data:int32)(value-type:byte) (where pointer-to-data is the same as the one in the header, offsetting from the start of the top-level document).
  • values - the values serialized according to their type.

The link is stored as two 64 bit integers: the cluster id and the record’s position in the cluster.

  1. (cluster-id:int64)(record-position:int64)

Link collections (lists and sets) are serialized as the size of the collection and then a list of LINKs.

  1. (size:varint)(links:LINK[])

LINKMAP

Maps of links can have keys with the following types:

  • STRING
  • SHORT
  • INTEGER
  • LONG
  • BYTE
  • DATE
  • DATETIME
  • DECIMAL
  • FLOAT
  • DOUBLE

As of now though, all keys are converted to STRINGs.

A LINKMAP is serialized as the number of key-value pairs and then the list of key-value pairs.

  1. (size:varint)(key-value-pairs:key-value[])

A key-value pair is serialized as:

  1. (key-type:byte)(key-value:byte[])(link:LINK)
  • key-type - the type id of the type of the key
  • key-value - the value of the key, serialized according to key-type
  • link - the link value

DECIMAL

Decimals are converted to integers and stored as the scale and the value. For example, 10234.546 is storead as scale 3 and value 10234546.

  1. (scale:int32)(value-size:int32)(value:byte[])
  • scale - the scale of the decimal.
  • value-size - the number of bytes that form the value.
  • value - the bytes representing the value of the decimal (in big-endian order).

LINKBAG

No documentation yet. :(