Indexes | Tarantool

Indexes

An index is a special data structure that stores a group of key values and pointers. It is used for efficient manipulations with data.

As with spaces, you should specify the index name and let Tarantool come up with a unique numeric identifier (“index id”).

An index always has a type. The default index type is TREE. TREE indexes are provided by all Tarantool engines, can index unique and non-unique values, support partial key searches, comparisons, and ordered results. Additionally, the memtx engine supports HASH, RTREE and BITSET indexes.

An index may be multi-part, that is, you can declare that an index key value is composed of two or more fields in the tuple, in any order. For example, for an ordinary TREE index, the maximum number of parts is 255.

An index may be unique, that is, you can declare that it would be illegal to have the same key value twice.

The first index defined on a space is called the primary key index, and it must be unique. All other indexes are called secondary indexes, and they may be non-unique.

Indexes have certain limitations. See details on page Limitations.

To create a generator for indexes, you can use a sequence object. Learn how to do it in the tutorial.

Not to be confused with index types – the types of the data structure that is an index. See more about index types below.

Indexes restrict values that Tarantool can store with MsgPack. This is why, for example, 'unsigned' and 'integer' are different field types, although in MsgPack they are both stored as integer values. An 'unsigned' index contains only non-negative integer values, while an ‘integer’ index contains any integer values.

The default field type is 'unsigned' and the default index type is TREE. Although 'nil' is not a legal indexed field type, indexes may contain nil as a non-default option.

To learn more about field types, check the Field type details section.

Field type name string Field type Index type
'boolean' boolean TREE or HASH
'integer' (may also be called 'int') integer, which may include unsigned values TREE or HASH
'unsigned' (may also be called 'uint' or 'num', but 'num' is deprecated) unsigned TREE, BITSET, or HASH
'double' double TREE or HASH
'number' number, which may include integer, double, or decimal values TREE or HASH
'decimal' decimal TREE or HASH
'string' (may also be called 'str') string TREE, BITSET, or HASH
'varbinary' varbinary TREE, HASH, or BITSET (since version 2.7.1)
'uuid' uuid TREE or HASH
'datetime' datetime TREE
'array' array RTREE
'map' table Cannot be indexed
'scalar' may include nil, boolean, integer, unsigned, number, decimal, string, varbinary, or uuid values |
When a scalar field contains values of different underlying types, the key order is: nils, then booleans, then numbers, then strings, then varbinaries, then uuids.
TREE or HASH

An index always has a type. Different types are intended for different usage scenarios.

We give an overview of index features in the following table:

Feature TREE HASH RTREE BITSET
unique + + - -
non-unique + - + +
is_nullable + - - -
can be multi-part + + - -
multikey + - - -
partial-key search + - - -
can be primary key + + - -
exclude_null (version 2.8+) + - - -
Pagination (the after option) + - - -
iterator types ALL, EQ, REQ, GT, GE, LT, LE ALL, EQ ALL, EQ, GT, GE, LT, LE, OVERLAPS, NEIGHBOR ALL, EQ, BITS_ALL_SET, BITS_ANY_SET, BITS_ALL_NOT_SET

Note

In 2.11.0, the GT index type is deprecated for HASH indexes.

The default index type is ‘TREE’. TREE indexes are provided by memtx and vinyl engines, can index unique and non-unique values, support partial key searches, comparisons and ordered results.

This is a universal type of indexes, for most cases it will be the best choice.

Additionally, memtx engine supports HASH, RTREE and BITSET indexes.

HASH indexes require unique fields and loses to TREE in almost all respects. So we do not recommend to use it in the applications. HASH is now present in Tarantool mainly because of backward compatibility.

Here are some tips. Do not use HASH index:

  • just if you want to
  • if you think that HASH is faster with no performance metering
  • if you want to iterate over the data
  • for primary key
  • as an only index

Use HASH index:

  • if it is a secondary key
  • if you 100% won’t need to make it non-unique
  • if you have taken measurements on your data and you see an accountable increase in performance
  • if you save every byte on tuples (HASH is a little more compact)

RTREE is a multidimensional index supporting up to 20 dimensions. It is used especially for indexing spatial information, such as geographical objects. In this example we demonstrate spatial searches via RTREE index.

RTREE index could not be primary, and could not be unique. The option list of this type of index may contain dimension and distance options. The parts definition must contain the one and only part with type array. RTREE index can accept two types of distance functions: euclid and manhattan.

Warning

Currently, the isolation level of RTREE indexes in MVCC transaction mode is read-committed (not serializable, as stated). If a transaction uses these indexes, it can read committed or confirmed data (depending on the isolation level). However, the indexes are subject to different anomalies that can make them unserializable.

Example 1:

my_space = box.schema.create_space("tester")
my_space:format({ { type = 'number', name = 'id' }, { type = 'array', name = 'content' } })
hash_index = my_space:create_index('primary', { type = 'tree', parts = {'id'} })
rtree_index = my_space:create_index('spatial', { type = 'RTREE', unique = false, parts = {'content'} })

Corresponding tuple field thus must be an array of 2 or 4 numbers. 2 numbers mean a point {x, y}; 4 numbers mean a rectangle {x1, y1, x2, y2}, where (x1, y1) and (x2, y2) - diagonal point of the rectangle.

my_space:insert{1, {1, 1}}
my_space:insert{2, {2, 2, 3, 3}}

Selection results depend on a chosen iterator. The default EQ iterator searches for an exact rectangle, a point is treated as zero width and height rectangle:

tarantool> rtree_index:select{1, 1}
---
- - [1, [1, 1]]
...

tarantool> rtree_index:select{1, 1, 1, 1}
---
- - [1, [1, 1]]
...

tarantool> rtree_index:select{2, 2}
---
- []
...

tarantool> rtree_index:select{2, 2, 3, 3}
---
- - [2, [2, 2, 3, 3]]
...

Iterator ALL, which is the default when no key is specified, selects all tuples in arbitrary order:

tarantool> rtree_index:select{}
---
- - [1, [1, 1]]
  - [2, [2, 2, 3, 3]]
...

Iterator LE (less or equal) searches for tuples with their rectangles within a specified rectangle:

tarantool> rtree_index:select({1, 1, 2, 2}, {iterator='le'})
---
- - [1, [1, 1]]
...

Iterator LT (less than, or strictly less) searches for tuples with their rectangles strictly within a specified rectangle:

tarantool> rtree_index:select({0, 0, 3, 3}, {iterator = 'lt'})
---
- - [1, [1, 1]]
...

Iterator GE searches for tuples with a specified rectangle within their rectangles:

tarantool> rtree_index:select({1, 1}, {iterator = 'ge'})
---
- - [1, [1, 1]]
...

Iterator GT searches for tuples with a specified rectangle strictly within their rectangles:

tarantool> rtree_index:select({2.1, 2.1, 2.9, 2.9}, {iterator = 'gt'})
---
- []
...

Iterator OVERLAPS searches for tuples with their rectangles overlapping specified rectangle:

tarantool> rtree_index:select({0, 0, 10, 2}, {iterator='overlaps'})
---
- - [1, [1, 1]]
  - [2, [2, 2, 3, 3]]
...

Iterator NEIGHBOR searches for all tuples and orders them by distance to the specified point:

tarantool> for i=1,10 do
         >    for j=1,10 do
         >        my_space:insert{i*10+j, {i, j, i+1, j+1}}
         >    end
         > end
---
...

tarantool> rtree_index:select({1, 1}, {iterator = 'neighbor', limit = 5})
---
- - [11, [1, 1, 2, 2]]
  - [12, [1, 2, 2, 3]]
  - [21, [2, 1, 3, 2]]
  - [22, [2, 2, 3, 3]]
  - [31, [3, 1, 4, 2]]
...

Example 2:

3D, 4D and more dimensional RTREE indexes work in the same way as 2D except that user must specify more coordinates in requests. Here’s short example of using 4D tree:

tarantool> my_space = box.schema.create_space("tester")
tarantool> my_space:format{ { type = 'number', name = 'id' }, { type = 'array', name = 'content' } }
tarantool> primary_index = my_space:create_index('primary', { type = 'TREE', parts = {'id'} })
tarantool> rtree_index = my_space:create_index('spatial', { type = 'RTREE', unique = false, dimension = 4, parts = {'content'} })
tarantool> my_space:insert{1, {1, 2, 3, 4}} -- insert 4D point
tarantool> my_space:insert{2, {1, 1, 1, 1, 2, 2, 2, 2}} -- insert 4D box

tarantool> rtree_index:select{1, 2, 3, 4} -- find exact point
---
- - [1, [1, 2, 3, 4]]
...

tarantool> rtree_index:select({0, 0, 0, 0, 3, 3, 3, 3}, {iterator = 'LE'}) -- select from 4D box
---
- - [2, [1, 1, 1, 1, 2, 2, 2, 2]]
...

tarantool> rtree_index:select({0, 0, 0, 0}, {iterator = 'neighbor'}) -- select neighbours
---
- - [2, [1, 1, 1, 1, 2, 2, 2, 2]]
  - [1, [1, 2, 3, 4]]
...

Note

Keep in mind that select NEIGHBOR iterator with unset limits extracts the entire space in order of increasing distance. And there can be tons of data, and this can affect the performance.

And another frequent mistake is to specify iterator type without quotes, in such way: rtree_index:select(rect, {iterator = LE}). This leads to silent EQ select, because LE is undefined variable and treated as nil, so iterator is unset and default used.

Bitset is a bit mask. You should use it when you need to search by bit masks. This can be, for example, storing a vector of attributes and searching by these attributes.

Warning

Currently, the isolation level of BITSET indexes in MVCC transaction mode is read-committed (not serializable, as stated). If a transaction uses these indexes, it can read committed or confirmed data (depending on the isolation level). However, the indexes are subject to different anomalies that can make them unserializable.

Example 1:

The following script shows creating and searching with a BITSET index. Notice that BITSET cannot be unique, so first a primary-key index is created, and bit values are entered as hexadecimal literals for easier reading.

tarantool> my_space = box.schema.space.create('space_with_bitset')
tarantool> my_space:create_index('primary_index', {
         >   parts = {1, 'string'},
         >   unique = true,
         >   type = 'TREE'
         > })
tarantool> my_space:create_index('bitset_index', {
         >   parts = {2, 'unsigned'},
         >   unique = false,
         >   type = 'BITSET'
         > })
tarantool> my_space:insert{'Tuple with bit value = 01', 0x01}
tarantool> my_space:insert{'Tuple with bit value = 10', 0x02}
tarantool> my_space:insert{'Tuple with bit value = 11', 0x03}
tarantool> my_space.index.bitset_index:select(0x02, {
         >   iterator = box.index.EQ
         > })
---
- - ['Tuple with bit value = 10', 2]
...
tarantool> my_space.index.bitset_index:select(0x02, {
         >   iterator = box.index.BITS_ANY_SET
         > })
---
- - ['Tuple with bit value = 10', 2]
  - ['Tuple with bit value = 11', 3]
...
tarantool> my_space.index.bitset_index:select(0x02, {
         >   iterator = box.index.BITS_ALL_SET
         > })
---
- - ['Tuple with bit value = 10', 2]
  - ['Tuple with bit value = 11', 3]
...
tarantool> my_space.index.bitset_index:select(0x02, {
         >   iterator = box.index.BITS_ALL_NOT_SET
         > })
---
- - ['Tuple with bit value = 01', 1]
...

Example 2:

tarantool> box.schema.space.create('bitset_example')
tarantool> box.space.bitset_example:create_index('primary')
tarantool> box.space.bitset_example:create_index('bitset',{unique = false, type = 'BITSET', parts = {2,'unsigned'}})
tarantool> box.space.bitset_example:insert{1,1}
tarantool> box.space.bitset_example:insert{2,4}
tarantool> box.space.bitset_example:insert{3,7}
tarantool> box.space.bitset_example:insert{4,3}
tarantool> box.space.bitset_example.index.bitset:select(2, {iterator = 'BITS_ANY_SET'})

The result will be:

---
- - [3, 7]
  - [4, 3]
...

because (7 AND 2) is not equal to 0, and (3 AND 2) is not equal to 0.

Additionally, there exist index iterator operations. They can only be used with code in Lua and C/C++. Index iterators are for traversing indexes one key at a time, taking advantage of features that are specific to an index type. For example, they can be used for evaluating Boolean expressions when traversing BITSET indexes, or for going in descending order when traversing TREE indexes.

Found what you were looking for?
Feedback