table <table_name>[:<parent table name>] {
...
}

table <table name> {
  type = plain
  path = /path/to/table
  source = <source_name>
  source = <another source_name>
  [stored_fields = <comma separated list of full-text fields that should be stored, all are stored by default, can be empty>]
}

table <table name> {
  type = rt
  path = /path/to/table

  rt_field = <full-text field name>
  rt_field = <another full-text field name>
  [rt_attr_uint = <integer field name>]
  [rt_attr_uint = <another integer field name, limit by N bits>:N]
  [rt_attr_bigint = <bigint field name>]
  [rt_attr_bigint = <another bigint field name>]
  [rt_attr_multi = <multi-integer (MVA) field name>]
  [rt_attr_multi = <another multi-integer (MVA) field name>]
  [rt_attr_multi_64 = <multi-bigint (MVA) field name>]
  [rt_attr_multi_64 = <another multi-bigint (MVA) field name>]
  [rt_attr_float = <float field name>]
  [rt_attr_float = <another float field name>]
  [rt_attr_float_vector = <float vector field name>]
  [rt_attr_float_vector = <another float vector field name>]
  [rt_attr_bool = <boolean field name>]
  [rt_attr_bool = <another boolean field name>]
  [rt_attr_string = <string field name>]
  [rt_attr_string = <another string field name>]
  [rt_attr_json = <json field name>]
  [rt_attr_json = <another json field name>]
  [rt_attr_timestamp = <timestamp field name>]
  [rt_attr_timestamp = <another timestamp field name>]

  [stored_fields = <comma separated list of full-text fields that should be stored, all are stored by default, can be empty>]

  [rt_mem_limit = <RAM chunk max size, default 128M>]
  [optimize_cutoff = <max number of RT table disk chunks>]

}

type = plain

type = rt

Table type: "plain" or "rt" (real-time)

Value: plain (default), rt

path = path/to/table

The path to where the table will be stored or located, either absolute or relative, without the extension.

Value: The path to the table, mandatory

stored_fields = title, content

CREATE TABLE products(title text, content text stored indexed, name text indexed, price float)

POST /cli -d "
CREATE TABLE products(title text, content text stored indexed, name text indexed, price float)"

$params = [
    'body' => [
        'columns' => [
            'title'=>['type'=>'text'],
            'content'=>['type'=>'text', 'options' => ['indexed', 'stored']],
            'name'=>['type'=>'text', 'options' => ['indexed']],
            'price'=>['type'=>'float']
        ]
    ],
    'table' => 'products'
];
$index = new \Manticoresearch\Index($client);
$index->create($params);

utilsApi.sql('CREATE TABLE products(title text, content text stored indexed, name text indexed, price float)')

res = await utilsApi.sql('CREATE TABLE products(title text, content text stored indexed, name text indexed, price float)');

utilsApi.sql("CREATE TABLE products(title text, content text stored indexed, name text indexed, price float)");

utilsApi.Sql("CREATE TABLE products(title text, content text stored indexed, name text indexed, price float)");

table products {
  stored_fields = title, content # we want to store only "title" and "content", "name" shouldn't be stored

  type = rt
  path = tbl
  rt_field = title
  rt_field = content
  rt_field = name
  rt_attr_uint = price
}

stored_only_fields = title,content

List of fields that will be stored in the table, but not indexed. This setting works similarly to stored_fields except that when a field is specified in stored_only_fields t will only be stored, not indexed, and cannot be searched using full-text queries. It can only be retrieved in search results.

The value is a comma-separated list of fields that should be stored only, not indexed. By default, this value is empty. If a real-time table is being defined, the fields listed in stored_only_fields must also be declared as rt_field.

Note also, that you don't need to list attributes instored_only_fields,since their original values are stored anyway. If to compare stored_only_fields to string attributes the former (stored field):

• is stored on disk and does not take up memory
• is stored in a compressed format
• can only be fetched, it cannot be used for sorting, filtering or grouping

In contrast, the latter (string attribute):

• is stored on disk and in memory
• is stored in an uncompressed format
• can be used for sorting, grouping, filtering, and any other actions you want to take with attributes.

json_secondary_indexes = json_attr

CREATE TABLE products(title text, j json secondary_index='1')

POST /cli -d "
CREATE TABLE products(title text, j json secondary_index='1')"

$params = [
    'body' => [
        'columns' => [
            'title'=>['type'=>'text'],
            'j'=>['type'=>'json', 'options' => ['secondary_index' => 1]]
        ]
    ],
    'table' => 'products'
];
$index = new \Manticoresearch\Index($client);
$index->create($params);

utilsApi.sql('CREATE TABLE products(title text, j json secondary_index='1')')

res = await utilsApi.sql('CREATE TABLE products(title text, j json secondary_index=\'1\')');

utilsApi.sql("CREATE TABLE products(title text, j json secondary_index='1')");

utilsApi.Sql("CREATE TABLE products(title text, j json secondary_index='1')");

table products {
  json_secondary_indexes = j

  type = rt
  path = tbl
  rt_field = title
  rt_attr_json = j
}

The maximum number of disk chunks for the RT table. Learn more here.

rt_field = subject

This field declaration determines the full-text fields that will be indexed. The field names must be unique, and the order is preserved. When inserting data, the field values must be in the same order as specified in the configuration.

This is a multi-value, optional field.

rt_attr_uint = gid

This declaration defines an unsigned integer attribute.

Value: the field name or field_name:N (where N is the maximum number of bits to keep).

rt_attr_bigint = gid

This declaration defines a BIGINT attribute.

Value: field name, multiple records allowed.

rt_attr_multi = tags

Declares a multi-valued attribute (MVA) with unsigned 32-bit integer values.

Value: field name. Multiple records allowed.

rt_attr_multi_64 = wide_tags

Declares a multi-valued attribute (MVA) with signed 64-bit BIGINT values.

Value: field name. Multiple records allowed.

rt_attr_float = lat
rt_attr_float = lon

Declares floating point attributes with single precision, 32-bit IEEE 754 format.

Value: field name. Multiple records allowed.

rt_attr_float_vector = image_vector

Declares a vector of floating-point values.

Value: field name. Multiple records allowed.

rt_attr_bool = available

Declares a boolean attribute with 1-bit unsigned integer values.

Value: field name.

rt_attr_string = title

String attribute declaration.

Value: field name.

rt_attr_json = properties

Declares a JSON attribute.

Value: field name.

rt_attr_timestamp = date_added

Declares a timestamp attribute.

Value: field name.

rt_mem_limit = 512M

Memory limit for a RAM chunk of the table. Optional, default is 128M.

RT tables store some data in memory, known as the "RAM chunk," and also maintain a number of on-disk tables, referred to as "disk chunks." This directive allows you to control the size of the RAM chunk. When there is too much data to keep in memory, RT tables will flush it to disk, activate a newly created disk chunk, and reset the RAM chunk.

Please note that the limit is strict, and RT tables will never allocate more memory than what is specified in the rt_mem_limit. Additionally, memory is not preallocated, so specifying a 512MB limit and only inserting 3MB of data will result in allocating only 3MB, not 512MB.

The rt_mem_limit is never exceeded, but the actual RAM chunk size can be significantly lower than the limit. RT tables adapt to the data insertion pace and adjust the actual limit dynamically to minimize memory usage and maximize data write speed. This is how it works:

• By default, the RAM chunk size is 50% of the rt_mem_limit, referred to as the "rt_mem_limit".
• As soon as the RAM chunk accumulates data equivalent to rt_mem_limit * rate data (50% of rt_mem_limit by default), Manticore starts saving the RAM chunk as a new disk chunk.
• While a new disk chunk is being saved, Manticore assesses the number of new/updated documents.
• After saving a new disk chunk, the rt_mem_limit rate is updated.
• The rate is reset to 50% each time you restart the searchd.

For instance, if 90MB of data is saved to a disk chunk and an additional 10MB of data arrives while the save is in progress, the rate would be 90%. Next time, the RT table will collect up to 90% of rt_mem_limit before flushing the data. The faster the insertion pace, the lower the rt_mem_limit rate. The rate varies between 33.3% to 95%. You can view the current rate of a table using the SHOW TABLE STATUS command.

In real-time mode, you can adjust the size limit of RAM chunks and the maximum number of disk chunks using the ALTER TABLE statement. To set rt_mem_limit to 1 gigabyte for the table "t," run the following query: ALTER TABLE t rt_mem_limit='1G'. To change the maximum number of disk chunks, run the query: ALTER TABLE t optimize_cutoff='5'.

In the plain mode, you can change the values of rt_mem_limit and optimize_cutoff by updating the table configuration or running the command ALTER TABLE <table_name> RECONFIGURE

• Real-time tables are similar to distributed consisting of multiple local tables, also known as disk chunks.
• Each RAM chunk is made up of multiple segments, which are special RAM-only tables.
• While disk chunks are stored on disk, RAM chunks are stored in memory.
• Each transaction made to a real-time table generates a new segment, and RAM chunk segments are merged after each transaction commit. It is more efficient to perform bulk INSERTs of hundreds or thousands of documents rather than multiple separate INSERTs with one document to reduce the overhead from merging RAM chunk segments.
• When the number of segments exceeds 32, they will be merged to keep the count below 32.
• Real-time tables always have one RAM chunk (which may be empty) and one or more disk chunks.
• Merging larger segments takes longer, so it's best to avoid having a very large RAM chunk (and therefore rt_mem_limit).
• The number of disk chunks depends on the data in the table and the rt_mem_limit setting.
• Searchd flushes the RAM chunk to disk (as a persisted file, not as a disk chunk) on shutdown and periodically according to the rt_flush_period setting. Flushing several gigabytes to disk may take some time.
• A large RAM chunk puts more pressure on storage, both when flushing to disk into the .ram file and when the RAM chunk is full and dumped to disk as a disk chunk.
• The RAM chunk is backed up by a binary log until it is flushed to disk, and a larger rt_mem_limit, setting will increase the time it takes to replay the binary log and recover the RAM chunk.
• The RAM chunk may be slightly slower than a disk chunk.
• Although the RAM chunk itself doesn't take up more memory than rt_mem_limit Manticore may take up more memory in some cases, such as when you start a transaction to insert data and don't commit it for a while. In this case, the data you have already transmitted within the transaction will remain in memory.

source = srcpart1
source = srcpart2
source = srcpart3

The source field specifies the source from which documents will be obtained during indexing of the current table. There must be at least one source. The sources can be of different types (e.g. one could be MySQL, another PostgreSQL). For more information on indexing from external storages, indexing from external storages here

Value: The name of the source is mandatory. Multiple values are allowed.

killlist_target = main:kl

This setting determines the table(s) to which the kill-list will be applied. Matches in the targeted table that are updated or deleted in the current table will be suppressed. In :kl mode, the documents to suppress are taken from the kill-list. In :id mode, all document IDs from the current table are suppressed in the targeted one. If neither is specified, both modes will take effect. Learn more about kill-lists here

Value: not specified (default), target_table_name:kl, target_table_name:id, target_table_name. Multiple values are allowed

columnar_attrs = *
columnar_attrs = id, attr1, attr2, attr3

This configuration setting determines which attributes should be stored in the columnar storage instead of the row-wise storage.

You can set columnar_attrs = * to store all supported data types in the columnar storage.

Additionally, id is a supported attribute to store in the columnar storage.

columnar_strings_no_hash = attr1, attr2, attr3

By default, all string attributes stored in columnar storage store pre-calculated hashes. These hashes are used for grouping and filtering. However, they occupy extra space, and if you don't need to group by that attribute, you can save space by disabling hash generation.

CREATE TABLE [IF NOT EXISTS] name ( <field name> <field data type> [data type options] [, ...]) [table_options]

CREATE TABLE products (title text, price float) morphology='stem_en'

CREATE TABLE products (title text indexed, description text stored, author text, price float)

create table ... engine='columnar';
create table ... engine='rowwise';

The engine setting changes the default attribute storage for all attributes in the table. You can also specify engine separately for each attribute.

For information on how to enable columnar storage for a plain table, see columnar_attrs .

Values:

• columnar - Enables columnar storage for all table attributes, except for json
• rowwise (default) - Doesn't change anything and uses the traditional row-wise storage for the table.

The following settings are applicable for both real-time and plain tables, regardless of whether they are specified in a configuration file or set online using the CREATE or ALTER command.

Manticore supports two access modes for reading table data: seek+read and mmap.

In seek+read mode, the server uses the pread system call to read document lists and keyword positions, represented by the*.spd and *.spp files. The server uses internal read buffers to optimize the reading process, and the size of these buffers can be adjusted using the options read_buffer_docs and read_buffer_hits.There is also the option preopen that controls how Manticore opens files at start.

In mmap access mode, the search server maps the table's file into memory using the mmap system call, and the OS caches the file contents. The options read_buffer_docs and read_buffer_hits have no effect for corresponding files in this mode. The mmap reader can also lock the table's data in memory using themlock privileged call, which prevents the OS from swapping the cached data out to disk.

To control which access mode to use, the options access_plain_attrs, access_blob_attrs, access_doclists, access_hitlists and access_dict are available, with the following values:

Here is a table which can help you select your desired mode:

• For the fastest search response time and ample memory availability, use row-wise attributes and lock them in memory using mlock. Additionally, use mlock for doclists/hitlists.
• If you prioritize can't afford lower performance after start and are willing to sacrifice longer startup time, use the --force-preread. option. If you desire faster searchd restart, stick to the default mmap_preread option.
• If you are looking to conserve memory, while still having enough memory for all attributes, skip the use of mlock. The operating system will determine what should be kept in memory based on frequent disk reads.
• If row-wise attributes do not fit into memory, opt for columnar attributes
• If full-text search performance is not a concern, and you wish to save memory, use access_doclists/access_hitlists=file

The default mode offers a balance of:

• mmap,
• Prereading non-columnar attributes,
• Seeking and reading columnar attributes with no preread,
• Seeking and reading doclists/hitlists with no preread.

This provides a decent search performance, optimal memory utilization, and faster searchd restart in most scenarios.

attr_update_reserve = 256k

This setting reserves extra space for updates to blob attributes such as multi-value attributes (MVA), strings, and JSON. The default value is 128k. When updating these attributes, their length may change. If the updated string is shorter than the previous one, it will overwrite the old data in the *.spb file. If the updated string is longer, it will be written to the end of the *.spb file. This file is memory-mapped, making resizing it a potentially slow process, depending on the operating system's memory-mapped file implementation. To avoid frequent resizing, you can use this setting to reserve extra space at the end of the .spb file.

Value: size, default 128k.

docstore_block_size = 32k

This setting controls the size of blocks used by the document storage. The default value is 16kb. When original document text is stored using stored_fields or stored_only_fields, it is stored within the table and compressed for efficiency. To optimize disk access and compression ratios for small documents, these documents are concatenated into blocks. The indexing process collects documents until their total size reaches the threshold specified by this option. At that point, the block of documents is compressed. This option can be adjusted to achieve better compression ratios (by increasing the block size) or faster access to document text (by decreasing the block size).

Value: size, default 16k.

docstore_compression = lz4hc

This setting determines the type of compression used for compressing blocks of documents stored in document storage. If stored_fields or stored_only_fields are specified, the document storage stores compressed document blocks. 'lz4' offers fast compression and decompression speeds, while 'lz4hc' (high compression) sacrifices some compression speed for a better compression ratio. 'none' disables compression completely.

Values: lz4 (default), lz4hc, none.

docstore_compression_level = 12

The compression level used when 'lz4hc' compression is applied in document storage. By adjusting the compression level, you can find the right balance between performance and compression ratio when using 'lz4hc' compression. Note that this option is not applicable when using 'lz4' compression.

Value: An integer between 1 and 12, with a default of 9.

preopen = 1

This setting indicates that searchd should open all table files on startup or rotation, and keep them open while running. By default, the files are not pre-opened. Pre-opened tables require a few file descriptors per table, but they eliminate the need for per-query open() calls and are immune to race conditions that might occur during table rotation under high load. However, if you are serving many tables, it may still be more efficient to open them on a per-query basis in order to conserve file descriptors.

Value: 0 (default), or 1.

read_buffer_docs = 1M

Buffer size for storing the list of documents per keyword. Increasing this value will result in higher memory usage during query execution, but may reduce I/O time.

Value: size, default 256k, minimum value is 8k.

read_buffer_hits = 1M

Buffer size for storing the list of hits per keyword. Increasing this value will result in higher memory usage during query execution, but may reduce I/O time.

Value: size, default 256k, minimum value is 8k.

inplace_enable = {0|1}

table products {
  inplace_enable = 1

  path = products
  source = src_base
}

inplace_hit_gap = size

table products {
  inplace_hit_gap = 1M
  inplace_enable = 1

  path = products
  source = src_base
}

inplace_reloc_factor = 0.1

table products {
  inplace_reloc_factor = 0.1
  inplace_enable = 1

  path = products
  source = src_base
}

inplace_write_factor = 0.1

table products {
  inplace_write_factor = 0.1
  inplace_enable = 1

  path = products
  source = src_base
}

The following settings are supported. They are all described in section NLP and tokenization.

• bigram_freq_words
• bigram_index
• blend_chars
• blend_mode
• charset_table
• dict
• embedded_limit
• exceptions
• expand_keywords
• global_idf
• hitless_words
• html_index_attrs
• html_remove_elements
• html_strip
• ignore_chars
• index_exact_words
• index_field_lengths
• index_sp
• index_token_filter
• index_zones
• infix_fields
• killlist_target
• max_substring_len
• min_infix_len
• min_prefix_len
• min_stemming_len
• min_word_len
• morphology
• morphology_skip_fields
• ngram_chars
• ngram_len
• overshort_step
• phrase_boundary
• phrase_boundary_step
• prefix_fields
• regexp_filter
• stopwords
• stopword_step
• stopwords_unstemmed
• stored_fields
• stored_only_fields
• wordforms