Skip to content

altertable-ai/tantiny

 
 

Repository files navigation

Build workflow Tantiny License: MIT

This is a fork of the original Tantiny gem by Alexander Baygeldin. Following baygeldin#24 we agreed transfering ownership of the gem to Altertable so we can keep it up to date with the latest versions of Tantivy and Ruby.

Tantiny

Need a fast full-text search for your Ruby script, but don't want to host/operate a full-blown search engine yet?

You're in the right place. Tantiny is a minimalistic full-text search library for Ruby based on Tantivy (an awesome alternative to Apache Lucene written in Rust). It's great for cases when your task at hand requires a full-text search, but configuring a full-blown distributed search engine would take more time than the task itself. And even if you already use such an engine in your project (which is highly likely, actually), it still might be easier to just use Tantiny instead because unlike Solr, Elasticsearch, or any hosted search engine it doesn't need anything to work (no separate server, process, API or whatever), it's purely embeddable. So, when you find yourself in a situation when using your search engine of choice would be tricky/inconvinient or would require additional setup you can always revert back to a quick and dirty solution that is nontheless flexible and fast.

Tantiny is not exactly Ruby bindings to Tantivy, but it tries to be close. The main philosophy is to provide low-level access to Tantivy's inverted index, but with a nice Ruby-esque API, sensible defaults, and additional functionality sprinkled on top.

Take a look at the most basic example:

# Persisted index
index = Tantiny::Index.new("/path/to/index") { text :description }

# Or in-memory (no persistence)
index = Tantiny::Index.new { text :description }

index << { id: 1, description: "Hello World!" }
index << { id: 2, description: "What's up?" }
index << { id: 3, description: "Goodbye World!" }

index.reload

index.search("world") # 1, 3

Installation

Add this line to your application's Gemfile:

gem "tantiny"

And then execute:

$ bundle install

Or install it yourself as:

$ gem install tantiny

You don't have to have Rust installed on your system since Tantiny will try to download the pre-compiled binaries hosted on GitHub releases during the installation. However, if no pre-compiled binaries were found for your system (which is a combination of platform, architecture, and Ruby version) you will need to install Rust first.

Defining the index schema

Whether you want to use a persisted index or an in-memory index, you need to define the schema first:

Tantiny::Index.new(path_or_memory) do
  id :imdb_id
  string :title
  text :description
  integer :duration
  double :rating
  date :release_date
  facet :category
end

In-memory indexes

For small to medium datasets or temporary search needs (or tests!), you can create an in-memory index by omitting the path parameter:

index = Tantiny::Index.new do
  text :title
  text :description
  double :price
end

In-memory indexes are perfect when you don't need persistence between runs, or when you're building a search index from data that already exists in a database. They offer the same full-text search capabilities without any file I/O overhead.

Field types

Here are the descriptions for every field type:

Type Description
id Specifies where documents' ids are stored (defaults to :id).
facet Fields with values like /animals/birds (i.e. hierarchial categories).
string Fields with text that are not tokenized.
text Fields with text that are tokenized by the specified tokenizer.
integer Fields with integer values.
double Fields with float values.
date Fields with either DateTime type or something that converts to it.

Each field can either be a single value or an array of values.

Managing documents

You can feed the index any kind of object that has methods specified in your schema, but plain hashes also work:

rio_bravo = OpenStruct.new(
  imdb_id: "tt0053221",
  type: '/western/US',
  title: "Rio Bravo",
  description: "A small-town sheriff enlists a drunk, a kid and an old man to help him fight off a ruthless cattle baron.",
  duration: 141,
  rating: 8.0,
  release_date: Date.parse("March 18, 1959")
)

index << rio_bravo

hanabi = {
  imdb_id: "tt0119250",
  type: "/crime/Japan",
  title: "Hana-bi",
  description: "Nishi leaves the police in the face of harrowing personal and professional difficulties. Spiraling into depression, he makes questionable decisions.",
  duration: 103,
  rating: 7.7,
  release_date: Date.parse("December 1, 1998")
}

index << hanabi

brother = {
  imdb_id: "tt0118767",
  type: "/crime/Russia",
  title: "Brother",
  description: "An ex-soldier with a personal honor code enters the family crime business in St. Petersburg, Russia.",
  duration: 99,
  rating: 7.9,
  release_date: Date.parse("December 12, 1997")
}

index << brother

In order to update the document just add it again (as long as the id is the same):

rio_bravo.rating = 10.0
index << rio_bravo

You can also delete it if you want:

index.delete(rio_bravo.imdb_id)

Transactions

If you need to perform multiple writing operations (i.e. more than one) you should always use transaction:

index.transaction do
  index << rio_bravo
  index << hanabi
  index << brother
end

Transactions group changes and commit them to the index in one go. This is dramatically more efficient than performing these changes one by one. In fact, all writing operations (i.e. << and delete) are wrapped in a transaction implicitly when you call them outside of a transaction, so calling << 10 times outside of a transaction is the same thing as performing 10 separate transactions.

Concurrency and thread-safety

Tantiny is thread-safe meaning that you can safely share a single instance of the index between threads. You can also spawn separate processes that could write to and read from the same index. However, while reading from the index should be parallel, writing to it is not. Whenever you call transaction or any other operation that modify the index (i.e. << and delete) it will lock the index for the duration of the operation or wait for another process or thread to release the lock. The only exception to this is when there is another process with an index with an exclusive writer running somewhere in which case the methods that modify the index will fail immediately.

Thus, it's best to have a single writer process and many reader processes if you want to avoid blocking calls. The proper way to do this is to set exclusive_writer to true when initializing the index:

index = Tantiny::Index.new("/path/to/index", exclusive_writer: true) {}

This way the index writer will only be acquired once which means the memory for it and indexing threads will only be allocated once as well. Otherwise a new index writer is acquired every time you perform a writing operation.

Searching

Make sure that your index is up-to-date by reloading it first:

index.reload

And search it (finally!):

index.search("a drunk, a kid, and an old man")

By default it will return ids of 10 best matching documents, but you can customize it:

index.search("a drunk, a kid, and an old man", limit: 100)

You may wonder, how exactly does it conduct the search? Well, the default behavior is to use smart_query search (see below for details) over all text fields defined in your schema. So, you can pass the parameters that the smart_query accepts right here:

index.search("a dlunk, a kib, and an olt mab", fuzzy_distance: 1)

However, you can customize it by composing your own query out of basic building blocks:

popular_movies = index.range_query(:rating, 8.0..10.0)
about_sheriffs = index.term_query(:description, "sheriff")
crime_movies = index.facet_query(:cetegory, "/crime")
long_ass_movies = index.range_query(:duration, 180..9999)
something_flashy = index.smart_query(:description, "bourgeoisie")

index.search((popular_movies & about_sheriffs) | (crime_movies & !long_ass_movies) | something_flashy)

I know, weird taste! But pretty cool, huh? Take a look at all the available queries below.

Supported queries

Query Behavior
all_query Returns all indexed documents.
empty_query Returns exactly nothing (used internally).
term_query Documents that contain the specified term.
fuzzy_term_query Documents that contain the specified term within a Levenshtein distance.
phrase_query Documents that contain the specified sequence of terms.
regex_query Documents that contain a term that matches the specified regex.
prefix_query Documents that contain a term with the specified prefix.
range_query Documents that with an integer, double or date field within the specified range.
facet_query Documents that belong to the specified category.
smart_query A combination of term_query, fuzzy_term_query and prefix_query.

Searching on multiple fields

All queries can search on multuple fields (except for facet_query because it doesn't make sense there).

So, the following query:

index.term_query(%i[title description], "hello")

Is equivalent to:

index.term_query(:title, "hello") | index.term_query(:description, "hello")

Boosting queries

All queries support the boost parameter that allows to bump documents position in the search:

about_cowboys = index.term_query(:description, "cowboy", boost: 2.0)
about_samurai = index.term_query(:description, "samurai") # sorry, Musashi...

index.search(about_cowboys | about_samurai)

smart_query behavior

The smart_query search will extract terms from your query string using the respective field tokenizers and search the index for documents that contain those terms via the term_query. If the fuzzy_distance parameter is specified it will use the fuzzy_term_query. Also, it allows the last term to be unfinished by using the prefix_query.

So, the following query:

index.smart_query(%i[en_text ru_text], "dollars рубли eur", fuzzy_distance: 1)

Is equivalent to:

t1_en = index.fuzzy_term_query(:en_text, "dollar")
t2_en = index.fuzzy_term_query(:en_text, "рубли")
t3_en = index.fuzzy_term_query(:en_text, "eur")
t3_prefix_en = index.prefix_query(:en_text, "eur")

t1_ru = index.fuzzy_term_query(:ru_text, "dollars")
t2_ru = index.fuzzy_term_query(:ru_text, "рубл")
t3_ru = index.fuzzy_term_query(:ru_text, "eur")
t3_prefix_ru = index.prefix_query(:ru_text, "eur")

(t1_en & t2_en & (t3_en | t3_prefix_en)) | (t1_ru & t2_ru & (t3_ru | t3_prefix_ru))

Notice how words "dollars" and "рубли" are stemmed differently depending on the field we are searching. This is assuming we have en_text and ru_text fields in our schema that use English and Russian stemmer tokenizers respectively.

About regex_query

The regex_query accepts the regex pattern, but it has to be a Rust regex, not a Ruby Regexp. So, instead of index.regex_query(:description, /hel[lp]/) you need to use index.regex_query(:description, "hel[lp]"). As a side note, the regex_query is pretty fast because it uses the fst crate internally.

Tokenizers

So, we've mentioned tokenizers more than once already. What are they?

Tokenizers is what Tantivy uses to chop your text onto terms to build an inverted index. Then you can search the index by these terms. It's an important concept to understand so that you don't get confused when index.term_query(:description, "Hello") returns nothing because Hello isn't a term, but hello is. You have to extract the terms from the query before searching the index. Currently, only smart_query does that for you. Also, the only field type that is tokenized is text, so for string fields you should use the exact match (i.e. index.term_query(:title, "Hello")).

Specifying the tokenizer

By default the simple tokenizer is used, but you can specify the desired tokenizer globally via index options or locally via field specific options:

en_stemmer = Tantiny::Tokenizer.new(:stemmer)
ru_stemmer = Tantiny::Tokenizer.new(:stemmer, language: :ru)

Tantiny::Index.new "/tmp/index", tokenizer: en_stemmer do
  text :description_en
  text :description_ru, tokenizer: ru_stemmer
end

Simple tokenizer

Simple tokenizer chops the text on punctuation and whitespaces, removes long tokens, and lowercases the text.

tokenizer = Tantiny::Tokenizer.new(:simple)
tokenizer.terms("Hello World!") # ["hello", "world"]

Stemmer tokenizer

Stemmer tokenizers is exactly like simple tokenizer, but with additional stemming according to the specified language (defaults to English).

tokenizer = Tantiny::Tokenizer.new(:stemmer, language: :ru)
tokenizer.terms("Привет миру сему!") # ["привет", "мир", "сем"]

Take a look at the source to see what languages are supported.

Ngram tokenizer

Ngram tokenizer chops your text onto ngrams of specified size.

tokenizer = Tantiny::Tokenizer.new(:ngram, min: 5, max: 10, prefix_only: true)
tokenizer.terms("Morrowind") # ["Morro", "Morrow", "Morrowi", "Morrowin", "Morrowind"]

Retrieving documents

You may have noticed that search method returns only documents ids. This is by design. The documents themselves are not stored in the index. Tantiny is a minimalistic library, so it tries to keep things simple. If you need to retrieve a full document, use a key-value store like Redis alongside.

Highlighting

Tantiny supports highlighting of search results. This is useful when you want to display the search results in a more readable format.

Tantiny::Query.highlight(field_text, query_string)

It supports fuzzy highlighting by specifying the fuzzy distance.

Tantiny::Query.highlight(field_text, query_string, fuzzy_distance: 2)

As well as custom tokenizers, but make sure to use the same tokenizer that was used to index the field.

tokenizer = Tantiny::Tokenizer.new(:stemmer, language: :fr)
Tantiny::Query.highlight(field_text, query_string, tokenizer: tokenizer)

This will return the text with the terms highlighted:

Tantiny::Query.highlight("hellow world. you are welcome.", "hello you")
# "<b>hellow</b> world. <b>you</b> are welcome."

Examples

The examples directory contains practical demonstrations of Tantiny's capabilities. These examples are great starting points for understanding how to use Tantiny in real-world scenarios.

Simple Ranking Example

examples/simple_ranking.rb

A minimal demonstration of field-based ranking showing:

  • Creating an in-memory index
  • Using boost values to rank title matches higher than description matches
  • Side-by-side comparison of equal weights vs boosted fields

This is perfect for understanding the core concept of ranking in just a few lines of code.

Ecommerce Example

examples/ecommerce.rb

A comprehensive example demonstrating in-memory search for a product catalog:

  • In-memory indexing - Perfect for small to medium datasets without persistent storage
  • Product search - Indexing products with various attributes (title, description, category, price, stock)
  • Fuzzy search - Handling typos and misspellings (e.g., "loptop" → "laptop")
  • Field-based ranking - Boosting title matches to rank higher than description matches
  • Complex queries - Combining multiple conditions with AND/OR operators
  • Category filtering - Filtering products by exact category match
  • Price range queries - Finding products within a specific price range

See the examples README for more details.

Development

After checking out the repo, run bin/setup to install dependencies. Then, run rake build to build native extensions, and then rake spec to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

We use conventional commits to automatically generate the CHANGELOG, bump the semantic version, and to publish and release the gem. All you need to do is stick to the convention and CI will take care of everything else for you.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/altertable-ai/tantiny.

License

The gem is available as open source under the terms of the MIT License.

About

Tiny full-text search for Ruby powered by Tantivy

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Ruby 57.6%
  • Rust 42.3%
  • Shell 0.1%