gem-hunter

tantivy-cli is the the command line interface for the tantivy search engine. It provides indexing and search capabilities, and is suitable for smaller projects.

For a more complete solution around tantivy, you may use

• https://github.com/quickwit-oss/quickwit
• https://github.com/lnx-search/lnx

In this tutorial, we will create a brand new index with the articles of English wikipedia in it.

There are a couple ways to install tantivy-cli.

If you are a Rust programmer, you probably have cargo installed and you can just:

cargo install tantivy-cli --locked

Let's create a directory in which your index will be stored.

mkdir wikipedia-index

We will now initialize the index and create its schema. The schema defines the list of your fields, and for each field:

• its name
• its type, currently u64, i64 or str
• how it should be indexed.

You can find more information about the latter on tantivy's schema documentation page

In our case, our documents will contain

• a title
• a body
• a url

We want the title and the body to be tokenized and indexed. We also want to add the term frequency and term positions to our index.

Note: You can also infer them from a sample of your documents (tantivy new -i wikipedia-index --infer-from-ndjson wiki-articles.jsonl), but in this tutorial we will create the schema manually.

Running tantivy new will start a wizard that will help you define the schema of the new index.

Like all the other commands of tantivy, you will have to pass it your index directory via the -i or --index parameter as follows:

tantivy new -i wikipedia-index

Answer the questions as follows:

    Creating new index 
    Let's define its schema! 



    New field name  ? title
    Choose Field Type (Text/u64/i64/f64/Date/Facet/Bytes) ? Text
    Should the field be stored (Y/N) ? Y
    Should the field be indexed (Y/N) ? Y
    Should the term be tokenized? (Y/N) ? Y
    Should the term frequencies (per doc) be in the index (Y/N) ? Y
    Should the term positions (per doc) be in the index (Y/N) ? Y
    Add another field (Y/N) ? Y
    
    
    
    New field name  ? body
    Choose Field Type (Text/u64/i64/f64/Date/Facet/Bytes) ? Text
    Should the field be stored (Y/N) ? Y
    Should the field be indexed (Y/N) ? Y
    Should the term be tokenized? (Y/N) ? Y
    Should the term frequencies (per doc) be in the index (Y/N) ? Y
    Should the term positions (per doc) be in the index (Y/N) ? Y
    Add another field (Y/N) ? Y
    
    
    
    New field name  ? url
    Choose Field Type (Text/u64/i64/f64/Date/Facet/Bytes) ? Text
    Should the field be stored (Y/N) ? Y
    Should the field be indexed (Y/N) ? N
    Add another field (Y/N) ? N


    [
    {
        "name": "title",
        "type": "text",
        "options": {
            "indexing": "position",
            "stored": true
        }
    },
    {
        "name": "body",
        "type": "text",
        "options": {
            "indexing": "position",
            "stored": true
        }
    },
    {
        "name": "url",
        "type": "text",
        "options": {
            "indexing": "unindexed",
            "stored": true
        }
    }
    ]

After the wizard has finished, a meta.json should exist in wikipedia-index/meta.json. It is a fairly human readable JSON, so you can check its content.

It contains two sections:

• segments (currently empty, but we will change that soon)
• schema

Tantivy's index command offers a way to index a json file. The file must contain one JSON object per line. The structure of this JSON object must match that of our schema definition.

{"body": "some text", "title": "some title", "url": "http://somedomain.com"}

For this tutorial, you can download a corpus with the 5 million+ English Wikipedia articles in the right format here: wiki-articles.json (2.34 GB). Make sure to decompress the file. Also, you can avoid this if you have bzcat installed so that you can read it compressed.

bunzip2 wiki-articles.json.bz2

If you are in a rush you can download 100 articles in the right format here (11 MB).

The index command will index your document. By default it will use as 3 thread, each with a buffer size of 1GB split a across these threads.

cat wiki-articles.json | tantivy index -i ./wikipedia-index

You can change the number of threads by passing it the -t parameter, and the total buffer size used by the threads heap by using the -m. Note that tantivy's memory usage is greater than just this buffer size parameter.

On my computer (8 core Xeon(R) CPU X3450 @ 2.67GHz), on 8 threads, indexing wikipedia takes around 9 minutes.

While tantivy is indexing, you can peek at the index directory to check what is happening.

ls ./wikipedia-index

The main file is meta.json.

You should also see a lot of files with a UUID as filename, and different extensions. Our index is in fact divided in segments. Each segment acts as an individual smaller index. Its name is simply a uuid.

If you decided to index the complete wikipedia, you may also see some of these files disappear. Having too many segments can hurt search performance, so tantivy actually automatically starts merging segments.

Tantivy's cli also embeds a search server. You can run it with the following command.

tantivy serve -i wikipedia-index

By default, it will serve on port 3000.

You can search for the top 20 most relevant documents for the query Barack Obama by accessing the following url in your browser

http://localhost:3000/api/?q=barack+obama&nhits=20

By default this query is treated as barack OR obama. You can also search for documents that contains both term, by adding a + sign before the terms in your query.

http://localhost:3000/api/?q=%2Bbarack%20%2Bobama&nhits=20

Also, - makes it possible to remove documents the documents containing a specific term.

http://localhost:3000/api/?q=-barack%20%2Bobama&nhits=20

Finally tantivy handle phrase queries.

http://localhost:3000/api/?q=%22barack%20obama%22&nhits=20

You may also use the search command to stream all documents matching a specific query. The documents are returned in an unspecified order.

tantivy search -i wikipedia-index -q "barack obama"
tantivy search -i hdfs --query "*" --agg '{"severities":{"terms":{"field":"severity_text"}}}'

Tantivy's cli provides a simple benchmark tool. You can run it with the following command.

tantivy bench -i wikipedia-index -n 10 -q queries.txt