You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

6.3KB

Tantivy-cli is the project hosting the command line interface for tantivy, a search engine project.

Tutorial: Indexing Wikipedia with Tantivy CLI

Introduction

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

Installing the tantivy CLI.

There are a couple ways to add the tantivy CLI to your computer.

If you are a rust programmer, you probably have cargo installed and you can just run cargo install tantivy-cli.

Alternatively, if you are on 64-bit Linux, you can directly download a static binary: binaries/linux_x86_64/, and save it in a directory on your system's PATH.

Creating the index: new

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

    # create the directory
    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 u32 or str
  • how it should be indexed.

You can find more information about the latter on [tantivy's schema documentation page](http://fulmicoton.com/tantivy/tantivy/schema/index.html

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. (To be honest, phrase queries are not yet implemented in tantivy, so the positions won't be really useful in this tutorial.)

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 it's schema! 



    New field name  ? title
    Text or unsigned 32-bit Integer (T/I) ? T
    Should the field be stored (Y/N) ? Y
    Should the field be indexed (Y/N) ? Y
    Should the field 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
    Text or unsigned 32-bit Integer (T/I) ? T
    Should the field be stored (Y/N) ? Y
    Should the field be indexed (Y/N) ? Y
    Should the field 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
    Text or unsigned 32-bit Integer (T/I) ? T
    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 may check its content.

It contains two sections:

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

Indexing the document: index

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

    bunzip2 wiki-articles.json.bz2

If you are in a rush you can download 100 articles in the right format here.

The index command will index your document. By default it will use as many threads as there are cores on your machine. You can change the number of threads by passing it the -t parameter.

On my computer (8 core Xeon(R) CPU X3450 @ 2.67GHz), it will take around 6 minutes.

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

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

    ls ./wikipedia-index

If you indexed the 5 million articles, you should see a lot of new files, all with the following format The main file is meta.json.

Our index is in fact divided in segments. Each segment acts as an individual smaller index. Its named is simply a uuid.

Serve the search index: serve

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&explain=true&nhits=20

Optimizing the index: merge

Each of tantivy's indexer threads closes a new segment every 100K documents (this is completely arbitrary at the moment). You should have more than 50 segments in your dictionary.

Having that many segments hurts your query performance (well, mostly the fast ones). Tantivy merge will merge your segments into one.

    tantivy merge -i ./wikipedia-index

(The command takes around 7 minutes on my computer)

Note that your files are still there even after having run the command. However, meta.json only lists one of the segments. You will still need to remove the files manually.