Skip to content

Easily integrate Meilisearch into your MongoDB backed Rails models.

License

Notifications You must be signed in to change notification settings

masukomi/mongodb_meilisearch

Repository files navigation

Table of Contents

MongodbMeilisearch

A simple gem for integrating Meilisearch into Ruby applications that are backed by MongoDB.

Installation

Install the gem and add to the application’s Gemfile by executing:

$ bundle add mongodb_meilisearch

If bundler is not being used to manage dependencies, install the gem by executing:

$ gem install mongodb_meilisearch

Usage

A high level overview

Pre-Requisites

Configuration

Background

Meilisearch uses 3 different API keys.

  • A Master API Key
  • An Administrative API key
  • A Search API key

The Master API key should be used when starting Meilisearch,[fn:macos] but should never be used when searching or administering for security reasons.

When you start it with a master key, it will create default Search & Administrative API Keys.[fn:default_keys] These mongodb_meilisearch uses these for searching & administering respectively.

Details

MEILI_MASTER_KEY is the environment variable Meilisearch uses for setting its master key at startup if it isn’t provided via command line arguments.

MEILISEARCH_SEARCH_KEY & MEILISEARCH_ADMIN_KEY are for Search & administration respectively. Administration is everything that isn’t search, including things like adding new records to the index.

I recommend you define the Search & Administration keys. If you provide both of those the MEILI_MASTER_KEY will not be required, or used if present. If you only know the master key, there are details below for how to obtain the others.

Define the following variables in your environment (or .env file if you’re using dotenv). The url below is the default one Meilisearch uses when run locally. Change it if you’re running it on a different box or have otherwise changed the default location setup.

SEARCH_ENABLED=true
MEILI_MASTER_KEY=<your search master key>
MEILISEARCH_SEARCH_KEY=<your search api key here>
MEILISEARCH_ADMIN_KEY=<your admin api key here>
MEILISEARCH_URL=http://127.0.0.1:7700

Optional configuration

MEILISEARCH_TIMEOUT=10
MEILISEARCH_MAX_RETRIES=2

👋 Environment variables starting with MEILI_ are ones defined by Meilisearch. The ones starting with MEILISEARCH_ are specific to this library.

Finding your Search & Admin keys

Meilisearch doesn’t make it easy to find out what your default search & administration API keys are, and I haven’t found a way to define them yourself.

Fortunately for you, this library has an easy way, and you’ve got this library. 😉

Search::Client.instance.get_default_keys("your master key here")

This will return a hash with the keys you’ll need.

{:search=>"your search api key",
 :admin=>"your admin api key"}

Use those keys in your environment. The next time the Search::Client is initialized it’ll use them.

Validating Search & Admin keys

When switching between Meilisearch instances the same master key doesn’t result in the same default admin and search keys. Being able to easily validate this is useful in situations like testing code against Meilisearch instances that are inside a Docker container and in the Host OS.

Practically speaking you may want to simply not provide these keys during development, and only use the MEILI_MASTER_KEY. Just be sure to include them when talking to a production instance.

Search::Client.instance.validate_default_keys(ENV['MEILI_MASTER_KEY'])

This returns a hash like the following.

  • status is always either "provided" or "missing" and indicates if the MEILISEARCH_SEARCH_KEY or MEILISEARCH_ADMIN_KEY were found in the environment.
  • matches is always either true or false, and indicates if the provided Search & Admin keys (if any) match the defaults the current instance is returning.
{
  :search_key=>{
   :status=>"provided",
   :matches=>true
  },
  :admin_key=>{
    :status=>"provided",
    :matches=>true
  }
}

Model Integration

Add the following near the top of your model. Only the extend and include lines are required. This assumes your model also includes Mongoid::Document

include Search::InstanceMethods
extend Search::ClassMethods

If you want Rails to automatically add, update, and delete records from the index, add the following to your model.

You can override these methods if needed, but you’re unlikely to want to.

# enabled?() is controlled by the SEARCH_ENABLED environment variable
if Search::Client.instance.enabled?
  after_create  :add_to_search
  after_update  :update_in_search
  after_destroy :remove_from_search
end

Assuming you’ve done the above a new index will be created with a name that corresponds to your model’s name, only in snake case. All of your models fields will be indexed and filterable.

Example Rails Model

Here’s what it looks like when you put it all together in a Rails model with the default behavior.

class Person
  include Mongoid::Document
  extend Search::ClassMethods

  if Search::Client.instance.enabled?
    after_create  :add_to_search
    after_update  :update_in_search
    after_destroy :remove_from_search
  end

  # normal Mongoid attributes
  field :name, type: String
  field :description, type: String
  field :age, type: Integer
end

Note that that unless you configure it otherwise the ids of belongs_to objects will not be searchable. This is because they’re random strings that no human’s ever going to be searching for, and we don’t want to waste RAM or storage.

Going Beyond The Defaults

This module strives for sensible defaults, but you can override them with the following optional constants:

  • PRIMARY_SEARCH_KEY - a Symbol matching one of your model’s attributes that is guaranteed unique. This defaults to _id
  • SEARCH_INDEX_NAME - a String - useful if you want to have records from multiple classes come back in the same search results. This defaults to the underscored form of the current class name.
  • SEARCH_OPTIONS - a hash of key value pairs in JS style
    {
      attributesToCrop: ['title'],
      cropLength: 2,
      filter: 'genre = adventure',
      attributesToHighlight: ['title'],
      limit: 2
    }
        
  • SEARCH_RANKING_RULES - an array of strings that correspond to meilisearch rules see meilisearch ranking rules docs You probably don’t want to change this.

Indexes

Searching is limited to records that have been added to a given index. This means, if you want to perform one search and get back records from multiple models you’ll need to add them to the same index.

In order to do that add the SEARCH_INDEX_NAME constant to the model whose search stuff you want to end up in the same index. You can name this just about anything. The important thing is that all the models that share this index have the same SEARCH_INDEX_NAME constant defined. You may want to just add it to a module they all import.

SEARCH_INDEX_NAME='general_search'

If multiple models are using the same index, you should also add CLASS_PREFIXED_SEARCH_IDS=true. This causes the id field to be <ClassName>_<_id> For example, a Note record might have an index of "Note_64274543906b1d7d02c1fcc6". If undefined this will default to false. This is not needed if you can absolutely guarantee that there will be no overlap in ids amongst all the models using a shared index.

CLASS_PREFIXED_SEARCH_IDS=true

Setting CLASS_PREFIXED_SEARCH_IDS to true will also cause the original Mongoid _id field to be indexed as original_document_id. This is useful if you want to be able to retrieve the original record from the database.

Searchable Data

You probably don’t want to index all the fields. For example, unless you intend to allow users to sort by when a record was created, there’s no point in recording it’s created_at in the search index. It’ll just waste bandwidth, memory, and disk space.

Define a SEARCHABLE_ATTRIBUTES constant with an array of strings to limit things. These are the field names, and/or names of methods you wish to have indexed.

By default these will also be the fields you can filter on.

Note that Meilisearch requires there to be an id field and it must be a string. If you don’t define one it will use string version of the _id your document’s BSON::ObjectId.

# explicitly define the fields you want to be searchable
# this should be an array of symbols
SEARCHABLE_ATTRIBUTES = %w[title body]
# OR explicitly define the fields you DON'T want searchable 
SEARCHABLE_ATTRIBUTES = searchable_attributes - [:created_at]

Including Foreign Key data

If, for example, your Person belongs_to: group and you wanted that group’s id to be searchable you would include group_id in the list.

If you don’t specify any SEARCHABLE_ATTRIBUTES, the default list will exclude any fields that are Mongoid::Fields::ForeignKey objects.

Getting Extra Specific

If your searchable data needs to by dynamically generated instead of just taken directly from the Mongoid::Document’s attributes or existing methods you can define a search_indexable_hash method on your class.

Before you do, please note that as of v1.1 your SEARCHABLE_ATTRIBUTES constant can contain fields and method names in its array of values. Making a method for each thing dynamically generated thing you want in the search and then including it in SEARCHABLE_ATTRIBUTES is going to be the easiest way of accomplishing this.

Your search_indexable_hash must return a hash, and that hash must include the following keys: - "id" - a string that uniquely identifies the record - "object_class" the name of the class that this record corresponds to.

The value of "object_class" is usually just self.class.name. This is something specific to this gem, and not Meilisearch itself.

See InstanceMethods#search_indexable_hash for an example.

Filterable Fields

If you’d like to only be able to filter on a subset of those then you can define FILTERABLE_ATTRIBUTE_NAMES but it must be a subset of SEARCHABLE_ATTRIBUTES. This is enforced by the gem to guarantee no complaints from Meilisearch. These must be symbols.

If you have no direct need for filterable results, set UNFILTERABLE_IN_SEARCH=true in your model. This will save on index size and speed up indexing, but you won’t be able to filter search results, and that’s half of what makes Meilisearch so great. It should be noted, that even if this is set to true this gem will still add "object_class" as a filterable attribute.

This is the magic that allows you to have an index shared by multiple models and still be able to retrieve results specifically for one.

If you decide to re-enable filtering you can remove that constant, or set it to false. Then call the following. If FILTERABLE_ATTRIBUTE_NAMES is defined it will use that, otherwise it will use whatever .searchable_attributes returns.

MyModel.set_filterable_attributes! # synchronous 
MyModel.set_filterable_attributes  # asynchronous

This will cause Meilisearch to reindex all the records for that index. If you have a large number of records this could take a while. Consider running it on a background thread. Note that filtering is managed at the index level, not the individual record level. By setting filterable attributes you’re giving Meilisearch guidance on what to do when indexing your data.

Note that you will encounter problems in a shared index if you try and filter on a field that one of the contributing models doesn’t have set as a filterable field, or doesn’t have at all.

Sortable Fields

Sortable fields work in essentially the same way as filterable fields. By default it’s the same as your FILTERABLE_ATTRIBUTE_NAMES which, in turn, defaults to your SEARCHABLE_ATTRIBUTES You can override it by setting SORTABLE_ATTRIBUTE_NAMES.

Note that you will encounter problems in a shared index if you try and sort on a field that one of the contributing models doesn’t have set as a sortable field, or doesn’t have at all.

MyModel.set_sortable_attributes! # synchronous 
MyModel.set_sortable_attributes  # asynchronous

Indexing things

Important note: By default anything you do that updates the search index (adding, removing, or changing) happens asynchronously.

Sometimes, especially when debugging something on the console, you want to update the index synchronously. The convention used in this codebase - and in the meilisearch-ruby library we build on - is that the synchronous methods are the ones with the bang. Similar to how mutating state is potentially dangerous and noted with a bang, using synchronous methods is potentially problematic for your users, and thus noted with a bang.

For example:

MyModel.reindex  # runs asyncronously

vs

MyModel.reindex! # runs synchronously

Reindexing, Adding, Updating, and Deleting

Reindexing Calling MyModel.reindex! deletes all the existing records from the current index, and then reindexes all the records for the current model. It’s safe to run this even if there aren’t any records. In addition to re-indexing your models, it will update/set the “sortable” and “filterable” fields on the relevant indexes.

Note: reindexing behaves slightly differently than all the other methods. It runs semi-asynchronously by default. The Asynchronous form will first, attempt to synchronously delete all the records from the index. If that fails an exception will be raised. Otherwise you’d think everything was fine when actually it had failed miserably. If you call .reindex! it will be entirely synchronous.

Note: adding, updating, and deleting should happen automatically if you’ve defined after_create, after_update, and after_destroy as instructed above. You’ll mostly only want to use these when manually mucking with things in the console.

Adding Be careful to not add documents that are already in the index.

  • Add everything: MyClass.add_all_to_search
  • Add a specific instance: my_instance.add_to_search
  • Add a specific subset of documents: MyClass.add_documents(documents_hashes) IMPORTANT: documents_hashes must be an array of hashes that were each generated via search_indexable_hash

Updating

  • Update everything: call reindex
  • Update a specific instance: my_instance.update_in_search
  • Update a specific subset of documents: MyClass.update_documents(documents_hashes) IMPORTANT: documents_hashes must be an array of hashes that were generated via search_indexable_hash The PRIMARY_SEARCH_KEY (_id by default) will be used to find records in the index to update.

Deleting

  • Delete everything: MyClass.delete_all_documents!
  • Delete a specific record: my_instance.remove_from_search
  • Delete the index: MyClass.delete_index! WARNING: if you think you should use this, you’re probably mistaken.

Indexes

By default every model gets its own search index. This means that Foo.search("some text") will only search Foo objects. To have a search cross objects you’ll need to use a “Shared Index” (see below).

The name of the index isn’t important when not using shared indexes. By default a model’s index is the snake cased form of the class name. For example, data for MyWidget models will be stored in the my_widget index.

Shared indexes

Imagine you have a Note and a Comment model, sharing an index so that you can perform a single search and have search results for both models that are ranked by relevance.

In this case both models would define a SEARCH_INDEX_NAME constant with the same value. You might want to just put this, and the other search stuff in a common module that they all include.

Then, when you search you can say Note.search("search term") and it will only bring back results for Note records. If you want to include results that match Comment records too, you can set the optional filtered_by_class parameter to false.

For example: Note.search("search term", filtered_by_class: false) will return all matching Note results, as well as results for all the other models that share the same index as Note.

⚠ Models sharing the same index must share the same primary key field as well. This is a known limitation of the system.

Searching

To get a list of all the matching objects in the order returned by the search engine run MyModel.search("search term") Note that this will restrict the results to records generated by the model you’re calling this on. If you have an index that contains data from multiple models and wish to include all of them in the results pass in the optional filtered_by_class parameter with a false value. E.g. MyModel.search("search term", filtered_by_class: false)

Searching returns a hash, with the class name of the results as the key and an array of String ids, or Mongoid::Document objects as the value. By default it assumes you want Mongoid::Document objects. The returned hash also includes a key of "search_result_metadata" which includes the metadata provided by Meilisearch regarding your request. You’ll need this for pagination if you have lots of results. To exclude the metadata pass include_metadata: false as an option. E.g. MyModel.search("search term", include_metadata: false)

Useful Keyword Parameters

  • ids_only
    • only return matching ids. These will be an array under the "matches" key.
    • defaults to false
  • filtered_by_class
    • limit results to the class you initiated the search from. E.g. Note.search("foo") will only return results from the Note class even if there are records from other classes in the same index.
    • defaults to true
  • include_metadata
    • include the metadata about the search results provided by Meilisearch. If true (default) there will be a "search_result_metadata" key, with a hash of the Meilisearch metadata.
    • You’ll likely need this in order to support pagination, however if you just want to return a single page worth of data, you can set this to false to discard it.
    • defaults to true

Example Search Results

Search results, ids only, for a class where CLASS_PREFIXED_SEARCH_IDS=false.

Note.search('foo', ids_only: true) # => returns 
{ 
  "matches" =>  [
    "64274a5d906b1d7d02c1fcc7",
    "643f5e1c906b1d60f9763071",
    "64483e63906b1d84f149717a"
  ],
  "search_result_metadata" => {
          "query"=>query_string, 
          "processingTimeMs"=>1, 
          "limit"=>50,
          "offset"=>0, 
          "estimatedTotalHits"=>33, 
          "nbHits"=>33
  }
}

If CLASS_PREFIXED_SEARCH_IDS=true the above would have ids like "Note_64274a5d906b1d7d02c1fcc7"

Without ids_only you get full objects in a matches array.

Note.search('foo') # or Note.search('foo', ids_only: false) # => returns 
{ 
  "matches" => [
    #<Note _id: 64274a5d906b1d7d02c1fcc7, created_at: 2023-03-15 00:00:00 UTC, updated_at: 2023-03-31 21:02:21.108 UTC, title: "A note from the past", body: "a body", type: "misc", context: "dachary">,
    #<Note _id: 643f5e1c906b1d60f9763071, created_at: 2023-04-18 00:00:00 UTC, updated_at: 2023-04-19 03:21:00.41 UTC, title: "offline standup ", body: "onother body", type: "misc", context: "WORK">,
    #<Note _id: 64483e63906b1d84f149717a, created_at: 2023-04-25 00:00:00 UTC, updated_at: 2023-04-26 11:23:38.125 UTC, title: "Standup Notes (for wed)", body: "very full bodied", type: "misc", context: "WORK">
  ],
  "search_result_metadata" => {
          "query"=>query_string, "processingTimeMs"=>1, "limit"=>50,
          "offset"=>0, "estimatedTotalHits"=>33, "nbHits"=>33
  }
}

If Note records shared an index with Task and they both had CLASS_PREFIXED_SEARCH_ID=true you’d get a result like this.

Note.search('foo') #=> returns 
{ 
  "matches" => [
      #<Note _id: 64274a5d906b1d7d02c1fcc7, created_at: 2023-03-15 00:00:00 UTC, updated_at: 2023-03-31 21:02:21.108 UTC, title: "A note from the past", body: "a body", type: "misc", context: "dachary">,
      #<Note _id: 643f5e1c906b1d60f9763071, created_at: 2023-04-18 00:00:00 UTC, updated_at: 2023-04-19 03:21:00.41 UTC, title: "offline standup ", body: "onother body", type: "misc", context: "WORK">,
      #<Task _id: 64483e63906b1d84f149717a, created_at: 2023-04-25 00:00:00 UTC, updated_at: 2023-04-26 11:23:38.125 UTC, title: "Do the thing", body: "very full bodied", type: "misc", context: "WORK">
  ],
  "search_result_metadata" => {
          "query"=>query_string, "processingTimeMs"=>1, "limit"=>50,
          "offset"=>0, "estimatedTotalHits"=>33, "nbHits"=>33
  }
  
}

Custom Search Options

To invoke any of Meilisearch’s custom search options (see their documentation). You can pass them in via an options hash.

MyModel.search("search term", options: <my custom options>)

Currently the Meilisearch-ruby gem can convert keys from snake case to camel case. For example hits_per_page will become hitsPerPage. Meilisearch ultimately wants camel case (camelCase) parameter keys, but meilisearch-ruby wants snake case (snake_case).

Follow Meilisearch’s documentation to see what’s available and what type of options to pass it, but convert them to snake case first. Note that your options keys and values must all be simple JSON values.

If for some reason that still isn’t enough, you can work with the meilisearch-ruby index directly via Search::Client.instance.index(search_index_name)

Pagination

This gem has no specific pagination handling, as there are multiple libraries for handling pagination in Ruby. Here’s an example of how to get started with Pagy.

current_page_number = 1
max_items_per_page = 10

search_results = Note.search('foo')

Pagy.new(
    count: search_results["search_result_metadata"]["nbHits"], 
    page: current_page_number, 
    items: max_items_per_page
)

Upgrading

From v1.x

primary difference between v1.x and v2 is that mongodb_meilisearch no longer uses the Meilisearch master key for security reasons. The only thing it will use it for is to look up the default search API key and / or the default administrative API key.

It now initializes 2 Meilisearch::Client objects: one for searching, one for administration, and all of the internal methods that interact with an index have been rewritten to interact with it via the correct client.

Practically speaking, there are 3 significant differences from v1.x

  1. MEILISEARCH_API_KEY is no longer supported. It has been replaced with the official Meilisearch environment variable for the same purpose: MEILI_MASTER_KEY
  2. The master key is only used when an API key for search and / or administration is not provided. It is recommended that you provide MEILISEARCH_SEARCH_KEY & MEILISEARCH_ADMIN_KEY for those purposes. See “Finding your Search & Admin keys” above for more details.
  3. Search::Client.instance.client has been replaced with 2 clients with specific purposes
    • Search::Client::instance.search_client
    • Search::Client::instance.admin_client
  4. ClassMethods.search_index has been replaced with
    • ClassMethods.searchable_index
    • ClassMethods.administratable_index

Development

To contribute to this gem.

  • Run bundle install to install all the dependencies.
  • run lefthook install to set up lefthook This will do things like make sure the tests still pass, and run rubocop before you commit.
  • Start hacking.
  • Add RSpec tests.
  • Add your name to CONTRIBUTORS.md
  • Make PR.

NOTE: by contributing to this repository you are offering to transfer copyright to the current maintainer of the repository.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and the created tag, and push the .gem file to rubygems.org.

Bug reports and pull requests are welcome on GitHub at https://github.com/masukomi/mongodb_meilisearch. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the code of conduct.

License

The gem is available as open source under the terms of the Server Side Public License. For those unfamiliar, the short version is that if you use it in a server side app you need to share all the code for that app and its infrastructure. It’s like AGPL on steroids. Commercial licenses are available if you want to use this in a commercial setting but not share all your source.

Code of Conduct

Everyone interacting in this project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.

It is furthermore fully acknowledged that “master” is a problematic term that shouldn’t be used. However, we’re stuck with it until Meilisearch changes the name of that key.

Footnotes

[fn:macos] MacOS users I have not found a way to successfully automatically Meilisearch with launchctl (brew services uses this) that allows it to see the MEILI_MASTER_KEY environment variable. As such, I can not recommend using launchctl for running Meilisearch. Running it without a master key only works in development mode, and introduces complications.

[fn:default_keys] Using the master key you can access the /keys endpoint in Meilisearch to retrieve the default Search & Administrative API keys