Riak KV Key/Value Modeling

While Riak enables you to take advantage of a wide variety of features that can be useful in application development secondary indexes (2i), and Riak Data Types, Riak almost always performs best when you build your application around basic CRUD operations (create, read, update, and delete) on objects, i.e. when you use Riak as a “pure” key/value store.

In this tutorial, we’ll suggest some strategies for naming and modeling for key/value object interactions with Riak. If you’d like to use some of Riak’s other features, we recommend checking out the documentation for each of them or consulting our guide to building applications with Riak for a better sense of which features you might need.

Advantages of Key/Value Operations

Riak’s key/value architecture enables it to be more performant than relational databases in many scenarios because Riak doesn’t need to perform lock, join, union, or other operations when working with objects. Instead, it interacts with objects on a one-by-one basis, using primary key lookups.

Primary key lookups store and fetch objects in Riak on the basis of three basic locators:

The object’s key, which can be anything you want as long as it is Unicode compliant
The bucket which houses the object and its key (bucket names are also Unicode compliant)
The bucket type that determines the bucket’s replication and other properties

It may be useful to think of this system as analogous to a nested key/value hash as you would find in most programming languages. Below is an example from Ruby. The hash simpsons contains keys for all of the available seasons, while each key houses a hash for each episode of that season:

simpsons = {
  'season 1': {
    { 'episode 1': 'Simpsons Roasting on an Open Fire' },
    { 'episode 2': 'Bart the Genius' },
    # ...
  },
  'season 2': {
    { 'episode 1': 'Bart Gets an "F"' },
    # ...
  },
  # ...
}

If we want to find out the title of an episode, we can retrieve it based on hash keys:

simpsons['season 4']['episode 12']

# => "Marge vs. the Monorail"

Storing data in Riak is a lot like this. Let’s say that we want to store JSON objects with a variety of information about every episode of the Simpsons. We could store each season in its own bucket and each episode in its own key within that bucket. Here’s what the URL structure would look like (for the HTTP API):

GET/PUT/DELETE /bucket/<season>/keys/<episode number>

The most important benefit of sorting Riak objects this way is that these types of lookup operations are extremely fast. Riak doesn’t need to search through columns or tables to find an object. If it knows the bucket/key “address” of the object, so to speak, it can locate that object just about as quickly with billions of objects in a cluster as when the cluster holds only a handful of objects.

Overcoming the Limitations of Key/Value Operations

Using any key/value store can be tricky at first, especially if you’re used to relational databases. The central difficulty is that your application cannot run arbitrary selection queries like SELECT * FROM table, and so it needs to know where to look for objects in advance.

One of the best ways to enable applications to discover objects in Riak more easily is to provide structured bucket and key names for objects. This approach often involves wrapping information about the object in the object’s location data itself.

Here are some example sources for bucket or key names:

Timestamps, e.g. 2013-11-05T08:15:30-05:00
UUIDs, e.g. 9b1899b5-eb8c-47e4-83c9-2c62f0300596
Geographical coordinates, e.g. 40.172N-21.273E

We could use these markers by themselves or in combination with other markers. For example, sensor data keys could be prefaced by sensor_ or temp_sensor1_ followed by a timestamp (e.g. sensor1_2013-11-05T08:15:30-05:00), or user data keys could be prefaced with user_ followed by a UUID (e.g. user_9b1899b5-eb8c-47e4-83c9-2c62f0300596).

Any of the above suggestions could apply to bucket names as well as key names. If you were building Twitter using Riak, for example, you could store tweets from each user in a different bucket and then construct key names using a combination of the prefix tweet_ and then a timestamp. In that case, all the tweets from the user BashoWhisperer123 could be housed in a bucket named BashoWhisperer123, and keys for tweets would look like tweet_<timestamp>.

The possibilities are essentially endless and, as always, defined by the use case at hand.

Object Discovery with Riak Sets

Let’s say that we’ve created a solid bucket/key naming scheme for a user information store that enables your application to easily fetch user records, which are all stored in the bucket users with each user’s username acting as the key. The problem at this point is this: how can Riak know which user records actually exist?

One way to determine this is to list all keys in the bucket users. This approach, however, is not recommended, because listing all keys in a bucket is a very expensive operation that should not be used in production. And so another strategy must be employed.

A better possibility is to use Riak sets to store lists of keys in a bucket. Riak sets are a Riak Data Type that enable you to store lists of binaries or strings in Riak. Unlike normal Riak objects, you can interact with Riak sets much like you interact with sets in most programming languages, i.e. you can add and remove elements at will.

Going back to our user data example, instead of simply storing user records in our users bucket, we could set up our application to store each key in a set when a new record is created. We’ll store this set in the bucket user_info_sets (we’ll keep it simple) and in the key usernames. The following will also assume that we’ve set up a bucket type called sets.

We can interact with that set on the basis of its location:

Location userIdSet = new Location(new Namespace("sets", "user_info_sets"), "usernames");

// With this Location, we can construct fetch operations like this:
FetchSet fetchUserIdSet = new FetchSet.Builder(userIdSet).build();

require 'riak'

set_bucket = client.bucket('user_info_sets')

# We'll make this set global because we'll use it
# inside of a function later on

$user_id_set = Riak::Crdt::Set.new(set_bucket, 'usernames', 'sets')

$command = (new \Basho\Riak\Command\Builder\FetchSet($riak))
    ->buildLocation('usernames', 'user_info_sets', 'sets')
    ->build();

from riak.datatypes import Set

bucket = client.bucket_type('sets').bucket('user_info_sets')
user_id_set = Set(bucket, 'usernames')

Getting started with Riak clients

If you are connecting to Riak using one of Basho’s official client libraries, you can find more information about getting started with your client in Developing with Riak KV: Getting Started.

Then, we can create a function that stores a user record’s key in that set every time a record is created:

// A User class for constructing user records
class User {
  public String username;
  public String info;

  public User(String username, String info) {
    this.username = username;
    this.info = info;
  }
}

// A function for storing a user record that has been created
public void storeUserRecord(User user) throws Exception {
  // User records themselves will be stored in the bucket "users"
  Location userObjectLocation =
    new Location(new Namespace("users"), user.username);
  RiakObject userObject = new RiakObject()
      // We'll keep it simple and store User object data as plain text
      .setContentType("text/plain")
      .setValue(user.info);
  StoreValue store = new StoreValue.Builder(userObjectLocation, userObject)
      .build();
  client.execute(store);

  Location userIdSet =
    new Location(new Namespace("sets", "user_info_sets"), "usernames");
  SetUpdate su = new SetUpdate()
      .add(BinaryValue.create(user.username));
  UpdateSet update = new UpdateSet.Builder(su, update)
      .build();
  client.execute(update);
}

class User
  attr_accessor :username, :info
end

def store_record(user)
  # First we create an empty object and specify its bucket and key
  obj = Riak::RObject.new(client.bucket('users'), user.username)

  # We'll keep it simple by storing plain text for each user's info
  obj.content_type = 'text/plain'
  obj.raw_data = user.info
  obj.store

  # Finally, we'll add the user's username to the set
  user_id_set.add(user.username)
end

class User
{
  public $user_name;
  public $info;

  public function __construct($user_name, $info)
  {
    $this->user_name = $user_name;
    $this->info = $info;
  }
}

function store_user(User $user)
{
  (new \Basho\Riak\Command\Builder\StoreObject)
    ->buildLocation($user->user_name, 'users')
    ->buildJsonObject($user)
    ->build()
    ->execute();

  (new \Basho\Riak\Command\Builder\UpdateSet)
    ->buildLocation('usernames', 'user_info_sets', 'sets')
    ->add($user->user_name)
    ->build()
    ->execute();
}

class User:
    def __init__(self, username, info):
        this.username = username
        this.info = info

# Using the "user_id_set" object from above
def store_record(user):
  # First we create an empty object and specify its bucket and key
    obj = RiakObject(client, 'users', user.username)

    # We'll keep it simple by storing plain text for each user's info
    obj.content_type = 'text/plain'
    obj.data = user.info
    obj.store()

    # Finally, we'll add the user's username to the set
    user_id_set.add(username)
    user_id_set.store()

Now, let’s say that we want to be able to pull up all user records in the bucket at once. We could do so by iterating through the usernames stored in our set and then fetching the object corresponding to each username:

public Set<User> fetchAllUserRecords() {
    // Empty builder sets for usernames and User objects
    Set<String> userIdSet = new HashSet<String>();
    Set<User> userSet = new HashSet<User>();

    // Turn the Riak username set into a set of Strings
    Location userIdSet =
        new Location(new Namespace("sets", "sets"), "usernames");
    FetchSet fetchUserIdSet = new FetchSet.Builder(userIdSet).build();
    RiakSet set = client.execute(fetchUserIdSet).getDatatype();
    set.viewAsSet().forEach((BinaryValue username) -> {
        userIdSet.add(username.toString());
    });

    // Fetch User objects for each of the usernames stored in the set
    userIdSet.forEach((String username) -> {
        Location userLocation = new Location(new Namespace("users"), username);
        FetchValue fetch = new FetchValue.Builder(userLocation).build();
        User user = client.execute(fetch).getValue(User.class);
        userSet.add(user);
    });
    return userSet;
}

# Using the "user_id_set" set from above

def fetch_all_user_records
  users_bucket = $client.bucket('users')
  user_records = Array.new
  $user_id_set.members.each do |user_id|
    user_record = users_bucket.get(user_id).data
    user_records.push(user_record)
  end
  user_records
end

function fetch_users()
{
  $users = [];

  $response = (new \Basho\Riak\Command\Builder\UpdateSet)
    ->buildLocation('usernames', 'user_info_sets', 'sets')
    ->build()
    ->execute();

  $user_names = $response->getSet()->getData();
  foreach($user_names as $user_name) {
    $response = (new \Basho\Riak\Command\Builder\FetchObject)
      ->buildLocation($user_name, 'users')
      ->build()
      ->execute();

    $users[$user_name] = $response->getObject()->getData();
  }

  return $users;
}

# We'll create a generator object that will yield a list of Riak objects
def fetch_all_user_records():
  users_bucket = client.bucket('users')
    user_id_list = list(user_id_set.reload().value)
    for user_id in user_id_list:
      yield users_bucket.get(user_id)

# We can retrieve that list of Riak objects later on
list(fetch_all_user_records())

Naming and Object Verification

Another advantage of structured naming is that you can prevent queries for objects that don’t exist or that don’t conform to how your application has named them. For example, you could store all user data in the bucket users with keys beginning with the fragment user_ followed by a username, e.g. user_coderoshi or user_macintux. If an object with an inappropriate key is stored in that bucket, it won’t even be seen by your application because it will only ever query keys that begin with user_:

// Assuming that we've created a class User:

public User getUserByUsername(String username) {
    String usernameKey = String.format("user_%s", username)
    Location loc = new Location("users")
            .setKey(usernameKey);
    FetchValue fetchUser = new FetchValue.Builder(loc).build();
    FetchValue.Response res = client.execute(fetchUser);
    User userObject = res.getValue(User.class);
    return userObject;
}

def get_user_by_username(username)
  bucket = client.bucket('users')
  obj = bucket.get('user_#{username}')
  return obj.raw_data
end

function fetchUser($user_name)
{
    $response = (new \Basho\Riak\Command\Builder\FetchObject)
      ->buildLocation($user_name, 'users')
      ->build()
      ->execute();

    return $response->getObject()->getData();
}

def get_user_by_username(username):
  bucket = client.bucket('users')
  obj = bucket.get('user_{}'.format(username))
  return obj.data

Bucket Types as Additional Namespaces

Riak bucket types have two essential functions: they enable you to manage bucket configurations in an efficient and streamlined way and, more importantly for our purposes here, they act as a third namespace in Riak in addition to buckets and keys. Thus, in Riak versions 2.0 and later you have access to a third layer of information for locating objects if you wish.

While bucket types are typically used to assign different bucket properties to groups of buckets, you can also create named bucket types that simply extend Riak’s defaults or multiple bucket types that have the same configuration but have different names.

Here’s an example of creating four bucket types that only extend Riak’s defaults:

riak admin bucket-type create john
riak admin bucket-type create robert
riak admin bucket-type create jimmy
riak admin bucket-type create john-paul

Or you can create five different bucket types that all set n_val to 2 but have different names:

riak admin bucket-type create earth '{"props":{"n_val":2}}'
riak admin bucket-type create fire '{"props":{"n_val":2}}'
riak admin bucket-type create wind '{"props":{"n_val":2}}'
riak admin bucket-type create water '{"props":{"n_val":2}}'
riak admin bucket-type create heart '{"props":{"n_val":2}}'

Bucket Types Example

To extend our Simpsons example from above, imagine that we become dissatisfied with our storage scheme because we want to separate the seasons into good seasons and bad seasons (we’ll leave it up to you to make that determination).

One way to improve our scheme might be to change our bucket naming system and preface each bucket name with good or bad, but a more elegant way would be to use bucket types instead. So instead of this URL structure…

GET/PUT/DELETE /bucket/<season>/keys/<episode number>

…we can use this structure:

GET/PUT/DELETE /types/<good or bad>/buckets/<season>/keys/<episode number>

That adds an additional layer of namespacing and enables us to think about our data in terms of a deeper hash than in the example above:

simpsons = {
  'good': {
    'season X': {
      { 'episode 1': '<title>' },
      # ...
    }
  },
  'bad': {
    'season Y': {
      { 'episode 1': '<title>' },
      # ...
    }
  }
}

We can fetch the title of season 8, episode 6:

# For the sake of example, we'll classify season 8 as good:

simpsons['good']['season 8']['episode 6']

# => "A Milhouse Divided"

If your data is best modeled as a three-layered hash, you may want to consider using bucket types in the way shown above.

Resources

More on key/value modeling in Riak can be found in this presentation by Basho evangelist Hector Castro, with the presentation slides available on Speaker Deck.