HyperLogLogs

The examples in this section will show you how to use hyperloglogs on their own.

Set Up a Bucket Type

If you've already created and activated a bucket type with the datatype parameter set to hyperloglog, skip to the next section.

Start by creating a bucket type with the datatype parameter set to hyperloglog:

riak_admin bucket-type create hlls '{"props":{"datatype":"hll"}}'

note

The hlls bucket type name provided above is an example and is not required to be hlls. You are free to name bucket types whatever you like, with the exception of default.

After creating a bucket with a Riak data type, confirm the bucket property configuration associated with that type is correct:

riak-admin bucket-type status hlls

This returns a list of bucket properties and their values in the form of property: value.

If our hlls bucket type has been set properly we should see the following pair in our console output:

datatype: hll

Once we have confirmed the bucket type is properly configured, we can activate the bucket type to be used in Riak KV:

riak-admin bucket-type activate hlls

We can check if activation has been successful by using the same bucket-type status command shown above:

riak-admin bucket-type status hlls

After creating and activating our new hlls bucket type, we can setup our client to start using the bucket type as detailed in the next section.

Client Setup

First, we need to direct our client to the bucket type/bucket/key location that contains our counter.

For this example we'll use the hlls bucket type created and activated above and a bucket called hlls:

Erlang

%% Buckets are simply named binaries in the Erlang client. See the
%% examples below for more information

Java

// In the Java client, a bucket/bucket type combination is specified
// using a Namespace object. To specify bucket, bucket type, and key,
// use a Location object that incorporates the Namespace object, as is
// done below.

Location hllLocation =
  new Location(new Namespace("<bucket_type>", "<bucket>"), "<key>");

Python

bucket_type = client.bucket_type('hlls')
bucket = bucket_type.bucket('my_hlls')
hll = bucket.new(key)

# or

from riak.datatypes import Hll
hll = Hll(bucket, key)

Go

// Buckets and bucket types are simply strings in the Go client.

// See the examples below for more information, or the full example at
// https://github.com/basho/riak-go-client/blob/master/examples/dev/using/data-types/hyperloglog.go

// We will need the follow imports to run the examples:
import (
	"fmt"
	"os"
	"time"

	riak "github.com/basho/riak-go-client"
	"errors"
)

C#

// In the C# client, buckets are just string parameters to operations.
// See the examples below for more information.

JS

// In the Node.js client, buckets are just string parameters to operations.
// See the examples below for more information.

PHP

$command = (new Command\Builder\FetchHll($riak_client))
    ->buildLocation('<key>', '<bucket>', 'hlls')
    ->build();

Ruby

bucket = client.bucket_type('hlls').bucket('my_hlls')

CURL

curl http://localhost:8098/types/<bucket_type>/buckets/<bucket>/datatypes/<key>

# Note that this differs from the URL structure for non-Data-Type
# requests, which end in /keys/<key>

Create a HyperLogLog data type

To create a hyperloglog data structure, you need to specify a bucket/key pair to hold that hyperloglog. Here is the general syntax for doing so:

Erlang

HLL = riakc_hll:new().

%% Hyperloglogs in the Erlang client are opaque data structures that
%% collect operations as you mutate them. We will associate the data
%% structure with a bucket type, bucket, and key later on.

Java

// In the Java client, you specify the location of Data Types
// before you perform operations on them:

Location hllLocation =
  new Location(new Namespace("hlls", "hello"), "darkness");

// In the Java client, there is no intermediate "empty" hyperloglog data type.
// Hyperloglogs can be created when an element is added to them, as in the examples below.

Python

bucket_type = client.bucket_type('hlls')
bucket = bucket_type.bucket('my_hlls')
hll = bucket.new(key)

# or

from riak.datatypes import Hll
hll = Hll(bucket, key)

Go

// In the Go client, there is no intermediate "empty" hyperloglog data type.
// Hyperloglogs can be created when an element is added to them, as in the examples below.

C#

// In the C# client, there is no intermediate "empty" hyperloglog data type.
// Hyperloglogs can be created when an element is added to them, as in the examples below.

JS

// In the Node.js client, there is no intermediate "empty" hyperloglog data type.
// Hyperloglogs can be created when an element is added to them, as in the examples below.

PHP

// Note that "hlls" is just an example HLL bucket type name used
// in these examples

$command = (new Command\Builder\UpdateHll($riak_client))
    ->add('gosabres poked you.')
    ->add('phprocks viewed your profile.')
    ->add('phprocks started following you.')
    ->buildBucket('<bucket>', 'hlls')
    ->build();

$response = $command->execute();

Ruby

key = "darkness"
hll = Riak::Crdt::HyperLogLog.new(bucket, key)

CURL

# You cannot create an empty hyperloglog data structure through the HTTP
# interface.
# Hyperloglogs can only be created when an element is added to them, as in the
# examples below.

Upon creation, our hyperloglog data structure is empty:

Erlang

HLL.

%% which will return:
%% {hll,0,[]}

Java

FetchHll fetch = new FetchHll.Builder(hllLocation)
    .build();
RiakHll hll = client.execute(fetch);
boolean isEmpty = hll.getCardinality() == 0;

Python

is_empty = hll.value == 0

Go

var resp *riak.FetchHllResponse

builder := riak.NewFetchHllCommandBuilder()
cmd, err := builder.WithBucketType("hlls").
  WithBucket("hello").
  WithKey("darkness").
  Build()
if err != nil {
  return err
}
if err = cluster.Execute(cmd); err != nil {
  return err
}
if fc, ok := cmd.(*riak.FetchHllCommand); ok {
  if fc.Response == nil {
    return errors.New("expected non-nil Response")
  }
  resp = fc.Response
}

fmt.Println("Hyperloglog cardinality: ", resp.Cardinality)
fmt.Println("Hyperloglog isNotFound: ", resp.IsNotFound)
return nil

JS

var options = {
    bucketType: 'hlls',
    bucket: 'hello',
    key: 'darkness'
};

client.fetchHll(options, function (err, rslt) {
    if (err) {
        throw new Error(err);
    }

    if (rslt.notFound) {
        logger.info("Not Found");
    }
});
// Prints "Not Found" to logger.info.

C#

 var fetch = new FetchHll.Builder()
                    .WithBucketType("hlls")
                    .WithBucket("hello")
                    .WithKey("darkness")
                    .Build();

RiakResult rslt = client.Execute(fetch);
HllResponse response = fetch.Response;
if (response.NotFound)
{
    Console.WriteLine("Not Found");
}
// Prints "Not Found" to the console.

PHP

$command = (new Command\Builder\FetchHll($riak_client))
    ->buildLocation('darkness', 'hello', 'hlls')
    ->build();

$response = $command->execute();

$response->getCode() == '404';

Ruby

puts hll.cardinality
# Prints "0"

CURL

curl http://localhost:8098/types/hlls/buckets/hello/datatypes/darkness

# Response
{"type":"hll","error":"notfound"}

Add elements to a HyperLogLog data type

Erlang

HLL1 = riakc_hll:add_element(<<"Jokes">>, HLL),
RepeatHLL1 = riakc_hll:add_element(<<"Jokes">>, HLL),
HLL2 = riakc_hll:add_elements([<<"Are">>, <<"Better">>, <<"Explained">>], HLL1),

HLL2.

%% which will return:
%% {hll,0,[<<"Are">>,<<"Better">>,<<"Explained">>, <<"Jokes">>]}

Java

HllUpdate hllUpdate = new HllUpdate()
                        .add("Jokes")
                        .add("Are")
                        .addAll(Arrays.asList("Better", "Explained", "Jokes"));

hllUpdate.getElementAdds();
// Returns the set of ["Jokes", "Are", "Better", "Explained"]                     

Python

bucket_type = client.bucket_type('hlls')
bucket = bucket_type.bucket('my_hlls')
myhll = datatypes.Hll(bucket, 'hll_one')
myhll.add('Jokes')
myhll.add('Are')
myhll.add('Better')
myhll.add('Explained')
myhll.add('Jokes')
myhll.store()
# myhll.value == 4

Go

// We will add values in the next example

C#

// We will add values in the next example

JS

// We will add values in the next example

PHP

$command = (new Command\Builder\UpdateHll($riak_client))
    ->add('Jokes')
    ->add('Are')
    ->add('Better')
    ->add('Explained')
    ->add('Jokes')
    ->buildBucket('my_hlls', 'hlls')
    ->build();

$response = $command->execute();

Ruby

CURL

curl -XPOST http://localhost:8098/types/hlls/buckets/hello/datatypes/darkness \
  -H "Content-Type: application/json" \
  -d '{"add_all":["my", "old", "friend"]}'

However, when using a non-HTTP client, the approximate cardinality/value of our data structure will be 0, locally, until its pushed to the server and then fetched from the server.

Erlang

riakc_hll:value(HLL2) == 0.

%% which will return:
%% true

Port = 8087,
{ok, Pid} = riakc_pb_socket:start_link("127.0.0.1", Port),
Key = <<"Holy Diver">>,
BucketType = <<"hlls">>,
Bucket = {BucketType, <<"rainbow in the dark">>},

ok = riakc_pb_socket:update_type(Pid, Bucket, Key, riakc_hll:to_op(HLL2)).
ok = riakc_pb_socket:update_type(Pid, Bucket, Key, riakc_hll:to_op(RepeatHLL1)).

Java

// Using hllUpdate and hllLocation from above examples

UpdateHll update = new UpdateHll.Builder(hllLocation, hllUpdate)
        .build();
client.execute(update);

Python

bucket_type = client.bucket_type('hlls')
bucket = bucket_type.bucket('my_hlls')
myhll = datatypes.Hll(bucket, 'hll_one')
myhll.add('Jokes')
myhll.add('Are')
myhll.add('Better')
myhll.add('Explained')
myhll.add('Jokes')
myhll.store()
# myhll.value == 4

Go

adds := [][]byte{
  []byte("Jokes"),
  []byte("Are"),
  []byte("Better"),
  []byte("Explained"),
  []byte("Jokes"),
}

builder := riak.NewUpdateHllCommandBuilder()
cmd, err := builder.WithBucketType("hlls").
  WithBucket("hello").
  WithKey("darkness").
  WithAdditions(adds...).
  Build()
if err != nil {
  return err
}

return cluster.Execute(cmd)

JS

var options = {
    bucketType: 'hlls',
    bucket: 'hello',
    key: 'darkness',
    additions: ['Jokes', 'Are', 'Better', 'Explained', 'Jokes'],
};

client.updateHll(options, function (err, rslt) {
    if (err) {
        throw new Error(err);
    }
});

C#

var adds = new HashSet<string> { "Jokes", "Are", "Better", "Explained", "Jokes" };

var update = new UpdateHll.Builder(adds)
    .WithBucketType("hlls")
    .WithBucket("hello")
    .WithKey("darkness")
    .WithReturnBody(true)
    .Build();

RiakResult rslt = client.Execute(update);

PHP

$command = (new Command\Builder\UpdateHll($riak_client))
    ->add('Jokes')
    ->add('Are')
    ->add('Better')
    ->add('Explained')
    ->add('Jokes')
    ->buildLocation('darkness', 'hello', 'hlls')
    ->build();

$response = $command->execute();

Ruby

hll.add('Jokes')
hll.batch do |s|
  s.add 'Are'
  s.add 'Better'
  s.add 'Explained'
  s.add 'Jokes'
end

Retrieve a HyperLogLog data type

Now, we can check the approximate count-of (a.k.a. the cardinality of the elements added to) our hyperloglog data structure:

Erlang

{ok, HLL3} = riakc_pb_socket:fetch_type(Pid, Bucket, Key),
riakc_hll:value(HLL3) == 4.

%% which would return:
%% true

%% We added <<"Jokes">> twice, but, remember, the algorithm only counts the
%% unique elements we've added to the data structure.

Java

FetchHll hllFetchCmd = new FetchHll.Builder(location).build();
RiakHll hll = client.execute(hllFetchCmd);
hll.getCardinality();
// Which returns 4

// We added "Jokes" twice, but, remember, the algorithm only counts the
// unique elements we've added to the data structure.

Python

bucket_type = client.bucket_type('hlls')
bucket = bucket_type.bucket('my_hlls')
myhll = bucket.get('hll_one')
# myhll.value == 4

Go

var resp *riak.FetchHllResponse

builder := riak.NewFetchHllCommandBuilder()
cmd, err := builder.WithBucketType("hlls").
  WithBucket("hello").
  WithKey("darkness").
  Build()
if err != nil {
  return err
}
if err = cluster.Execute(cmd); err != nil {
  return err
}
if fc, ok := cmd.(*riak.FetchHllCommand); ok {
  if fc.Response == nil {
    return errors.New("expected non-nil Response")
  }
  resp = fc.Response
}

// We added "Jokes" twice, but, remember, the algorithm only counts the
// unique elements we've added to the data structure.
fmt.Println("Hyperloglog cardinality: ", resp.Cardinality)
return nil

JS

var options = {
    bucketType: 'hlls',
    bucket: 'hello',
    key: 'darkness'
};

client.fetchHll(options, function (err, rslt) {
    if (err) {
        throw new Error(err);
    }

    if (rslt.notFound) {
        logger.info("Not Found");
    }
    logger.info("Hyperloglog cardinality is: " + rslt.cardinality);
});
// Prints "Hyperloglog cardinality is: 4"
// We added "Jokes" twice, but, remember, the algorithm only counts the
// unique elements we've added to the data structure.

C#

var fetch = new FetchHll.Builder()
                    .WithBucketType("hlls")
                    .WithBucket("hello")
                    .WithKey("darkness")
                    .Build();

RiakResult rslt = client.Execute(fetch);
Assert.IsTrue(rslt.IsSuccess, rslt.ErrorMessage);

HllResponse response = fetch.Response;
if (response.NotFound)
{
    Console.WriteLine("Not Found");
}
else
{
    Console.WriteLine("Hyperloglog cardinality is: " + response.Cardinality);  
}

// Prints "Hyperloglog cardinality is: 4"
// We added "Jokes" twice, but, remember, the algorithm only counts the
// unique elements we've added to the data structure.

PHP

$command = (new Command\Builder\FetchHll($riak_client))
    ->buildLocation('darkness', 'hello', 'hlls')
    ->build();

$result = $command->execute();

// Note: as though we are in a PHP unit test
$this->assertTrue(is_int($response->getHll()->getData()));
$this->assertEquals(4, $response->getHll()->getData());

// We added "Jokes" twice, but, remember, the algorithm only counts the
// unique elements we've added to the data structure.

Ruby

puts hll.cardinality
# Prints "4"

CURL

curl http://localhost:8098/types/hlls/buckets/hello/datatypes/darkness

# Response
{"type":"hll","value":"4"}