HomeMogDBMogDB StackUqbar
v1.1

Documentation:v1.1

Supported Versions:

Other Versions:

HLL

HyperLoglog (HLL) is an approximation algorithm for efficiently counting the number of distinct values in a data set. It features faster computing and lower space usage. You only need to store HLL data structures, instead of data sets. When new data is added to a data set, make hash calculation on the data and insert the result to an HLL. Then, you can obtain the final result based on the HLL.

Table 1 compares HLL with other algorithms.

Table 1 Comparison between HLL and other algorithms

Item Sorting Algorithm Hash Algorithm HLL
Time complexity O(nlogn) O(n) O(n)
Space complexity O(n) O(n) 1280 bytes
Error rate 0 0 ≈2%
Storage space requirement Size of original data Size of original data 1280 bytes

HLL has advantages over others in the computing speed and storage space requirement. In terms of time complexity, the sorting algorithm needs O(nlogn) time for sorting, and the hash algorithm and HLL need O(n) time for full table scanning. In terms of storage space requirements, the sorting algorithm and hash algorithm need to store raw data before collecting statistics, whereas the HLL algorithm needs to store only the HLL data structures rather than the raw data, and thereby occupying a fixed space of only 1280 bytes.

img NOTICE:

  • In default specifications, the maximum number of distinct values is 1.6e plus 12, and the maximum error rate can reach 2.3%. If a calculation result exceeds the maximum number, the error rate of the calculation result will increase, or the calculation will fail and an error will be reported.
  • When using this feature for the first time, you need to evaluate the distinct values of the service, properly select configuration parameters, and perform verification to ensure that the accuracy meets requirements.
  • When default parameter configuration is used, the calculated number of distinct values is 1.6e plus 12. If the calculated result is NaN, you need to adjust log2m and regwidth to accommodate more distinct values.
  • The hash algorithm has an extremely low probability of collision. However, you are still advised to select 2 or 3 hash seeds for verification when using the hash algorithm for the first time. If there is only a small difference between the distinct values, you can select any one of the seeds as the hash seed.

Table 2 describes main HLL data structures.

Table 2 Main HLL data structures

Data Type Description
hll Its size is always 1280 bytes, which can be directly used to calculate the number of distinct values.

The following describes HLL application scenarios.

  • Scenario 1: "Hello World"

    The following example shows how to use the HLL data type:

    -- Create a table with the HLL data type:
    mogdb=# create table helloworld (id integer, set hll);
    
    -- Insert an empty HLL to the table:
    mogdb=# insert into helloworld(id, set) values (1, hll_empty());
    
    -- Add a hashed integer to the HLL:
    mogdb=# update helloworld set set = hll_add(set, hll_hash_integer(12345)) where id = 1;
    
    -- Add a hashed string to the HLL:
    mogdb=# update helloworld set set = hll_add(set, hll_hash_text('hello world')) where id = 1;
    
    -- Obtain the number of distinct values of the HLL:
    mogdb=# select hll_cardinality(set) from helloworld where id = 1;
     hll_cardinality
    -----------------
                   2
    (1 row)
    
    -- Delete the table.
    mogdb=# drop table helloworld;
  • Scenario 2: Collect statistics about website visitors.

    The following example shows how an HLL collects statistics on the number of users visiting a website within a period of time:

    -- Create a raw data table to show that a user has visited the website at a certain time:
    mogdb=# create table facts (
             date            date,
             user_id         integer
    );
    
     -- Create a raw data table to show that a user has visited the website at a certain time:
    mogdb=# insert into facts values ('2019-02-20', generate_series(1,100));
    mogdb=# insert into facts values ('2019-02-21', generate_series(1,200));
    mogdb=# insert into facts values ('2019-02-22', generate_series(1,300));
    mogdb=# insert into facts values ('2019-02-23', generate_series(1,400));
    mogdb=# insert into facts values ('2019-02-24', generate_series(1,500));
    mogdb=# insert into facts values ('2019-02-25', generate_series(1,600));
    mogdb=# insert into facts values ('2019-02-26', generate_series(1,700));
    mogdb=# insert into facts values ('2019-02-27', generate_series(1,800));
    
    -- Create another table and specify an HLL column:
    mogdb=# create table daily_uniques (
        date            date UNIQUE,
        users           hll
    );
    
    -- Group data by date and insert the data into the HLL:
    mogdb=# insert into daily_uniques(date, users)
        select date, hll_add_agg(hll_hash_integer(user_id))
        from facts
        group by 1;
    
    -- Calculate the numbers of users visiting the website every day:
    mogdb=# select date, hll_cardinality(users) from daily_uniques order by date;
            date         | hll_cardinality
    ---------------------+------------------
     2019-02-20 00:00:00 |              100
     2019-02-21 00:00:00 | 203.813355588808
     2019-02-22 00:00:00 | 308.048239950384
     2019-02-23 00:00:00 | 410.529188080374
     2019-02-24 00:00:00 | 513.263875705319
     2019-02-25 00:00:00 | 609.271181107416
     2019-02-26 00:00:00 | 702.941844662509
     2019-02-27 00:00:00 | 792.249946595237
    (8 rows)
    
    -- Calculate the number of users who had visited the website in the week from February 20, 2019 to February 26, 2019:
    mogdb=# select hll_cardinality(hll_union_agg(users)) from daily_uniques where date >= '2019-02-20'::date and date <= '2019-02-26'::date;
     hll_cardinality
    ------------------
     702.941844662509
    (1 row)
    
    -- Calculate the number of users who had visited the website yesterday but have not visited the website today:
    mogdb=# SELECT date, (#hll_union_agg(users) OVER two_days) - #users AS lost_uniques FROM daily_uniques WINDOW two_days AS (ORDER BY date ASC ROWS 1 PRECEDING);
            date         | lost_uniques
    ---------------------+--------------
     2019-02-20 00:00:00 |            0
     2019-02-21 00:00:00 |            0
     2019-02-22 00:00:00 |            0
     2019-02-23 00:00:00 |            0
     2019-02-24 00:00:00 |            0
     2019-02-25 00:00:00 |            0
     2019-02-26 00:00:00 |            0
     2019-02-27 00:00:00 |            0
    (8 rows)
    
    -- Delete the table.
    mogdb=# drop table facts;
    mogdb=# drop table daily_uniques;
  • Scenario 3: The data to be inserted does not meet the requirements of the HLL data structure.

    When inserting data into a column of the HLL type, ensure that the data meets the requirements of the HLL data structure. If the data does not meet the requirements after being parsed, an error will be reported. In the following example, E\\1234 to be inserted does not meet the requirements of the HLL data structure after being parsed. As a result, an error is reported.

    mogdb=# create table test(id integer, set hll);
    mogdb=# insert into test values(1, 'E\\1234');
    ERROR:  unknown schema version 4
    mogdb=# drop table test;
Copyright © 2011-2024 www.enmotech.com All rights reserved.