1 billion rows challenge in MySQL

Earlier this month I wrote a piece on solving Gunnar Morling interesting 1 billion rows challenge in PostgreSQL and ClickHouse. Since Aiven provides also MySQL, I wanted to give it a try. TLDR; The results are much slower than PG and ClickHouse, do you have any suggestion on how to improve?

Alert: the following is NOT a benchmark! The test is done with default installations of both databases and NO optimization. The blog only shows the technical viability of a solution.

Install MySQL on Mac

The first step is to have an handy MySQL, I could use the Aiven FREE tier, but, as for the previous PostgreSQL and ClickHouse example, decided to install it locally. To do it, you can execute:

brew install mysql

then we can start with

brew services start mysql

Connect to MySQL and setup a database

Once the installation process is completed, we can connect to the local MySQL with the following command:

mysql -u root

For this test we’re going to use the root user. However is highly suggested to create a non root user for any sensible work with a database.

We can now create a database with:

CREATE DATABASE 1brow;

And start using it with:

USE 1brow;

Loading the data in MySQL

In order to properly solve the challenge we need to load the data into MySQL. We can create a table called TEST with the needed columns with:

CREATE TABLE TEST (
  id BIGINT AUTO_INCREMENT PRIMARY KEY,
  city VARCHAR(100),
  temperature FLOAT
);

And populate it with:

LOAD DATA LOCAL INFILE "measurements.txt.new" IGNORE
INTO TABLE TEST
COLUMNS TERMINATED BY ';'
LINES TERMINATED BY '\n'
IGNORE 1 LINES;

If you receive the error ERROR 3948 (42000): Loading local data is disabled; this must be enabled on both the client and server sides, we can enable it with:

SET GLOBAL local_infile=1;

Compared to PostgreSQL and ClickHouse, MySQL doesn’t seem to have a native option to read from an external CSV file. The MySQL CSV Engine is available, but it requires you to move the CSV within MySQL’s data folder.

MySQL Results

Once the data is loaded, we can use a similar query to the PostgreSQL and ClickHouse ones to perform the correct ordering.

WITH AGG AS(
    SELECT city,
           MIN(temperature) min_measure,
           cast(AVG(temperature) AS DECIMAL(8,1)) mean_measure,
           MAX(temperature) max_measure
    FROM test
    GROUP BY city
    )
SELECT GROUP_CONCAT(CITY, '=', CONCAT(min_measure,'/', mean_measure,'/', max_measure) ORDER BY CITY SEPARATOR ', ')
FROM AGG
;

The above query is using GROUP_CONCAT allows to return the concatenated list of cities once min/mean/max measures have been calculated.

MySQL Timing

To get the timing I created a file called test.sql with the entire set of commands:

CREATE TABLE TEST (
  id BIGINT AUTO_INCREMENT PRIMARY KEY,
  city VARCHAR(100),
  temperature float
) ENGINE=MEMORY;

SET GLOBAL local_infile=1;

LOAD DATA LOCAL INFILE "measurements.txt" IGNORE 
INTO TABLE TEST
COLUMNS TERMINATED BY ';'
LINES TERMINATED BY '\n'
IGNORE 1 LINES
(city, temperature);

WITH AGG AS(
    SELECT city,
           MIN(temperature) min_measure,
           cast(AVG(temperature) AS DECIMAL(8,1)) mean_measure,
           MAX(temperature) max_measure
    FROM test
    GROUP BY city
    )
SELECT GROUP_CONCAT(CITY, '=', CONCAT(min_measure,'/', mean_measure,'/', max_measure) ORDER BY CITY SEPARATOR ', ')
FROM AGG
;

And a then timed with:

time mysql --local-infile=1 -u root -vvv 1brow < test.sql

The timing is 43 min 57.99 sec with over 33 min 32.15 sec spent on the ingestion phase and 10 min 25.77 sec on the query. The file test.ibd being over 40GB almost 4x times the original size of the measurements.txt file (13GB). You can check the size of the table file with the following command:

ls -lh /usr/local/var/mysql/1brow/test.idb

Alert: the following is NOT a benchmark! The test is done with default installations of both databases and NO optimization. The blog only shows the technical viability of a solution.

You might need to tweak the connect_timeout parameter in my.cnf to raise the connection timeout to 100 minutes in order to wait for the loading and query process to finish.

Speed up MySQL timing by using the MEMORY Engine

MySQL has a Memory storage engine which will avoid writing to disk and store the entire table in RAM. As per documentation, this is not a safe option for production workloads since:

Because the data is vulnerable to crashes, hardware issues, or power outages, only use these tables as temporary work areas or read-only caches for data pulled from other tables.

The only change we have to perform, in order to use the MEMORY storage engine is to redefine our test table as:

CREATE TABLE TEST (
  id BIGINT AUTO_INCREMENT PRIMARY KEY,
  city VARCHAR(100),
  temperature FLOAT
) ENGINE = MEMORY;

But, when making the change and retrying the script you might hit the ERROR 1114 (HY000) at line 9: The table 'test' is full since the data doesn’t fit the memory allocated to MySQL. To avoid this we can set the following flags to raise the allocated memory to 64GB:

SET max_heap_table_size = 1024 * 1024 * 1024 * 64;

However, also this didn’t work, and stopped my test after waiting for over 40 minutes for just the loading time.

Do you have ideas on how to speed up the process in MySQL? I’m all 👂 on X (ex Twitter) @ftisiot!