MySQL Benchmarking

I would like to evaluate the performance of a different MySQL instance by re-executing real queries from log files. I am aware of tools like mysqlslap, that produces random and autogenerated queries, but I would rather like to test the system with realistic queries.

In order to benchmark the new system, I would ideally collect all queries either using the slow log or the general log and then replay the queries to the new system.

It should be possible to filter queries, for instance I would like to replay SELECT statements only. I am wondering if there exists an actively maintained procedure or a best practice that how to achieve warm-up or benchmark functionality.


  • Replay (re-execute) real queries collected from a running MySQL instance
  • Filter statements by statement type (INSERT, UPDATE, DELETE) and DML
  • Get statistics (how long did the query run before, after the migration etc)

Problem: Tools are outdated

Percona offered several tools that seemed to be designed for that purpose. All of them have been removed from the current Percona Toolkit. These tools were:

  • pt-log-player
  • The replay option for pt-query-digest
  • query-playback

Potential solutions

Using a Debian Wheezy docker container gives easy access to an older Percona Toolkit version where the pt-log-player is available. Running a container with docker run -it --network="host" --name wheezy debian:wheezy /bin/bash and installing the legacy versions inside works:

apt-get update
apt-get install percona-toolkit mysql-client

I also tried percona-playback, but it does not seem to be very actively maintained either. That would be a great tool, but it is hardly documented and I could not find a method for filtering SELECT statements only, as I do not want to run other DML statements. I only saw that Percona Server offers a flag for enabling a read only option, but this is not available with vanilla MySQL or AWS RDS.

However I would rather like to use a procedure that does not rely on unmaintained tools.

Imagined Workflow

  1. Enable slow query log or general log
  2. Parse and analyze log file
  3. Filter relevant queries
  4. Re-execute queries
  5. Compare results

Are there any tools how I could achieve this? What is the best way to filter the slow query log? How to replay the queries sanely?

3 Answers 3


I found a solution for my problem. It is a combination of Percona pt-query-digest and Percona playback.

First, I filter the slow log by database name and exclude all statements that do not start with select.

cat mysql-slow.log |  pt-query-digest --filter '(($event->{db}) =~ /my_database/) && $event->{arg} =~ m/^select/i' --output slowlog > my_database_selects.log

As output, I get a new slow log file that only contains the desired queries.

percona-playback --mysql-max-retries 1 --mysql-host example.org --mysql-port 3306 --mysql-username warmup --mysql-password S-E-C-R-E-T --ignore-row-result-diffs --query-log-file ~/my_database_selects.log

As result, I get an overview of the query execution time and a warmed up database.

Detailed Report
SELECTs  : 41666 queries (33379 faster, 8287 slower)
INSERTs  : 0 queries (0 faster, 0 slower)
UPDATEs  : 0 queries (0 faster, 0 slower)
DELETEs  : 0 queries (0 faster, 0 slower)
REPLACEs : 0 queries (0 faster, 0 slower)
DROPs    : 0 queries (0 faster, 0 slower)

Executed 41666 queries
Spent 00:00:45.795945 executing queries versus an expected 00:02:07.966942 time.
33379 queries were quicker than expected, 8287 were slower
A total of 0 queries had errors.
Expected 79870 rows, got 79868 (a difference of 2)
Number of queries where number of rows differed: 0.

Average of 265.39 queries per connection (157 connections).

I hope this helps someone with a similar problem.

Installing Percona Playback

There is a bug in the current version that requires to change a library path.

sudo apt-get install libtbb-dev libmysqlclient-dev libboost-program-options-dev libboost-thread-dev libboost-regex-dev libboost-system-dev libboost-chrono-dev pkg-config cmake  libssl-dev
git clone https://github.com/Percona-Lab/query-playback.git
cd query-playback/

Edit the the file CMakeLists.txt (in the directory ~/git/query-playback/percona_playback/mysql_client/CMakeLists.txt ) and replace find_library(MYSQL_LIB"mysqlclient_r" PATH_SUFFIXES "mysql") with find_library(MYSQL_LIB "mysqlclient" PATH_SUFFIXES "mysql") (remove the _r suffix).

mkdir build_dir
cd build_dir
cmake -DCMAKE_BUILD_TYPE=RelWithDebInfo ..
sudo make install

Here is a link with some background

  • How did you install percona-playback? I tried a bunch of install instructions I found online, and they all seemed to be broken. Feb 28, 2019 at 17:07
  • I wanted to use percona-playback on CentOS 7, Percona Server 8.0.15 to replay general log/slow logs from a MySQL 5.6 Server. Unable to build percona-playback on CentOS 7. Can't get past CMake Error: The following variables are used in this project, but they are set to NOTFOUND. Please set them or make sure they are set and tested correctly in the CMake files: MYSQL_LIB linked by target "mysql_client" in directory /root/query-playback/percona_playback/mysql_client
    – thatsaru
    Jul 9, 2019 at 12:17
  • I edited my answer. It now contains some info how to compile Percona Playback
    – Stefan
    Jul 11, 2019 at 6:12

Since pt-query-digest did not work for me on general log generated for MySQL 5.6 nor MariaDB 10.3, I looked around and found mysql-utilities which is doing simple parsing but expect date/time on each Query. So in the end I simply redid the parsing which is quite simple:

rm -rf /tmp/sql-reqs ; mkdir /tmp/sql-reqs ; perl -lne 'if (/^(\d{6} (\d{1,2}:\d{2}:\d{2})|\t)\t *(\d+) (\w+)\t(.*)/) { open $F, ">>", "/tmp/sql-reqs/$thread.sql"; print $F "$arg;" if $type eq "Query" && $arg =~ /^(select|set)/i; $thread = $3; $type = $4; $arg = $5 } else { $arg .= " $_" } ' /var/lib/mysql/xxx.log

Then you can replay doing a simple:

time sh -c 'for i in /tmp/sql-reqs/*.sql; do cat $i | mysql _MY_DB_ >/dev/null & done; wait'

NB: tested on an application (moodle) which never uses prepared statements.


We're using https://github.com/devops-works/slowql here and it works like a charm.

It reads a slow query log and replays it on a server at the desired speed. It does not filetr the slow quelry log though.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.