mysql-haskell

mysql-haskell is a MySQL driver written entirely in haskell.

Is it fast?

In short, reading is as fast as the C client (libmysqlclient) on one connection and up to 1.7 times slower with ten connections in parallel, and 5 to 8 times faster than mysql-simple. Writing one statement at a time is 1.2 to 1.4 times slower than C, and on par with C through executeMany.

Median wall time in milliseconds over 10 runs, for 1 to 10 threads that each use their own connection:

threads	1	2	3	4	10
libmysqlclient select	76	80	88	108	178
mysql-haskell select	80	95	106	130	309
mysql-haskell select over TLS (tls)	120	130	148	172	436
libmysqlclient select, prepared	82	86	94	120	179
mysql-haskell select, prepared	77	88	102	124	275
mysql-simple select	462	504	578	698	2399
libmysqlclient insert	29	30	36	38	62
mysql-haskell insert	40	40	45	46	76
mysql-haskell insert, executeMany	27	29	36	37	63
mysql-haskell insert, prepared	38	38	45	46	82

Each thread either reads all 300,024 rows of the sample employees table with select * from employees, or inserts 1000 rows into a 29-column table with auto-commit off. The programs are the ones in benchmark/. They ran on mysql-haskell 1.3.3 with GHC 9.10.3 and +RTS -N4, against MySQL 8.0.45 and its own client library (mysql-simple goes through MariaDB Connector/C 3.3.5), with TLS off unless stated, on an AMD Ryzen AI 7 350 (October 2026). The server kept its data in RAM, so the inserts measure the clients rather than the disk.

Leave the allocation area (-A) at its default when running with several capabilities: -A128M with -N4 makes the runtime fault in a 128 MB allocation area for each capability it touches, which costs 35 to 280 ms per run in these benchmarks on every GHC version tried. Since GHC 8.2 it touches all four even when one thread does the work, where GHC 8.0, used for the 2016 figures this README showed until 2026, touched one or two. Without the flag, mysql-haskell 0.6.0.0 and 1.3.3 run plain and prepared selects equally fast, and 1.3.3 is 1.4 to 1.6 times faster over TLS.

Motivation

While MySQL may not be the most advanced sql database, it's widely used among China companies, including but not limited to Baidu, Alibaba, Tecent etc., but haskell's MySQL support is not ideal, we only have a very basic MySQL binding written by Bryan O'Sullivan, and some higher level wrapper built on it, which have some problems:

• lack of prepared statment and binary protocol support.

• limited concurrency due to FFI.

• no replication protocol support.

mysql-pure is intended to solve these problems, and provide foundation for higher level libraries such as groundhog and persistent, so that accessing MySQL is both fast and easy in haskell.

Guide

The Database.MySQL.Base module provides everything you need to start making queries:

{-# LANGUAGE OverloadedStrings #-}

module Main where

import Database.MySQL.Base
import qualified System.IO.Streams as Streams

main :: IO () 
main = do
    conn <- connect
        defaultConnectInfo {ciUser = "username", ciPassword = "password", ciDatabase = "dbname"}
    (defs, is) <- query_ conn "SELECT * FROM some_table"
    print =<< Streams.toList is

query/query_ will return a column definition list, and an InputStream of rows, you should consume this stream completely before start new queries.

It's recommanded to use prepared statement to improve query speed:

    ...
    s <- prepareStmt conn "SELECT * FROM some_table where person_age > ?"
    ...
    (defs, is) <- queryStmt conn s [MySQLInt32U 18]
    ...

If you want to do batch inserting/deleting/updating, you can use executeMany to save considerable time.

To decode rows straight into your own types, skipping MySQLValue, give queryRows_, queryRows or queryStmtRows a decoder from Database.MySQL.Decoder. Each column is checked against its decoder once, before the first row:

import qualified Database.MySQL.Decoder as Decode

data Person = Person { name :: Text, age :: Word32 }

person :: Decode.RowDecoder Person
person = Person <$> Decode.field Decode.text <*> Decode.field Decode.word32

    ...
    (defs, people) <- queryRows_ person conn "SELECT name, person_age FROM some_table"

The Database.MySQL.BinLog module provides binlog listenning functions and row-based event decoder, following program will automatically get last binlog position, and print every row event it receives:

{-# LANGUAGE LambdaCase #-}
module Main where

import           Control.Monad         (forever)
import qualified Database.MySQL.BinLog as MySQL
import qualified System.IO.Streams     as Streams

main :: IO () 
main = do
    conn <- MySQL.connect 
        MySQL.defaultConnectInfo
          { MySQL.ciUser = "username"
          , MySQL.ciPassword = "password"
          , MySQL.ciDatabase = "dbname"
          }
    MySQL.getLastBinLogTracker conn >>= \ case
        Just tracker -> do
            es <- MySQL.decodeRowBinLogEvent =<< MySQL.dumpBinLog conn 1024 tracker False
            forever $ do
                Streams.read es >>= \ case
                    Just v  -> print v
                    Nothing -> return ()
        Nothing -> error "can't get latest binlog position"

Build Test Benchmark

Just use the old way:

git clone https://github.com/winterland1989/mysql-pure.git
cd mysql-pure
cabal install --enable-tests --only-dependencies
cabal build

Running tests require:

• A local MySQL server, a user testMySQLHaskell and a database testMySQLHaskell, you can do it use following script:

mysql -u root -e "CREATE DATABASE IF NOT EXISTS testMySQLHaskell;"
mysql -u root -e "CREATE USER 'testMySQLHaskell'@'localhost' IDENTIFIED BY ''"
mysql -u root -e "GRANT ALL PRIVILEGES ON testMySQLHaskell.* TO 'testMySQLHaskell'@'localhost'"
mysql -u root -e "FLUSH PRIVILEGES"

• Enable binlog by adding log_bin = filename to my.cnf or add --log-bin=filename to the server, and grant replication access to testMySQLHaskell with:

mysql -u root -e "GRANT REPLICATION SLAVE, REPLICATION CLIENT ON *.* TO 'testMySQLHaskell'@'localhost';"

• Set binlog_format to ROW.

• Set max_allowed_packet to larger than 256M(for test large packet).

Enter benchmark directory and run ./bench.sh to benchmark 1) c++ version 2) mysql-pure 3) FFI version mysql, you may need to:

• Modify bench.sh(change the include path) to get c++ version compiled.
• Modify mysql-pure-bench.cabal(change the openssl's lib path) to get haskell version compiled.
• Setup MySQL's TLS support, modify MySQLHaskellOpenSSL.hs/MySQLHaskellTLS.hs to change the CA file's path, and certificate's subject name.
• Adjust rts options -N to get best results, and leave -A at its default (see "Is it fast?").

With -N10 on my company's 24-core machine, binary protocol performs almost identical to c version!

The .cabal files in benchmark/ date from 2016 and no longer build; the figures in "Is it fast?" came from the same programs compiled against the current library.

Reference

MySQL official site provided intensive document, but without following project, mysql-pure may not be written at all:

• mysql-binlog-connector-java

• canal

• go mysql toolkit

• python binlog parser

License

Copyright (c) 2016, winterland1989

All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

* Redistributions of source code must retain the above copyright
  notice, this list of conditions and the following disclaimer.

* Redistributions in binary form must reproduce the above
  copyright notice, this list of conditions and the following
  disclaimer in the documentation and/or other materials provided
  with the distribution.

* Neither the name of winterland1989 nor the names of other
  contributors may be used to endorse or promote products derived
  from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.