25% developed

Converting MySQL to PostgreSQL

From Wikibooks, open books for an open world
Jump to navigation Jump to search

Very Short Intro[edit | edit source]

You may have read a bunch of short articles with the same name on the web, but they were just snippets of information you needed. It's time to put it all together.

You have a project in MySQL and suddenly you find out that you need to switch to PostgreSQL. Suddenly you see that there are many flavours of SQL and that your seemingly basic constructions throw a lot of errors. You don't have time to really rewrite your code from scratch, it may come later...

Actually, there may be some good reasons to switch...

With PostgreSQL you may still feel a little like a second-class citizen, but not really the ignored one. There are some major projects like Asterisk, Horde or DBMail that have recognized its qualities and although MySQL was their first choice database, they are showing effort to make things run here too.

Check that the Server is Running[edit | edit source]

Most likely you don't need this chapter, but very briefly: after you've installed your package with PostgreSQL on your Linux machine (be it from a package or following these notes), you need to do something like

su -
su - postgres
createdb test
psql test
=# create user username password ' password ';
-- To change a password:
=# alter role username password ' password ';
=# create database databasename with encoding 'utf8';
=# grant all privileges on database databasename to username;
=# \l
=# \c databasename
=# \q

vi /etc/postgresql/pg_hba.conf

host    all         all         0.0.0.0           0.0.0.0            md5

be SURE to cover this security issue with iptables!

/etc/init.d/postgresql reload or /usr/lib/postgresql/bin/pg_ctl reload

postmaster successfully signaled

psql -h server -d databasename -U username

databasename=>

Convert and Import[edit | edit source]

Using pgloader[edit | edit source]

Have a look at http://pgloader.io and you can migrate your MySQL database over to PostgreSQL in a single command:

pgloader mysql://user@localhost/dbname postgresql:///dbname

This will handle type casting with a default casting rules set, and also schema discovery in MySQL and creation in PostgreSQL, including tables, columns, constraints (primary keys, foreign keys, NOT NULL), default values, and secondary indexes. The data are transformed on the fly to be accepted by PostgreSQL, which includes getting rid of zero-dates (there's no year zero in our calendar, neither month nor day zero, and while MySQL doesn't care about that PostgreSQL is quite strongly opinionated that if you use year zero then what you're dealing with is not a date).

For more advanced options or if you want to change the default settings pgloader MySQL support[1] allows you to write a full command using its own language with different rules to describe how you want your migration done.

Common way with SQL dump[edit | edit source]

Dump your tables with

mysqldump -u username -p --compatible=postgresql databasename > outputfile.sql

but even then you will have to change escaped chars (replacing \t with ^I, \n with ^M, single quote (') with doubled single quote and double (escaped) backslash (\\) with a single backslash). This can't be trivially done with sed command, you may need to write a script for it (Ruby, Perl, etc). There is a MySQL to PostgreSQL python convert script (you need to use --default-character-set=utf8 when exporting your mysqldump to make it work). It is much better and proven solution to prepend your dump with the following lines

SET standard_conforming_strings = 'off';
SET backslash_quote = 'on';

These options will force PostgreSQL parser to accept non-ANSI-SQL-compatible escape sequences (Postgre will still issue HINTs on it; you can safely ignore them). Do not set these options globally: this may compromise security of the server!

You also have to manually modify the data types etc. as discussed later.

After you convert your tables, import them the same way you were used to in MySQL, that is

psql -h server -d databasename -U username -f data.sql

Export using CSV-files[edit | edit source]

When you have a large sql dump containing binary data, it will not be easy to modify the data structure, so there is another way to export your data to PostgreSQL. Mysql have an option to export each table from the database as a separate .sql file with table structure and .txt file with table's data in CSV-format:

mysqldump -u username -p --compatible=postgresql -T /path/to/export databasename

Notice that /path/to/export should be writeable by user who runs mysqld, in most case it mysqld. After that you should modify your table structure according PostgreSQL format:

  • convert data types
  • create separate keys definitions
  • replace escape characters

When table structure will be ready, you should load it as it was shown above. You should prepare data files: replace carriage return characters to "\r" and remove invalid characters for your data encoding. Here is an example bash script how you can do this and load all the data in your database:

#!/bin/bash

CHARSET="utf-8" #your current database charset
DATADIR="/path/to/export"
DBNAME="databasename"

for file in $DATADIR/*.txt; do
  TMP=${file%.*}
  TABLE=${TMP##*/}
  echo "preparing $TABLE"

  #replace carriage return
  sed 's/\r/\\r/g' $file > /tmp/$TABLE.export.tmp

  #cleanup non-printable and wrong sequences for current charset
  iconv -t $CHARSET -f $CHARSET -c < /tmp/$TABLE.export.tmp > /tmp/$TABLE.export.tmp.out

  echo "loading $TABLE"
  /usr/bin/psql $DBNAME -c "copy $TABLE from '/tmp/$TABLE.export.tmp.out'"

  #clean up
  rm /tmp/$TABLE.export.tmp /tmp/$TABLE.export.tmp.out
done

The Environment[edit | edit source]

Perl[edit | edit source]

You will need to install an appropriate DBD package. In Debian/Ubuntu run apt-get install libdbd-pg-perl.

Changing The Code Quick And Dirty[edit | edit source]

Perl[edit | edit source]

MySQL PostgreSQL comments
$db=DBI->connect("dbi:mysql:database= ... )
$db=DBI->connect("dbi:Pg:database= ... )
All you have to do is changing mysql to Pg. Beware the case sensitivity.

SQL[edit | edit source]

Syntax[edit | edit source]

MySQL PostgreSQL comments
#
--
MySQL accepts nonstandard # to begin a comment line; PostgreSQL uses ANSI standard double dash; use the ANSI standard, both databases understand it. (However, MySQL requires a space after --, whilst it is not mandatory in PostgreSQL)
' " vs. `
' vs. "
MySQL uses ' or " to quote values (i.e. WHERE name = "John"). This is not the ANSI standard for databases. PostgreSQL uses only single quotes for this (i.e. WHERE name = 'John'). Double quotes are used to quote system identifiers; field names, table names, etc. (i.e. WHERE "last name" = 'Smith'). MySQL uses ` (accent mark or backtick) to quote system identifiers, which is decidedly non-standard. Note: you can make MySQL interpret quotes like PostgreSQL using SET sql_mode='ANSI_QUOTES'.
... WHERE lastname="smith"
... WHERE lower(lastname)='smith'
PostgreSQL is case-sensitive for string comparisons. The value 'Smith' is not the same as 'smith'. This is a big change for many users from MySQL (in MySQL, VARCHAR and TEXT columns are case-insensitive unless the "binary" flag is set) and other small database systems, like Microsoft Access. In PostgreSQL, you can either:
  • Use the correct case in your query. (i.e. WHERE lastname='Smith')
  • Use a conversion function, like lower() to search. (i.e. WHERE lower(lastname)='smith')
  • Use a case-insensitive operator, like ILIKE or *~
`LastName` = `lastname`

and maybe not?

"LastName" <> "lastname"
Database, table, field and columns names in PostgreSQL are case-independent, unless you created them with double-quotes around their name, in which case they are case-sensitive. In MySQL, table names can be case-sensitive or not, depending on which operating system you are using.
Note that PostgreSQL actively converts all non-quoted names to lower case and so returns lower case in query results!
'foo' || 'bar'
means OR
'foo' || 'bar'
means string concatenation (= 'foobar')
MySQL accepts C-language operators for logic, SQL requires AND, OR; use the SQL standard keywords for logic, both databases understand it.

Data Types[edit | edit source]

The ideas for this table were partially derived from automated dump converting script [1]. Official documentation:

List of available data types can be reached also by using psql's internal slash command \dT.

MySQL PostgreSQL ANSI Standard SQL comments
TINYINT
SMALLINT
MEDIUMINT
BIGINT
SMALLINT
SMALLINT
INTEGER
BIGINT
INTEGER
INTEGER
INTEGER
NUMERIC(20)
see [2]; integer size in PostgreSQL is 4 Bytes signed (-2147483648 – +2147483647)
TINYINT UNSIGNED
SMALLINT UNSIGNED
MEDIUMINT UNSIGNED
INT UNSIGNED
BIGINT UNSIGNED
SMALLINT
INTEGER
INTEGER
BIGINT
NUMERIC(20)
INTEGER
INTEGER
INTEGER
NUMERIC(10)
NUMERIC(20)
SQL doesn't know UNSIGNED, all numbers are signed.
FLOAT
FLOAT UNSIGNED
REAL
REAL
FLOAT4
FLOAT4
DOUBLE
DOUBLE PRECISION
FLOAT8
BOOLEAN
BOOLEAN
BOOLEAN
MySQL Booleans are an alias for TINYINT(1); PostgreSQL doesn't auto-convert numbers into booleans.
TINYTEXT
TEXT
MEDIUMTEXT
LONGTEXT
TEXT
TEXT
TEXT
TEXT
TEXT
TEXT
TEXT
TEXT
BINARY(n)
VARBINARY(n)
TINYBLOB
BLOB
MEDIUMBLOB
LONGBLOB
BYTEA
BYTEA
BYTEA
BYTEA
BYTEA
BYTEA
BIT(n)
BIT VARYING(n)
TEXT
TEXT
TEXT
TEXT
ZEROFILL
not available
not available
DATE
TIME
DATETIME
TIMESTAMP
DATE
TIME [WITHOUT TIME ZONE]
TIMESTAMP [WITHOUT TIME ZONE]
TIMESTAMP [WITHOUT TIME ZONE]
DATE
TIME
TIMESTAMP
TIMESTAMP
column SERIAL

equals to:

column BIGINT UNSIGNED NOT NULL AUTO_INCREMENT UNIQUE

or:

column INT DEFAULT SERIAL

equals to:

column INT NOT NULL AUTO_INCREMENT UNIQUE
column SERIAL

equals to:

CREATE SEQUENCE name;
CREATE TABLE table ( 
	column INTEGER NOT NULL
	DEFAULT nextval(name)
);
column GENERATED BY DEFAULT
Note for PostgresSQL:

SERIAL = 1 – 2147483647
BIGSERIAL = 1 – 9223372036854775807

SERIAL is in fact an entity named SEQUENCE. It exists independently on the rest of your table. If you want to cleanup your system after dropping a table, you also have to DROP SEQUENCE name. More on that topic...

Note for MySQL:

column SERIAL PRIMARY KEY

or

column SERIAL,
PRIMARY KEY(column)

Will result in having 2 indexes for column. One will be generated by the PRIMARY KEY constraint, and one by the implicit UNIQUE constraint present in the SERIAL alias. This has been reported as a bug and might be corrected.

column ENUM (value1, value2, [...])
column VARCHAR(255) NOT NULL,
CHECK (column IN (value1, value2, [...]))

or

CREATE TYPE mood AS ENUM ('sad','ok','happy');
CREATE TABLE person ( current_mood mood ... )
column VARCHAR(255) NOT NULL,
CHECK (column IN (value1, value2, [...]))
PostgreSQL doesn't have the ENUM types prior to 8.3, so you need to simulate it with constraints when using < 8.3.

Language Constructs[edit | edit source]

MySQL PostgreSQL comments
DESCRIBE table
Using psql:
\d table

or

SELECT
	a.attname AS Field,
	t.typname || '(' || a.atttypmod || ')' AS Type,
	CASE WHEN a.attnotnull = 't' THEN 'YES' ELSE 'NO' END AS Null,
	CASE WHEN r.contype = 'p' THEN 'PRI' ELSE '' END AS Key,
	(SELECT substring(pg_catalog.pg_get_expr(d.adbin, d.adrelid), '\'(.*)\'')
		FROM
			pg_catalog.pg_attrdef d
		WHERE
			d.adrelid = a.attrelid
			AND d.adnum = a.attnum
			AND a.atthasdef) AS Default,
	'' as Extras
FROM
	pg_class c 
	JOIN pg_attribute a ON a.attrelid = c.oid
	JOIN pg_type t ON a.atttypid = t.oid
	LEFT JOIN pg_catalog.pg_constraint r ON c.oid = r.conrelid 
		AND r.conname = a.attname
WHERE
	c.relname = 'tablename'
	AND a.attnum > 0
	
ORDER BY a.attnum
PostgreSQL doesn't implement an SQL extension; it uses psql's internal slash command instead. (Be careful: in the mysql client, \d is shorthand for DROP TABLE)
DROP TABLE IF EXISTS table
DROP TABLE IF EXISTS table 
IF EXISTS in DROP TABLE clause only available since PostgreSQL 8.2.
REPLACE [INTO] table [(column, [...])] VALUES (value, [...])

or

INSERT INTO table
	(column1, column2, [...])
VALUES
	(value1, value2, [...])
ON DUPLICATE KEY UPDATE
	column1 = value1, column2 = value2
CREATE FUNCTION someplpgsqlfunction() RETURNS void AS $$
BEGIN
	IF EXISTS( SELECT * FROM phonebook WHERE name = 'john doe' ) THEN
		UPDATE phonebook
		SET extension = '1234' WHERE name = 'john doe';
	ELSE
		INSERT INTO phonebook VALUES( 'john doe', '1234' );
	END IF;

	RETURN;
END;
$$ LANGUAGE plpgsql;
PostgreSQL doesn't implement REPLACE SQL extension. The presented solution uses PL/pgSQL.

(Note: MySQL REPLACE INTO deletes the old row and inserts the new, instead of updating in-place.)

SELECT ... INTO OUTFILE '/var/tmp/outfile'
COPY ( SELECT ... ) TO '/var/tmp/outfile'
SHOW DATABASES
Run psql with -l parameter

or using psql:

\l

or

SELECT datname AS Database FROM pg_database
	WHERE datistemplate = 'f'
PostgreSQL doesn't implement an SQL extension.
SHOW TABLES
Using psql:
\dt

or

SELECT c.relname AS Tables_in FROM pg_catalog.pg_class c
	LEFT JOIN pg_catalog.pg_namespace n ON n.oid = c.relnamespace
WHERE pg_catalog.pg_table_is_visible(c.oid)
	AND c.relkind = 'r'
	AND relname NOT LIKE 'pg_%'
ORDER BY 1
PostgreSQL doesn't implement an SQL extension; it uses psql's internal slash command instead.
SELECT ... LIMIT offset, limit

or

SELECT ... LIMIT limit OFFSET offset
SELECT ... LIMIT limit OFFSET offset
CREATE TABLE table (
	column ... ,
	{INDEX|KEY} [name] (column, [...])
)

or

CREATE INDEX name ON table (column, [...])
CREATE INDEX name ON table (column, [...])
USE database ;
Using psql:
\c database
UNLOCK TABLES;
-- nothing
"There is no UNLOCK TABLE command; locks are always released at transaction end." ( http://www.postgresql.org/docs/8.1/static/sql-lock.html )

Functions[edit | edit source]

MySQL PostgreSQL comments
LAST_INSERT_ID() CURRVAL('serial_variable') NOTE: it is not only "substitute string" solution as you need to know the name of SERIAL variable (unlike AUTO_INCREMENT in MySQL). Also note that PostgreSQL can play with the OID of the last row inserted by the most recent SQL command.

NOTE2: Even better way to replace LAST_INSERT_ID() is creating a rule, because this cannot suffer from race-conditions:

CREATE RULE get_{table}_id_seq AS ON INSERT TO {table} DO SELECT currval('{table}_id_seq'::text) AS id;

(usage is somehow strange, you get a result from an INSERT-statement, but it works very well)

NOTE3: Another, more readable way:

INSERT INTO mytable VALUES (...) RETURNING my_serial_column_name;

Common Errors[edit | edit source]

  • ERROR: relation "something" does not exist - usually table doesn't exist as you probably didn't make it with the new datatypes or syntax. Also watch out for case folding issues; PostgreSQL = postgresql != "PostgreSQL".
  • prepared statement "dbdpg_X" does not exist -

PL/pgSQL[edit | edit source]

Install[edit | edit source]

In versions prior to 9.0, you have to make it available explicitly for every database:

your_unix$ su - postgres
your_unix$ .../pgsql/bin/createlang plpgsql -h localhost -d databasename

(On BSD systems, the username is pgsql)

Running A Function[edit | edit source]

SELECT definedfunction();

Administration[edit | edit source]

To use the same backup technique as used with MySQL, in /etc/logrotate.d/postgresql-dumps:

/dumps/postgresql/*/*.dump.gz {
        daily
        rotate 20
        dateext
        nocompress
        sharedscripts
        create
        postrotate
                for i in $(su - postgres -c "psql --list -t" | awk '{print $1}' | grep -vE '^$|^template[0-9]'); do
                        if [ ! -e /dumps/postgresql/$i ]; then mkdir -m 700 /dumps/postgresql/$i; fi
                        # compress even in custom format, because it can be compressed more
                        su - postgres -c "pg_dump --format=custom $i" | gzip > /dumps/postgresql/$i/$i.dump.gz
                done
        endscript
}
/dumps/postgresql/*/*.sql.gz {
        daily
        rotate 20
        dateext
        nocompress
        sharedscripts
        create
        postrotate
                for i in $(su - postgres -c "psql --list -t" | awk '{print $1}' | grep -vE '^$|^template[0-9]'); do
                        if [ ! -e /dumps/postgresql/$i ]; then mkdir -m 700 /dumps/postgresql/$i; fi
                        su - postgres -c "pg_dump --format=plain $i" | gzip > /dumps/postgresql/$i/$i.sql.gz
                done
        endscript
}
/dumps/postgresql/*/*.tar.gz {
        daily
        rotate 20
        dateext
        nocompress
        sharedscripts
        create
        postrotate
                for i in $(su - postgres -c "psql --list -t" | awk '{print $1}' | grep -vE '^$|^template[0-9]'); do
                        if [ ! -e /dumps/postgresql/$i ]; then mkdir -m 700 /dumps/postgresql/$i; fi
                        su - postgres -c "pg_dump --format=tar $i" | gzip > /dumps/postgresql/$i/$i.tar.gz
                done
        endscript
}

Related wikibooks[edit | edit source]

References[edit | edit source]