Databases

Django attempts to support as many features as possible on all database backends. However, not all database backends are alike, and we’ve had to make design decisions on which features to support and which assumptions we can make safely.

This file describes some of the features that might be relevant to Django usage. Of course, it is not intended as a replacement for server-specific documentation or reference manuals.

General notes

Persistent connections

Persistent connections avoid the overhead of re-establishing a connection to the database in each request. They’re controlled by the CONN_MAX_AGE parameter which defines the maximum lifetime of a connection. It can be set independently for each database.

The default value is 0, preserving the historical behavior of closing the database connection at the end of each request. To enable persistent connections, set CONN_MAX_AGE to a positive number of seconds. For unlimited persistent connections, set it to None.

Connection management

Django opens a connection to the database when it first makes a database query. It keeps this connection open and reuses it in subsequent requests. Django closes the connection once it exceeds the maximum age defined by CONN_MAX_AGE or when it isn’t usable any longer.

In detail, Django automatically opens a connection to the database whenever it needs one and doesn’t have one already — either because this is the first connection, or because the previous connection was closed.

At the beginning of each request, Django closes the connection if it has reached its maximum age. If your database terminates idle connections after some time, you should set CONN_MAX_AGE to a lower value, so that Django doesn’t attempt to use a connection that has been terminated by the database server. (This problem may only affect very low traffic sites.)

At the end of each request, Django closes the connection if it has reached its maximum age or if it is in an unrecoverable error state. If any database errors have occurred while processing the requests, Django checks whether the connection still works, and closes it if it doesn’t. Thus, database errors affect at most one request; if the connection becomes unusable, the next request gets a fresh connection.

Caveats

Since each thread maintains its own connection, your database must support at least as many simultaneous connections as you have worker threads.

Sometimes a database won’t be accessed by the majority of your views, for example because it’s the database of an external system, or thanks to caching. In such cases, you should set CONN_MAX_AGE to a low value or even 0, because it doesn’t make sense to maintain a connection that’s unlikely to be reused. This will help keep the number of simultaneous connections to this database small.

The development server creates a new thread for each request it handles, negating the effect of persistent connections. Don’t enable them during development.

When Django establishes a connection to the database, it sets up appropriate parameters, depending on the backend being used. If you enable persistent connections, this setup is no longer repeated every request. If you modify parameters such as the connection’s isolation level or time zone, you should either restore Django’s defaults at the end of each request, force an appropriate value at the beginning of each request, or disable persistent connections.

Encoding

Django assumes that all databases use UTF-8 encoding. Using other encodings may result in unexpected behavior such as “value too long” errors from your database for data that is valid in Django. See the database specific notes below for information on how to set up your database correctly.

PostgreSQL notes

Django supports PostgreSQL 8.4 and higher.

PostgreSQL connection settings

See HOST for details.

Optimizing PostgreSQL’s configuration

Django needs the following parameters for its database connections:

  • client_encoding: 'UTF8',
  • default_transaction_isolation: 'read committed' by default, or the value set in the connection options (see below),
  • timezone: 'UTC' when USE_TZ is True, value of TIME_ZONE otherwise.

If these parameters already have the correct values, Django won’t set them for every new connection, which improves performance slightly. You can configure them directly in postgresql.conf or more conveniently per database user with ALTER ROLE.

Django will work just fine without this optimization, but each new connection will do some additional queries to set these parameters.

Autocommit mode

In previous versions of Django, database-level autocommit could be enabled by setting the autocommit key in the OPTIONS part of your database configuration in DATABASES:

DATABASES = {
    # ...
    'OPTIONS': {
        'autocommit': True,
    },
}

Since Django 1.6, autocommit is turned on by default. This configuration is ignored and can be safely removed.

Isolation level

Like PostgreSQL itself, Django defaults to the READ COMMITTED isolation level. If you need a higher isolation level such as REPEATABLE READ or SERIALIZABLE, set it in the OPTIONS part of your database configuration in DATABASES:

import psycopg2.extensions

DATABASES = {
    # ...
    'OPTIONS': {
        'isolation_level': psycopg2.extensions.ISOLATION_LEVEL_SERIALIZABLE,
    },
}

Note

Under higher isolation levels, your application should be prepared to handle exceptions raised on serialization failures. This option is designed for advanced uses.

Indexes for varchar and text columns

When specifying db_index=True on your model fields, Django typically outputs a single CREATE INDEX statement. However, if the database type for the field is either varchar or text (e.g., used by CharField, FileField, and TextField), then Django will create an additional index that uses an appropriate PostgreSQL operator class for the column. The extra index is necessary to correctly perform lookups that use the LIKE operator in their SQL, as is done with the contains and startswith lookup types.

MySQL notes

Version support

Django supports MySQL 5.0.3 and higher.

Django’s inspectdb feature uses the information_schema database, which contains detailed data on all database schemas.

The minimum version requirement of MySQL 5.0.3 was set in Django 1.5.

Django expects the database to support Unicode (UTF-8 encoding) and delegates to it the task of enforcing transactions and referential integrity. It is important to be aware of the fact that the two latter ones aren’t actually enforced by MySQL when using the MyISAM storage engine, see the next section.

Storage engines

MySQL has several storage engines. You can change the default storage engine in the server configuration.

Until MySQL 5.5.4, the default engine was MyISAM [1]. The main drawbacks of MyISAM are that it doesn’t support transactions or enforce foreign-key constraints. On the plus side, it was the only engine that supported full-text indexing and searching until MySQL 5.6.4.

Since MySQL 5.5.5, the default storage engine is InnoDB. This engine is fully transactional and supports foreign key references. It’s probably the best choice at this point. However, note that the the InnoDB autoincrement counter is lost on a MySQL restart because it does not remember the AUTO_INCREMENT value, instead recreating it as “max(id)+1”. This may result in an inadvertent reuse of AutoField values.

If you upgrade an existing project to MySQL 5.5.5 and subsequently add some tables, ensure that your tables are using the same storage engine (i.e. MyISAM vs. InnoDB). Specifically, if tables that have a ForeignKey between them use different storage engines, you may see an error like the following when running syncdb:

_mysql_exceptions.OperationalError: (
    1005, "Can't create table '\\db_name\\.#sql-4a8_ab' (errno: 150)"
)
[1]Unless this was changed by the packager of your MySQL package. We’ve had reports that the Windows Community Server installer sets up InnoDB as the default storage engine, for example.

MySQLdb

MySQLdb is the Python interface to MySQL. Version 1.2.1p2 or later is required for full MySQL support in Django.

Note

If you see ImportError: cannot import name ImmutableSet when trying to use Django, your MySQLdb installation may contain an outdated sets.py file that conflicts with the built-in module of the same name from Python 2.4 and later. To fix this, verify that you have installed MySQLdb version 1.2.1p2 or newer, then delete the sets.py file in the MySQLdb directory that was left by an earlier version.

Note

There are known issues with the way MySQLdb converts date strings into datetime objects. Specifically, date strings with value 0000-00-00 are valid for MySQL but will be converted into None by MySQLdb.

This means you should be careful while using loaddata/dumpdata with rows that may have 0000-00-00 values, as they will be converted to None.

Python 3

At the time of writing, the latest release of MySQLdb (1.2.4) doesn’t support Python 3. In order to use MySQL under Python 3, you’ll have to install an unofficial fork, such as MySQL-for-Python-3.

This port is still in alpha. In particular, it doesn’t support binary data, making it impossible to use django.db.models.BinaryField.

Time zone definitions

If you plan on using Django’s timezone support, use mysql_tzinfo_to_sql to load time zone tables into the MySQL database. This needs to be done just once for your MySQL server, not per database.

Creating your database

You can create your database using the command-line tools and this SQL:

CREATE DATABASE <dbname> CHARACTER SET utf8;

This ensures all tables and columns will use UTF-8 by default.

Collation settings

The collation setting for a column controls the order in which data is sorted as well as what strings compare as equal. It can be set on a database-wide level and also per-table and per-column. This is documented thoroughly in the MySQL documentation. In all cases, you set the collation by directly manipulating the database tables; Django doesn’t provide a way to set this on the model definition.

By default, with a UTF-8 database, MySQL will use the utf8_general_ci collation. This results in all string equality comparisons being done in a case-insensitive manner. That is, "Fred" and "freD" are considered equal at the database level. If you have a unique constraint on a field, it would be illegal to try to insert both "aa" and "AA" into the same column, since they compare as equal (and, hence, non-unique) with the default collation.

In many cases, this default will not be a problem. However, if you really want case-sensitive comparisons on a particular column or table, you would change the column or table to use the utf8_bin collation. The main thing to be aware of in this case is that if you are using MySQLdb 1.2.2, the database backend in Django will then return bytestrings (instead of unicode strings) for any character fields it receive from the database. This is a strong variation from Django’s normal practice of always returning unicode strings. It is up to you, the developer, to handle the fact that you will receive bytestrings if you configure your table(s) to use utf8_bin collation. Django itself should mostly work smoothly with such columns (except for the contrib.sessions Session and contrib.admin LogEntry tables described below), but your code must be prepared to call django.utils.encoding.smart_text() at times if it really wants to work with consistent data – Django will not do this for you (the database backend layer and the model population layer are separated internally so the database layer doesn’t know it needs to make this conversion in this one particular case).

If you’re using MySQLdb 1.2.1p2, Django’s standard CharField class will return unicode strings even with utf8_bin collation. However, TextField fields will be returned as an array.array instance (from Python’s standard array module). There isn’t a lot Django can do about that, since, again, the information needed to make the necessary conversions isn’t available when the data is read in from the database. This problem was fixed in MySQLdb 1.2.2, so if you want to use TextField with utf8_bin collation, upgrading to version 1.2.2 and then dealing with the bytestrings (which shouldn’t be too difficult) as described above is the recommended solution.

Should you decide to use utf8_bin collation for some of your tables with MySQLdb 1.2.1p2 or 1.2.2, you should still use utf8_general_ci (the default) collation for the django.contrib.sessions.models.Session table (usually called django_session) and the django.contrib.admin.models.LogEntry table (usually called django_admin_log). Those are the two standard tables that use TextField internally.

Please note that according to MySQL Unicode Character Sets, comparisons for the utf8_general_ci collation are faster, but slightly less correct, than comparisons for utf8_unicode_ci. If this is acceptable for your application, you should use utf8_general_ci because it is faster. If this is not acceptable (for example, if you require German dictionary order), use utf8_unicode_ci because it is more accurate.

Warning

Model formsets validate unique fields in a case-sensitive manner. Thus when using a case-insensitive collation, a formset with unique field values that differ only by case will pass validation, but upon calling save(), an IntegrityError will be raised.

Connecting to the database

Refer to the settings documentation.

Connection settings are used in this order:

  1. OPTIONS.
  2. NAME, USER, PASSWORD, HOST, PORT
  3. MySQL option files.

In other words, if you set the name of the database in OPTIONS, this will take precedence over NAME, which would override anything in a MySQL option file.

Here’s a sample configuration which uses a MySQL option file:

# settings.py
DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.mysql',
        'OPTIONS': {
            'read_default_file': '/path/to/my.cnf',
        },
    }
}


# my.cnf
[client]
database = NAME
user = USER
password = PASSWORD
default-character-set = utf8

Several other MySQLdb connection options may be useful, such as ssl, init_command, and sql_mode. Consult the MySQLdb documentation for more details.

Creating your tables

When Django generates the schema, it doesn’t specify a storage engine, so tables will be created with whatever default storage engine your database server is configured for. The easiest solution is to set your database server’s default storage engine to the desired engine.

If you’re using a hosting service and can’t change your server’s default storage engine, you have a couple of options.

  • After the tables are created, execute an ALTER TABLE statement to convert a table to a new storage engine (such as InnoDB):

    ALTER TABLE <tablename> ENGINE=INNODB;
    

    This can be tedious if you have a lot of tables.

  • Another option is to use the init_command option for MySQLdb prior to creating your tables:

    'OPTIONS': {
       'init_command': 'SET storage_engine=INNODB',
    }
    

    This sets the default storage engine upon connecting to the database. After your tables have been created, you should remove this option as it adds a query that is only needed during table creation to each database connection.

  • Another method for changing the storage engine is described in AlterModelOnSyncDB.

Table names

There are known issues in even the latest versions of MySQL that can cause the case of a table name to be altered when certain SQL statements are executed under certain conditions. It is recommended that you use lowercase table names, if possible, to avoid any problems that might arise from this behavior. Django uses lowercase table names when it auto-generates table names from models, so this is mainly a consideration if you are overriding the table name via the db_table parameter.

Savepoints

Both the Django ORM and MySQL (when using the InnoDB storage engine) support database savepoints.

If you use the MyISAM storage engine please be aware of the fact that you will receive database-generated errors if you try to use the savepoint-related methods of the transactions API. The reason for this is that detecting the storage engine of a MySQL database/table is an expensive operation so it was decided it isn’t worth to dynamically convert these methods in no-op’s based in the results of such detection.

Notes on specific fields

Character fields

Any fields that are stored with VARCHAR column types have their max_length restricted to 255 characters if you are using unique=True for the field. This affects CharField, SlugField and CommaSeparatedIntegerField.

DateTime fields

MySQL does not store fractions of seconds. Fractions of seconds are truncated to zero when the time is stored.

Row locking with QuerySet.select_for_update()

MySQL does not support the NOWAIT option to the SELECT ... FOR UPDATE statement. If select_for_update() is used with nowait=True then a DatabaseError will be raised.

Automatic typecasting can cause unexpected results

When performing a query on a string type, but with an integer value, MySQL will coerce the types of all values in the table to an integer before performing the comparison. If your table contains the values 'abc', 'def' and you query for WHERE mycolumn=0, both rows will match. Similarly, WHERE mycolumn=1 will match the value 'abc1'. Therefore, string type fields included in Django will always cast the value to a string before using it in a query.

If you implement custom model fields that inherit from Field directly, are overriding get_prep_value(), or use extra() or raw(), you should ensure that you perform the appropriate typecasting.

SQLite notes

SQLite provides an excellent development alternative for applications that are predominantly read-only or require a smaller installation footprint. As with all database servers, though, there are some differences that are specific to SQLite that you should be aware of.

Substring matching and case sensitivity

For all SQLite versions, there is some slightly counter-intuitive behavior when attempting to match some types of strings. These are triggered when using the iexact or contains filters in Querysets. The behavior splits into two cases:

1. For substring matching, all matches are done case-insensitively. That is a filter such as filter(name__contains="aa") will match a name of "Aabb".

2. For strings containing characters outside the ASCII range, all exact string matches are performed case-sensitively, even when the case-insensitive options are passed into the query. So the iexact filter will behave exactly the same as the exact filter in these cases.

Some possible workarounds for this are documented at sqlite.org, but they aren’t utilised by the default SQLite backend in Django, as incorporating them would be fairly difficult to do robustly. Thus, Django exposes the default SQLite behavior and you should be aware of this when doing case-insensitive or substring filtering.

Version 3.5.9

The Ubuntu “Intrepid Ibex” (8.10) SQLite 3.5.9-3 package contains a bug that causes problems with the evaluation of query expressions. If you are using Ubuntu “Intrepid Ibex”, you will need to update the package to version 3.5.9-3ubuntu1 or newer (recommended) or find an alternate source for SQLite packages, or install SQLite from source.

At one time, Debian Lenny shipped with the same malfunctioning SQLite 3.5.9-3 package. However the Debian project has subsequently issued updated versions of the SQLite package that correct these bugs. If you find you are getting unexpected results under Debian, ensure you have updated your SQLite package to 3.5.9-5 or later.

The problem does not appear to exist with other versions of SQLite packaged with other operating systems.

Version 3.6.2

SQLite version 3.6.2 (released August 30, 2008) introduced a bug into SELECT DISTINCT handling that is triggered by, amongst other things, Django’s DateQuerySet (returned by the dates() method on a queryset).

You should avoid using this version of SQLite with Django. Either upgrade to 3.6.3 (released September 22, 2008) or later, or downgrade to an earlier version of SQLite.

Using newer versions of the SQLite DB-API 2.0 driver

For versions of Python 2.5 or newer that include sqlite3 in the standard library Django will now use a pysqlite2 interface in preference to sqlite3 if it finds one is available.

This provides the ability to upgrade both the DB-API 2.0 interface or SQLite 3 itself to versions newer than the ones included with your particular Python binary distribution, if needed.

“Database is locked” errors

SQLite is meant to be a lightweight database, and thus can’t support a high level of concurrency. OperationalError: database is locked errors indicate that your application is experiencing more concurrency than sqlite can handle in default configuration. This error means that one thread or process has an exclusive lock on the database connection and another thread timed out waiting for the lock the be released.

Python’s SQLite wrapper has a default timeout value that determines how long the second thread is allowed to wait on the lock before it times out and raises the OperationalError: database is locked error.

If you’re getting this error, you can solve it by:

  • Switching to another database backend. At a certain point SQLite becomes too “lite” for real-world applications, and these sorts of concurrency errors indicate you’ve reached that point.

  • Rewriting your code to reduce concurrency and ensure that database transactions are short-lived.

  • Increase the default timeout value by setting the timeout database option:

    'OPTIONS': {
        # ...
        'timeout': 20,
        # ...
    }
    

    This will simply make SQLite wait a bit longer before throwing “database is locked” errors; it won’t really do anything to solve them.

QuerySet.select_for_update() not supported

SQLite does not support the SELECT ... FOR UPDATE syntax. Calling it will have no effect.

“pyformat” parameter style in raw queries not supported

For most backends, raw queries (Manager.raw() or cursor.execute()) can use the “pyformat” parameter style, where placeholders in the query are given as '%(name)s' and the parameters are passed as a dictionary rather than a list. SQLite does not support this.

Parameters not quoted in connection.queries

sqlite3 does not provide a way to retrieve the SQL after quoting and substituting the parameters. Instead, the SQL in connection.queries is rebuilt with a simple string interpolation. It may be incorrect. Make sure you add quotes where necessary before copying a query into an SQLite shell.

Oracle notes

Django supports Oracle Database Server versions 9i and higher. Oracle version 10g or later is required to use Django’s regex and iregex query operators. You will also need at least version 4.3.1 of the cx_Oracle Python driver.

Note that due to a Unicode-corruption bug in cx_Oracle 5.0, that version of the driver should not be used with Django; cx_Oracle 5.0.1 resolved this issue, so if you’d like to use a more recent cx_Oracle, use version 5.0.1.

cx_Oracle 5.0.1 or greater can optionally be compiled with the WITH_UNICODE environment variable. This is recommended but not required.

In order for the python manage.py syncdb command to work, your Oracle database user must have privileges to run the following commands:

  • CREATE TABLE
  • CREATE SEQUENCE
  • CREATE PROCEDURE
  • CREATE TRIGGER

To run Django’s test suite, the user needs these additional privileges:

  • CREATE USER
  • DROP USER
  • CREATE TABLESPACE
  • DROP TABLESPACE
  • CONNECT WITH ADMIN OPTION
  • RESOURCE WITH ADMIN OPTION

The Oracle database backend uses the SYS.DBMS_LOB package, so your user will require execute permissions on it. It’s normally accessible to all users by default, but in case it is not, you’ll need to grant permissions like so:

GRANT EXECUTE ON SYS.DBMS_LOB TO user;

Connecting to the database

To connect using the service name of your Oracle database, your settings.py file should look something like this:

DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.oracle',
        'NAME': 'xe',
        'USER': 'a_user',
        'PASSWORD': 'a_password',
        'HOST': '',
        'PORT': '',
    }
}

In this case, you should leave both HOST and PORT empty. However, if you don’t use a tnsnames.ora file or a similar naming method and want to connect using the SID (“xe” in this example), then fill in both HOST and PORT like so:

DATABASES = {
    'default': {
        'ENGINE': 'django.db.backends.oracle',
        'NAME': 'xe',
        'USER': 'a_user',
        'PASSWORD': 'a_password',
        'HOST': 'dbprod01ned.mycompany.com',
        'PORT': '1540',
    }
}

You should either supply both HOST and PORT, or leave both as empty strings. Django will use a different connect descriptor depending on that choice.

Threaded option

If you plan to run Django in a multithreaded environment (e.g. Apache using the default MPM module on any modern operating system), then you must set the threaded option of your Oracle database configuration to True:

'OPTIONS': {
    'threaded': True,
},

Failure to do this may result in crashes and other odd behavior.

INSERT ... RETURNING INTO

By default, the Oracle backend uses a RETURNING INTO clause to efficiently retrieve the value of an AutoField when inserting new rows. This behavior may result in a DatabaseError in certain unusual setups, such as when inserting into a remote table, or into a view with an INSTEAD OF trigger. The RETURNING INTO clause can be disabled by setting the use_returning_into option of the database configuration to False:

'OPTIONS': {
    'use_returning_into': False,
},

In this case, the Oracle backend will use a separate SELECT query to retrieve AutoField values.

Naming issues

Oracle imposes a name length limit of 30 characters. To accommodate this, the backend truncates database identifiers to fit, replacing the final four characters of the truncated name with a repeatable MD5 hash value. Additionally, the backend turns database identifiers to all-uppercase.

To prevent these transformations (this is usually required only when dealing with legacy databases or accessing tables which belong to other users), use a quoted name as the value for db_table:

class LegacyModel(models.Model):
    class Meta:
        db_table = '"name_left_in_lowercase"'

class ForeignModel(models.Model):
    class Meta:
        db_table = '"OTHER_USER"."NAME_ONLY_SEEMS_OVER_30"'

Quoted names can also be used with Django’s other supported database backends; except for Oracle, however, the quotes have no effect.

When running syncdb, an ORA-06552 error may be encountered if certain Oracle keywords are used as the name of a model field or the value of a db_column option. Django quotes all identifiers used in queries to prevent most such problems, but this error can still occur when an Oracle datatype is used as a column name. In particular, take care to avoid using the names date, timestamp, number or float as a field name.

NULL and empty strings

Django generally prefers to use the empty string (‘’) rather than NULL, but Oracle treats both identically. To get around this, the Oracle backend ignores an explicit null option on fields that have the empty string as a possible value and generates DDL as if null=True. When fetching from the database, it is assumed that a NULL value in one of these fields really means the empty string, and the data is silently converted to reflect this assumption.

TextField limitations

The Oracle backend stores TextFields as NCLOB columns. Oracle imposes some limitations on the usage of such LOB columns in general:

  • LOB columns may not be used as primary keys.
  • LOB columns may not be used in indexes.
  • LOB columns may not be used in a SELECT DISTINCT list. This means that attempting to use the QuerySet.distinct method on a model that includes TextField columns will result in an error when run against Oracle. As a workaround, use the QuerySet.defer method in conjunction with distinct() to prevent TextField columns from being included in the SELECT DISTINCT list.

Using a 3rd-party database backend

In addition to the officially supported databases, there are backends provided by 3rd parties that allow you to use other databases with Django:

The Django versions and ORM features supported by these unofficial backends vary considerably. Queries regarding the specific capabilities of these unofficial backends, along with any support queries, should be directed to the support channels provided by each 3rd party project.