I put this in settings.py:
DATABASES = {
'default': {
'ENGINE': 'django.db.backends.mysql',
'HOST': 'localhost',
'NAME': 'db',
'USER': 'root',
'PASSWORD': '',
'OPTIONS': {
'charset': 'utf8mb4',
'init_command': 'set collation_connection=utf8mb4_unicode_ci',
},
},
}
Then I used the shell to check that it worked:
$ ./manage.py shell
>>> from django.db import connection
>>> cursor = connection.cursor()
>>> cursor.execute("show variables like 'collation_connection'")
>>> print cursor.fetchall()
((u'collation_connection', u'utf8mb4_general_ci'),)
Unfortunately, what I've learned from inspecting my query log is that MySQLdb does this when it connects:
set collation_connection='utf8mb4_unicode_ci'
SET NAMES utf8mb4
That's right. It executes set names after my init command. This sets the collation back to the default.
Omitting the 'charset' option doesn't help. If I do, it will call set names utf8 instead, which is even worse. I tried making the set names command part of my 'init_command' in case it wouldn't clobber my collation if there was nothing to change, but no, it still clobbers it.
I can't fork the Python library MySQLdb because I'm running my app on Google App Engine and MySQLdb is part of App Engine.
I was able to accomplish this by monkey-patching Django:
from django.db.backends.mysql import base
old_get_new_connection = base.DatabaseWrapper.get_new_connection
def get_new_connection(self, conn_params):
conn = old_get_new_connection(self, conn_params)
conn.query("set names 'utf8mb4' collate 'utf8mb4_unicode_520_ci'")
return conn
base.DatabaseWrapper.get_new_connection = get_new_connection
Btw, this patch was necessary to even use the utf8mb4 character set on App Engine because they are using an old version of MySQL Connector C which doesn't support that character set. Without this character set, your app will crash if you attempt to insert any four-byte UTF8 characters, and any existing ones in the database will show up as question marks.