Reputation: 57198
A field on a model, foo = models.ForeignKey(Foo)
will automatically add a database index for the column, in order to make look-ups faster. That's good and well, but Django's docs don't state whether the fields in a model-meta's unique_together
receive the same treatment. I happen to have a model in which one char field which is listed in unique_together
requires an index for quick lookups. I know that it won't hurt anything to add a duplicate db_index=True
in the field definition, but I'm curious.
Upvotes: 35
Views: 12508
Reputation: 1208
unique_together
does not automatically add indexes for each field included in the list.
The new versions of Django suggest using Index & constraint meta options instead:
https://docs.djangoproject.com/en/3.2/ref/models/options/#unique-together
https://docs.djangoproject.com/en/3.2/ref/models/options/#index-together
https://docs.djangoproject.com/en/dev/ref/models/indexes/
And an example model from an open source project:
class GroupResult(models.Model): """Task Group result/status."""
group_id = models.CharField(
max_length=getattr(
settings,
"DJANGO_CELERY_RESULTS_TASK_ID_MAX_LENGTH",
255
),
unique=True,
verbose_name=_("Group ID"),
help_text=_("Celery ID for the Group that was run"),
)
date_created = models.DateTimeField(
auto_now_add=True,
verbose_name=_("Created DateTime"),
help_text=_("Datetime field when the group result was created in UTC"),
)
date_done = models.DateTimeField(
auto_now=True,
verbose_name=_("Completed DateTime"),
help_text=_("Datetime field when the group was completed in UTC"),
)
content_type = models.CharField(
max_length=128,
verbose_name=_("Result Content Type"),
help_text=_("Content type of the result data"),
)
content_encoding = models.CharField(
max_length=64,
verbose_name=_("Result Encoding"),
help_text=_("The encoding used to save the task result data"),
)
result = models.TextField(
null=True, default=None, editable=False,
verbose_name=_('Result Data'),
help_text=_('The data returned by the task. '
'Use content_encoding and content_type fields to read.'))
def as_dict(self):
return {
'group_id': self.group_id,
'result': self.result,
'date_done': self.date_done,
}
def __str__(self):
return f'<Group: {self.group_id}>'
objects = managers.GroupResultManager()
class Meta:
"""Table information."""
ordering = ['-date_done']
verbose_name = _('group result')
verbose_name_plural = _('group results')
indexes = [
models.Index(fields=['date_created']),
models.Index(fields=['date_done']),
]
Upvotes: 8
Reputation: 22841
For anyone coming here wondering if they need an index_together
in addition to unique_together
to get the index's performance benefit, the answer for Postgres is no, they are functionally the same.
Upvotes: 42
Reputation: 373
In Django 1.5 and higher, you can use the {Model}.Meta.index_together
class attribute. If you had two fields named foo
and bar
, you would add:
class Meta(object):
index_together = unique_together = [
['foo', 'bar']
]
If you have only one set of unique fields, you can use a one-dimensional iterable for unique_together
. However, the documentation does not indicate that the same applies to index_together
.
This would also be okay:
class Meta(object):
unique_together = 'foo', 'bar'
index_together = [
['foo', 'bar']
]
This, however, is NOT supported by the documentation:
class Meta(object):
unique_together = 'foo', 'bar'
index_together = 'foo', 'bar'
Upvotes: 7
Reputation: 309109
If unique_together
does add an index, it will be a multiple column index.
If you want one of the columns to be indexed individually, I believe you need to specify db_index=True
in the field definition.
Upvotes: 14
Reputation: 13496
According to the docs, it will only enforce uniqueness on database level. I think generally making a field unique does not imply it has an index. Though you could also simply check on db level if the index exists. Everything indicates though it does not.
Upvotes: 1