All Articles

Using Case..When in Django

I was working on a multi-tenant project and encountered a bug when using Django’s GenericForeignKey with django-tenants. It was using the public schema’s contenttype_id instead of the tenant schema’s contenttype_id.

So, if I have a model of Comment, my django_content_type table would have something like

public.django_content_type

id app_label model
15 comments comment

tenant.django_content_type

id app_label model
19 comments comment

There shouldn’t be a problem here since django-tenants should handle this because it chooses the id of the tenant first and then only use the public as a fall back value. But for some reason, it was sometimes using the public id so comments aren’t appearing at all!

In order to fix this, I opted to remove django_content_type table from all of my tenants and should only use the public’s django_content_type values.

What I had to do was to update the contents inside my models that were using GenericForeignKeys, which in my case is the comments table.

Here’s the model:

# comments/models.py

class Comment(models.Model):
    user = models.ForeignKey(settings.AUTH_USER_MODEL, on_delete=models.CASCADE)
    parent = models.ForeignKey("self", null=True, blank=True, on_delete=models.SET_NULL)
    path = models.CharField(max_length=350)
    text = models.TextField()
    timestamp = models.DateTimeField(auto_now_add=True)
    updated = models.DateTimeField(auto_now=True)
    active = models.BooleanField(default=True)
    flagged = models.BooleanField(default=False)

    target_content_type = models.ForeignKey(ContentType, null=True, blank=True, on_delete=models.SET_NULL)
    target_object_id = models.PositiveIntegerField(null=True, blank=True)
    target_object = GenericForeignKey("target_content_type", "target_object_id")

I need to update the target_content_type so that it uses the public id which is 15 instead of 19.

What I needed to do was:

  1. Determine which target ids need to be updated. For example, a comment can be in an Announcement or in a Post. So, we’d have to determine the content_type_id for Announcement and Post in the tenant’s schema and update its value so it uses the one in public.
  2. Update the values using Case..When.
  3. Drop the tenant.django_content_type table so it would always use public.django_content_type.

For number 1, I had to do a GROUP BY to determine which id’s I need to update then get its equivalent in the public schema.

For number 2, I had to use the Case..When syntax. So, for example when the target_content_type_id is 19, then update its value to 15. when the target_content_type_id is 20, then update its value to 12.

I think I need not explain number 3 since it only drops the table.

I created a management command for this so it can be easily executed in production. Here’s the code:

from django.apps import apps
from django.contrib.contenttypes.models import ContentType
from django.core.management.base import BaseCommand
from django.db import connection
from django.db.models import Case, F, Value, When

from django_tenants.utils import schema_context

from tenant.models import Tenant


def group_by_sql(schema, table, column):
    sql = f"""
        SELECT {column} FROM {schema}.{table}
        GROUP BY {column}
    """
    print(sql)
    return sql


class Command(BaseCommand):

    help = "One time management command execution to update tenant's content_type_ids"

    def handle(self, *args, **options):

        has_gfk_models = [
            {
                'app_label': 'comments',
                'model': 'comment',
                'col': 'target_content_type_id'
            },
            {
                'app_label': 'notifications',
                'model': 'notification',
                'col': 'target_content_type_id'
            },
            {
                'app_label': 'notifications',
                'model': 'notification',
                'col': 'action_content_type_id',
            },
            {
                'app_label': 'prerequisites',
                'model': 'prereq',
                'col': 'parent_content_type_id',
            },
        ]

        for tenant in Tenant.objects.exclude(schema_name='public'):
            for has_gfk_model in has_gfk_models:
                app_label, model, col = has_gfk_model.values()

                # Number 1
                with connection.cursor() as cursor:
                    cursor.execute(group_by_sql(
                        schema=tenant.schema_name,
                        table=f"{app_label}_{model}",
                        column=col))

                    # Remove null ids
                    tenant_target_content_type_ids = [_id[0] for _id in cursor.fetchall() if _id[0]]
                    # print(tenant_target_content_type_ids)

                    # tenant content_type_id : public content_type_id
                    ct_ids_map = {}
                    for ct_id in tenant_target_content_type_ids:
                        # Get what kind of model the given ID is
                        with schema_context(tenant.schema_name):
                            ct_tenant_app = ContentType.objects.get(id=ct_id)
                        # ... then fetch its equivalent in the public tenant
                        try:
                            ct_public = ContentType.objects.get(app_label=ct_tenant_app.app_label, model=ct_tenant_app.model)
                            ct_ids_map[ct_id] = ct_public.id
                        except ContentType.DoesNotExist:
                            # Just skip the apps that aren't installed anymore
                            print(f'{ct_tenant_app} has been removed from settings.APPS')
                            continue

                    # Number 2
                    Model = apps.get_model(app_label, model)
                    with schema_context(tenant.schema_name):
                        # Using CASE..WHEN is much faster compared to bulk_update in this case
                        # https://docs.djangoproject.com/en/dev/ref/models/conditional-expressions/#conditional-update
                        whens = []

                        for tenant_ct_id, public_ct_id in ct_ids_map.items():
                            # Build query
                            # when target_content_type_id is 19 then update it to 15
                            # When(target_content_type_id={tenant_ct_id}, then=Value({public_ct_id}))
                            when = {
                                col: tenant_ct_id,
                                'then': Value(public_ct_id),
                            }
                            whens.append(When(**when))

                        # If we are currently updating comments, the query would look something like
                        # Comment.objects.update(
                        #   target_content_type_id=Case(
                        #       When(target_content_type_id=17, then=Value(25)),
                        #       When(...),
                        #       default=F(target_content_type_id)))
                        # )
                        case_when = {
                            # When statements should be wrapped in a `Case` so we need to unpack the list `*whens`
                            col: Case(*whens, default=F(col)),
                        }

                        # Filter out the queryset so we don't bother updating other target ids
                        # The `default` is useless in this case because we are only updating the ids that are needed
                        # so it's safe to remove the `default=F(col)`.
                        qs = Model.objects.filter(**{f'{col}__in': ct_ids_map.keys()})
                        qs.update(**case_when)
                        print(connection.queries)

            # Drop the table so it only uses public.django_content_type
            drop_contenttype_table = f"DROP TABLE IF EXISTS {tenant.schema_name}.django_content_type CASCADE"
            with connection.cursor() as cursor:
                cursor.execute(drop_contenttype_table)

The code above could still be improved but it did the job for me. Also, here’s a link to a GitHub gist: https://gist.github.com/yujinyuz/d257d3ce978deb0bc7a1fecbd3f8d101