Ensure that m2m field indexes are created when using add_field (baserow#4491)

Author: bram2w

backend/src/baserow/contrib/database/db/schema.py

@@ -355,6 +355,41 @@ def delete_model(self, model):
             ):
                 self.deferred_sql.remove(sql)
 
+    def ensure_single_column_index(self, model, field):
+        """
+        Ensure an index exists on model(field.column). Safe to call repeatedly.
+        """
+
+        # If any index/constraint exists that is backed by an index on exactly this
+        # column, don't create another.
+        if self._constraint_names(model, [field.column], index=True):
+            return
+
+        stmt = self._create_index_sql(model, fields=[field])
+        self.execute(stmt, params=None)
+
+    def ensure_m2m_field_indexes(self, field):
+        if field.many_to_many and field.remote_field.through._meta.auto_created:
+            through = field.remote_field.through
+            # Ensure the two FK indexes exist (especially important for the "reverse"
+            # FK). m2m_field_name() / m2m_reverse_field_name() return the
+            # through-field names.
+            for through_field_name in (
+                field.m2m_field_name(),
+                field.m2m_reverse_field_name(),
+            ):
+                fk_field = through._meta.get_field(through_field_name)
+                self.ensure_single_column_index(through, fk_field)
+
+    def add_field(self, model, field):
+        return_value = super().add_field(model, field)
+        # Using the `create_model` to create a Baserow table, like what we do on
+        # `import_serialize` does actually create the indexes of the through table.
+        # However, when using `add_field` it does not. The code below will make sure
+        # that the needed indexes are created.
+        self.ensure_m2m_field_indexes(field)
+        return return_value
+
 
 @contextlib.contextmanager
 def safe_django_schema_editor(atomic=True, name=None, classes=None, **kwargs):

backend/src/baserow/contrib/database/migrations/0202_table_missing_m2m_indexes_added.py

@@ -0,0 +1,21 @@
+# Generated by Django 5.0.14 on 2025-12-19 21:52
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+    dependencies = [
+        ("database", "0201_increase_pendingsearchvalueupdate_statistics"),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name="table",
+            name="missing_m2m_indexes_added",
+            field=models.BooleanField(
+                db_default=False,
+                default=True,
+                help_text="Indicates whether potentially missing m2m foreign key indexes have been added.",
+            ),
+        ),
+    ]

backend/src/baserow/contrib/database/table/models.py

@@ -889,6 +889,20 @@ class Table(
         null=True,
         help_text="Indicates whether the table has had the field_rules_are_valid column added.",
     )
+    # The m2m indexes of the foreign keys were not added before because the
+    # `schema_editor.add_field` does not add them. The `schema_editor.create_model`
+    # does add those. This problem has been addressed, but there are tables out there
+    # without those indexes.
+    missing_m2m_indexes_added = models.BooleanField(
+        # The `db_default` must be false because this is used when an entry is created
+        # no default value is set. This is what happens when the field index changes
+        # are not yet deployed, so index does not exist.
+        db_default=False,
+        # However, if the field index changes are deployed, this default value is used,
+        # and in that case, the index has been applied.
+        default=True,
+        help_text="Indicates whether potentially missing m2m foreign key indexes have been added.",
+    )
 
     class Meta:
         ordering = ("order",)

backend/src/baserow/contrib/database/table/tasks.py

@@ -127,6 +127,25 @@ def setup_created_by_and_last_modified_by_column(self, table_id: int):
         TableHandler().create_created_by_and_last_modified_by_fields(table)
 
 
+@app.task(bind=True, queue="export")
+def setup_m2m_field_indexes_if_not_exist(self, table_id: int):
+    from baserow.contrib.database.db.schema import safe_django_schema_editor
+    from baserow.contrib.database.table.handler import TableHandler
+
+    with transaction.atomic():
+        table = TableHandler().get_table_for_update(table_id)
+        model = table.get_model()
+        fields = model.get_fields()
+
+        with safe_django_schema_editor(atomic=False) as schema_editor:
+            for field in fields:
+                model_field = model._meta.get_field(field.db_column)
+                schema_editor.ensure_m2m_field_indexes(model_field)
+
+        table.missing_m2m_indexes_added = True
+        table.save(update_fields=["missing_m2m_indexes_added"])
+
+
 @app.task(bind=True)
 def update_table_usage(self, table_id: int, row_count: int = 0):
     from baserow.contrib.database.table.handler import TableUsageHandler

backend/src/baserow/contrib/database/views/signals.py

@@ -80,6 +80,7 @@ def update_view_index_if_view_group_by_changes(sender, view_group_by, **kwargs):
 def view_loaded_create_indexes_and_columns(sender, view, table_model, **kwargs):
     from baserow.contrib.database.table.tasks import (
         setup_created_by_and_last_modified_by_column,
+        setup_m2m_field_indexes_if_not_exist,
     )
     from baserow.contrib.database.views.handler import ViewIndexingHandler
 
@@ -88,6 +89,8 @@ def view_loaded_create_indexes_and_columns(sender, view, table_model, **kwargs):
     table = view.table
     if not table.last_modified_by_column_added or not table.created_by_column_added:
         setup_created_by_and_last_modified_by_column.delay(table_id=view.table.id)
+    if not table.missing_m2m_indexes_added:
+        setup_m2m_field_indexes_if_not_exist.delay(table_id=view.table_id)
 
 
 @receiver(field_signals.fields_type_changed)

changelog/entries/unreleased/bug/ensure_m2m_indexes_are_set.json

@@ -0,0 +1,9 @@
+{
+  "type": "bug",
+  "message": "Ensure m2m field indexes are all set.",
+  "issue_origin": "github",
+  "issue_number": null,
+  "domain": "database",
+  "bullet_points": [],
+  "created_at": "2025-12-19"
+}