diff --git a/django/contrib/gis/utils/layermapping.py b/django/contrib/gis/utils/layermapping.py
index ad7b961883..3c78085489 100644
--- a/django/contrib/gis/utils/layermapping.py
+++ b/django/contrib/gis/utils/layermapping.py
@@ -101,7 +101,8 @@ class LayerMapping:
         self.layer = self.ds[layer]
 
         self.using = using if using is not None else router.db_for_write(model)
-        self.spatial_backend = connections[self.using].ops
+        connection = connections[self.using]
+        self.spatial_backend = connection.ops
 
         # Setting the mapping & model attributes.
         self.mapping = mapping
@@ -113,7 +114,7 @@ class LayerMapping:
 
         # Getting the geometry column associated with the model (an
         # exception will be raised if there is no geometry column).
-        if connections[self.using].features.supports_transform:
+        if connection.features.supports_transform:
             self.geo_field = self.geometry_field()
         else:
             transform = False
diff --git a/django/db/models/fields/related_descriptors.py b/django/db/models/fields/related_descriptors.py
index dba9e14c86..cb77a0c476 100644
--- a/django/db/models/fields/related_descriptors.py
+++ b/django/db/models/fields/related_descriptors.py
@@ -1100,8 +1100,8 @@ def create_forward_many_to_many_manager(superclass, rel, reverse):
             # user-defined intermediary models as they could have other fields
             # causing conflicts which must be surfaced.
             can_ignore_conflicts = (
-                connections[db].features.supports_ignore_conflicts and
-                self.through._meta.auto_created is not False
+                self.through._meta.auto_created is not False and
+                connections[db].features.supports_ignore_conflicts
             )
             # Don't send the signal when inserting duplicate data row
             # for symmetrical reverse entries.
diff --git a/django/db/models/query.py b/django/db/models/query.py
index 661ebd1f24..fc36945f56 100644
--- a/django/db/models/query.py
+++ b/django/db/models/query.py
@@ -498,7 +498,6 @@ class QuerySet:
         if not objs:
             return objs
         self._for_write = True
-        connection = connections[self.db]
         opts = self.model._meta
         fields = opts.concrete_fields
         objs = list(objs)
@@ -521,6 +520,7 @@ class QuerySet:
                 returned_columns = self._batched_insert(
                     objs_without_pk, fields, batch_size, ignore_conflicts=ignore_conflicts,
                 )
+                connection = connections[self.db]
                 if connection.features.can_return_rows_from_bulk_insert and not ignore_conflicts:
                     assert len(returned_columns) == len(objs_without_pk)
                 for obj_without_pk, results in zip(objs_without_pk, returned_columns):
@@ -551,9 +551,10 @@ class QuerySet:
             return 0
         # PK is used twice in the resulting update query, once in the filter
         # and once in the WHEN. Each field will also have one CAST.
-        max_batch_size = connections[self.db].ops.bulk_batch_size(['pk', 'pk'] + fields, objs)
+        connection = connections[self.db]
+        max_batch_size = connection.ops.bulk_batch_size(['pk', 'pk'] + fields, objs)
         batch_size = min(batch_size, max_batch_size) if batch_size else max_batch_size
-        requires_casting = connections[self.db].features.requires_casted_case_in_updates
+        requires_casting = connection.features.requires_casted_case_in_updates
         batches = (objs[i:i + batch_size] for i in range(0, len(objs), batch_size))
         updates = []
         for batch_objs in batches:
@@ -1308,13 +1309,14 @@ class QuerySet:
         """
         Helper method for bulk_create() to insert objs one batch at a time.
         """
-        if ignore_conflicts and not connections[self.db].features.supports_ignore_conflicts:
+        connection = connections[self.db]
+        if ignore_conflicts and not connection.features.supports_ignore_conflicts:
             raise NotSupportedError('This database backend does not support ignoring conflicts.')
-        ops = connections[self.db].ops
+        ops = connection.ops
         max_batch_size = max(ops.bulk_batch_size(fields, objs), 1)
         batch_size = min(batch_size, max_batch_size) if batch_size else max_batch_size
         inserted_rows = []
-        bulk_return = connections[self.db].features.can_return_rows_from_bulk_insert
+        bulk_return = connection.features.can_return_rows_from_bulk_insert
         for item in [objs[i:i + batch_size] for i in range(0, len(objs), batch_size)]:
             if bulk_return and not ignore_conflicts:
                 inserted_rows.extend(self._insert(
@@ -1523,10 +1525,8 @@ class RawQuerySet:
     def iterator(self):
         # Cache some things for performance reasons outside the loop.
         db = self.db
-        compiler = connections[db].ops.compiler('SQLCompiler')(
-            self.query, connections[db], db
-        )
-
+        connection = connections[db]
+        compiler = connection.ops.compiler('SQLCompiler')(self.query, connection, db)
         query = iter(self.query)
 
         try:
diff --git a/django/db/models/sql/query.py b/django/db/models/sql/query.py
index 388c60545f..e9fa0d7547 100644
--- a/django/db/models/sql/query.py
+++ b/django/db/models/sql/query.py
@@ -2332,10 +2332,10 @@ class Query(BaseExpression):
         # used. The proper fix would be to defer all decisions where
         # is_nullable() is needed to the compiler stage, but that is not easy
         # to do currently.
-        return (
-            connections[DEFAULT_DB_ALIAS].features.interprets_empty_strings_as_nulls and
-            field.empty_strings_allowed
-        ) or field.null
+        return field.null or (
+            field.empty_strings_allowed and
+            connections[DEFAULT_DB_ALIAS].features.interprets_empty_strings_as_nulls
+        )
 
 
 def get_order_dir(field, default='ASC'):