diff --git a/data/model/legacy.py b/data/model/legacy.py
index 34be41491..1fb719b00 100644
--- a/data/model/legacy.py
+++ b/data/model/legacy.py
@@ -1148,6 +1148,8 @@ def __translate_ancestry(old_ancestry, translations, repository, username, prefe
 
     return translations[old_id]
 
+  # TODO: PERFORMANCE IMPROVMENT: Select all the ancestor Docker IDs in a single query. The values
+  # are retrieved in the above translate call.
   old_ids = [int(id_str) for id_str in old_ancestry.split('/')[1:-1]]
   new_ids = [str(translate_id(old_id)) for old_id in old_ids]
   return '/%s/' % '/'.join(new_ids)
@@ -1164,6 +1166,9 @@ def _create_storage(location_name):
 def find_create_or_link_image(docker_image_id, repository, username, translations,
                               preferred_location):
   with config.app_config['DB_TRANSACTION_FACTORY'](db):
+    # TODO PERF IMPROVEMENT: Also make this only lookup the image directly, rather than
+    # joining image and image storage, etc. The common case doesn't need that information
+    # (but other callers might, so double check)
     repo_image = get_repo_image(repository.namespace_user.username, repository.name,
                                 docker_image_id)
     if repo_image:
diff --git a/endpoints/index.py b/endpoints/index.py
index 1c8e551a0..7394671af 100644
--- a/endpoints/index.py
+++ b/endpoints/index.py
@@ -228,6 +228,8 @@ def create_repository(namespace, repository):
                               for desc in image_descriptions])
   new_repo_images = dict(added_images)
 
+  # TODO PERF IMPROVEMENT: Doesn't need the locations OR the imagestorage, so just select the images
+  # directly. Also use a set here.
   for existing in model.get_repository_images(namespace, repository):
     if existing.docker_image_id in new_repo_images:
       added_images.pop(existing.docker_image_id)
diff --git a/endpoints/registry.py b/endpoints/registry.py
index 48943bf4c..85b4b7963 100644
--- a/endpoints/registry.py
+++ b/endpoints/registry.py
@@ -151,6 +151,9 @@ def get_image_layer(namespace, repository, image_id, headers):
         return resp
 
       profile.debug('Streaming layer data')
+
+      # TODO: DATABASE: We should disconnect from the database here, so that
+      # we're not holding the DB handle during this long download.
       return Response(store.stream_read(repo_image.storage.locations, path), headers=headers)
     except (IOError, AttributeError):
       profile.debug('Image not found')
@@ -212,6 +215,9 @@ def put_image_layer(namespace, repository, image_id):
   h, sum_hndlr = checksums.simple_checksum_handler(json_data)
   sr.add_handler(sum_hndlr)
 
+  # TODO: DATABASE: We should disconnect from the database here and reconnect AFTER, so that
+  # we're not holding the DB handle during this long upload.
+
   # Stream write the data to storage.
   store.stream_write(repo_image.storage.locations, layer_path, sr)
 
diff --git a/endpoints/verbs.py b/endpoints/verbs.py
index b410251d3..fa3bbe574 100644
--- a/endpoints/verbs.py
+++ b/endpoints/verbs.py
@@ -24,6 +24,8 @@ logger = logging.getLogger(__name__)
 def _open_stream(namespace, repository, tag, synthetic_image_id, image_json, image_list):
   store = Storage(app)
 
+  # TODO: PERFORMANCE: configure DB, load the images, cache them, then disconnect.
+
   def get_next_image():
     for current_image_id in image_list:
       yield model.get_repo_image(namespace, repository, current_image_id)
@@ -55,6 +57,8 @@ def _write_synthetic_image_to_storage(linked_storage_uuid, linked_locations, que
 
   queue_file.add_exception_handler(handle_exception)
 
+  # TODO: PERFORMANCE: disconnect from the DB and reconnect once the stream write finishes or on
+  # error.
   image_path = store.image_layer_path(linked_storage_uuid)
   store.stream_write(linked_locations, image_path, queue_file)
   queue_file.close()
@@ -132,6 +136,9 @@ def get_squashed_tag(namespace, repository, tag):
     storage_args = (derived.uuid, derived.locations, storage_queue_file)
     QueueProcess.run_process(_write_synthetic_image_to_storage, storage_args, finished=_cleanup)
 
+    # TODO: PERFORMANCE: close the database handle here for this process before we send the long
+    # download.
+
     # Return the client's data.
     return send_file(client_queue_file)