Defer loading task args

Encrypted task args are no longer loaded and decrypted in codepaths that
take no care in them. This can also help to clean up broken tasks in case
the encryption key was lost.

[noissue]

CHANGES/+deferred_task_args.feature

@@ -0,0 +1,2 @@
+Improved performance when handling tasks by deferred loading of encrypted args.
+This also allows seeing and purging tasks in case the symmetric db key was lost.

pulpcore/app/models/task.py

@@ -63,6 +63,14 @@ class Meta:
         unique_together = ("task", "name")
 
 
+class TaskManager(models.Manager):
+    def get_queryset(self):
+        # Always make encrypted args deferred.
+        # This will prevent a lot of issues when the fernet key is lost.
+        # Only task workers need to be able to read these args anyway.
+        return super().get_queryset().defer("enc_args", "enc_kwargs")
+
+
 class Task(BaseModel, AutoAddObjPermsMixin):
     """
     Represents a task
@@ -88,6 +96,8 @@ class Task(BaseModel, AutoAddObjPermsMixin):
         pulp_domain (models.ForeignKey): The domain the Task is a part of
     """
 
+    objects = TaskManager()
+
     state = models.TextField(choices=TASK_CHOICES)
     name = models.TextField()
     logging_cid = models.TextField(db_index=True)
@@ -285,8 +295,10 @@ def refresh_from_db(self, using=None, fields=None, **kwargs):
         # loaded.
         if fields is not None:
             fields = set(fields)
-            deferred_fields = self.get_deferred_fields()
-            # If any deferred field is going to be loaded
+            deferred_fields = {
+                field for field in self.get_deferred_fields() if not field.startswith("enc_")
+            }
+            # If any state related deferred field is going to be loaded
             if fields.intersection(deferred_fields):
                 # then load all of them
                 fields = fields.union(deferred_fields)

pulpcore/tasking/_util.py

@@ -95,7 +95,8 @@ def perform_task(task_pk, task_working_dir_rel_path):
     signal.signal(signal.SIGUSR1, child_signal_handler)
     # All processes need to create their own postgres connection
     connection.connection = None
-    task = Task.objects.select_related("pulp_domain").get(pk=task_pk)
+    # enc_args and enc_kwargs are deferred by default but we actually want them
+    task = Task.objects.defer(None).select_related("pulp_domain").get(pk=task_pk)
     user = get_users_with_perms(task, with_group_users=False).first()
     # Isolate from the parent asyncio.
     asyncio.set_event_loop(asyncio.new_event_loop())