fix: query count discrepancy between MySQL and SQLite

Claude explanation: In `_create_side_effects_for_change_log`, `change_log.records.all()` has no `ORDER BY`. On SQLite, this returns records in insertion order (PK order) — which is root-first, since `publish_from_drafts` inserts the initial draft first, then its dependencies layer by layer. On MySQL/InnoDB, the query uses the `oel_plr_uniq_pl_publishable` unique index on (`publish_log`, `entity`) to service the `WHERE publish_log_id = ?` filter, which returns records in `entity_id` order — i.e. leaf-first, because `child_entity2` has a lower entity PK than `parent_of_two`, `grandparent`, etc.

The `processed_entity_ids` optimization on line 899–904 is order-sensitive: when you process leaf records first, `processed_entity_ids` isn't yet populated with their ancestors' IDs, so the while loop traverses all the way up the tree redundantly on each leaf

Author: bradenmacdonald

src/openedx_content/applets/publishing/api.py

@@ -810,7 +810,7 @@ def _create_side_effects_for_change_log(change_log: DraftChangeLog | PublishLog)
     # It also guards against infinite parent-child relationship loops, though
     # those aren't *supposed* to happen anyhow.
     processed_entity_ids: set[int] = set()
-    for original_change in change_log.records.all():
+    for original_change in change_log.records.order_by("pk"):
         affected_by_original_change = branch_cls.objects.filter(
             version__dependencies=original_change.entity
         )

tests/openedx_content/applets/containers/test_api.py

@@ -1273,11 +1273,7 @@ def test_deep_publish_log(
         ]
 
     # Publish great_grandparent. Should publish the whole tree.
-    # FIXME: this is using 144 queries on MySQL vs 132 on SQLite. The disparity only happens when publishing four levels
-    # of hierarchy at once (e.g. publishing a Section-->Component or Great-Grandparent-->Child). i.e. if you change this
-    # to publish "grandparent" instead of "great_grandparent" or you pre-publish the leaves (child entities), there is
-    # no disparity between the databases.
-    with django_assert_num_queries(133):
+    with django_assert_num_queries(132):
         publish_log = publish_entity(great_grandparent)
         assert list(publish_log.records.order_by("entity__pk").values_list("entity__key", flat=True)) == [
             "child_entity2",
@@ -1286,16 +1282,6 @@ def test_deep_publish_log(
             "grandparent",
             "great_grandparent",
         ]
-        assert list(
-            PublishSideEffect.objects.filter(cause__publish_log=publish_log)
-            .values_list("cause__entity__key", "effect__entity__key")
-            .order_by("cause__entity__key")  # Note: order_by("pk") is different on MySQL vs SQLite
-        ) == [
-            ("child_entity2", "parent_of_two"),
-            ("grandparent", "great_grandparent"),
-            ("parent_of_three", "grandparent"),
-            ("parent_of_two", "grandparent"),
-        ]
 
 
 # get_entities_in_container

tests/openedx_content/applets/sections/test_api.py

@@ -154,7 +154,7 @@ def test_section_queries(self) -> None:
         """
         with self.assertNumQueries(37):
             section = self.create_section_with_subsections([self.subsection_1, self.subsection_2_v1])
-        with self.assertNumQueries(169):  # TODO: this seems high? FIXME: this is 181 on MySQL but 169 on SQLite?
+        with self.assertNumQueries(169):
             content_api.publish_from_drafts(
                 self.learning_package.id,
                 draft_qset=content_api.get_all_drafts(self.learning_package.id).filter(entity=section.pk),