vacuum_eventless_issuetags: tune batch-size

See #134

tags/tasks.py

@@ -103,7 +103,11 @@ def vacuum_eventless_issuetags(min_id=0):
     #
     # This task aims to reconcile that, in a delayed and resumable fashion.
 
-    BATCH_SIZE = 512  # close to 500, and a multiple of 64
+    # Empirically determined: at this size, each batch is approx .3s (local dev, sqlite); Note that we're "nearer to the
+    # edge of the object-graph" than for e.g. even-retention, so we can both afford bigger batches (less cascading
+    # effects per item) as well as need bigger batches (because there are more expected items in a fanning-out
+    # object-graph).
+    BATCH_SIZE = 2048
 
     # Community wisdom (says ChatGPT, w/o source): queries with dozens of OR clauses can slow down significantly. 64 is
     # a safe, batch size that avoids planner overhead and keeps things fast across databases.

tags/tests.py

@@ -323,10 +323,10 @@ class VacuumEventlessIssueTagsTestCase(TransactionTestCase):
 
     def test_many_tags_spanning_chunks(self):
         event = create_event(self.project, issue=self.issue)
-        store_tags(event, self.issue, {f"key-{i}": f"value-{i}" for i in range(512 + 1)})  # bigger than BATCH_SIZE
+        store_tags(event, self.issue, {f"key-{i}": f"value-{i}" for i in range(2048 + 1)})  # bigger than BATCH_SIZE
 
         # check setup: all issue tags are there
-        self.assertEqual(IssueTag.objects.filter(issue=self.issue).count(), 513)
+        self.assertEqual(IssueTag.objects.filter(issue=self.issue).count(), 2048 + 1)
 
         event.delete_deferred()
         vacuum_eventless_issuetags()