Drop run count from get_stats to avoid blocking the event loop

python/apsis/runs.py

@@ -635,8 +635,16 @@ def count_runs(
         )
 
     def get_stats(self):
+        # Note: deliberately no DB count here.  Counting runs in the lookback window
+        # requires a query over the runs table, which has no index on timestamp and
+        # stores wide JSON payloads inline; the scan reads GBs and takes tens of
+        # seconds when the page cache is cold.  Since get_stats() runs on the event
+        # loop, that blocks all of Apsis.  Report only in-memory counts, which are
+        # O(1).  Use count_runs() if you need a run count from the DB.
+        # TODO: expose a total run count that doesn't require a DB query on the
+        # event loop, e.g. maintained incrementally or sampled off-thread.
         return {
-            "num_runs": len(self.__expected_runs)
-            + self.__run_db.count_runs(min_timestamp=self.__min_timestamp),
+            "num_expected_runs": len(self.__expected_runs),
+            "num_active_runs": len(self.__active_runs),
             "publisher": self.publisher.get_stats(),
         }

test/unit/test_run_store.py

@@ -195,14 +195,36 @@ def test_run_store_finished_run_not_in_memory(tmp_path):
     assert run.run_id not in expected_map
 
 
-def test_run_store_num_runs_no_double_count(tmp_path):
-    """get_stats()['num_runs'] must count each physical run once."""
+def test_run_store_count_runs_no_double_count(tmp_path):
+    """count_runs() must count each physical run once."""
     store = _make_store(tmp_path)
     run = Run(Instance("job", {}), expected=True)
     _schedule(store, run)
     _transition(store, run, State.waiting)
 
-    assert store.get_stats()["num_runs"] == 1
+    assert store.count_runs() == 1
+
+
+def test_run_store_get_stats_does_not_query_db(tmp_path):
+    """
+    get_stats() must not count runs in the DB.
+
+    That query scans the runs table, which blocks the event loop for tens of
+    seconds on a large database.  See RunStore.get_stats().
+    """
+    store = _make_store(tmp_path)
+    run = Run(Instance("job", {}), expected=True)
+    _schedule(store, run)
+    _transition(store, run, State.waiting)
+
+    def fail(*args, **kwargs):
+        raise AssertionError("get_stats() must not count runs in the DB")
+
+    store._RunStore__run_db.count_runs = fail
+
+    stats = store.get_stats()
+    assert stats["num_active_runs"] == 1
+    assert "num_runs" not in stats
 
 
 def test_run_store_query_since_filters_expected(tmp_path):