Fix race condition in subscription TAP test 021_twophase

michaelpq · michaelpq · commit 6ea812f4dc63 · 2025-05-26T17:28:40.000+09:00
The test did not wait for all the subscriptions to have caught up when dropping the subscription "tab_copy". In a slow environment, it could be possible for the replay of the COMMIT PREPARED transaction "mygid" to not be confirmed yet, causing one prepared transaction to be left around before moving to the next steps of the test. One failure noticed is a transaction found in pg_prepared_xacts for the cases where copy_data = false and two_phase = true, but there should be none after dropping the subscription. As an extra safety measure, a check is added before dropping the subscription, scanning pg_prepared_xacts to make sure that no prepared transactions are left once both subscriptions have caught up. Issue introduced by a8fd13c, fixing a problem similar to eaf5321. Per buildfarm member kestrel. Author: Vignesh C <vignesh21@gmail.com> Reviewed-by: Amit Kapila <amit.kapila16@gmail.com> Discussion: https://postgr.es/m/CALDaNm329QaZ+bwU--bW6GjbNSZ8-38cDE8QWofafub7NV67oA@mail.gmail.com Backpatch-through: 15
diff --git a/src/test/subscription/t/021_twophase.pl b/src/test/subscription/t/021_twophase.pl
@@ -373,7 +373,14 @@
   $node_publisher->safe_psql('postgres', "SELECT count(*) FROM tab_copy;");
 is($result, qq(6), 'publisher inserted data');
 
+# Wait for both subscribers to catchup
 $node_publisher->wait_for_catchup($appname_copy);
+$node_publisher->wait_for_catchup($appname);
+
+# Make sure there are no prepared transactions on the subscriber
+$result = $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts;");
+is($result, qq(0), 'should be no prepared transactions on subscriber');
 
 $result =
   $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM tab_copy;");
