Fix rare issue in basebackup detection.

arssher · arssher · commit 36f5b595014f · 2021-03-16T23:54:54.000+03:00
diff --git a/src/state.c b/src/state.c
@@ -3696,13 +3696,14 @@ is_basebackuped(MtmConfig *mtm_cfg)
 	}
 	CommitTransactionCommand();
 
-	if (n_missing_slots == 0)
-		return false;
-	else if (n_missing_slots == mtm_cfg->n_nodes - 1) /* n_nodes includes me */
-		return true;
-	else
-		mtm_log(ERROR, "Missing %d replication slots out of %d",
-				n_missing_slots, mtm_cfg->n_nodes);
+	/*
+	 * XXX: we will be confused here if user accidently drops mm slot, it
+	 * would be nicer to have more accurate detector. Note that we can't check
+	 * 'n_missing_slots == n_nodes - 1' because 1) new node still doesn't know
+	 * its id; 2) it itself might or might not be initialized at donor on the
+	 * moment of backup.
+	 */
+	return n_missing_slots != 0;
 }
 
 static void
diff --git a/t/010_add_stop_node.pl b/t/010_add_stop_node.pl
@@ -101,8 +101,9 @@
 {
 	$cluster->pgbench(0, ('-N', '-n', -t => '100') );
 }
-# ensure monitor creates slot for new node on donor which will be used by
-# basebackup before proceeding
+# Ensure monitor creates slot for new node on donor. We don't use it for
+# basebackup anymore, but this is still a good idea (it would be even better to
+# wait for logical slot creation too).
 $cluster->poll_query_until(0, "select exists(select * from pg_replication_slots where slot_name = 'mtm_filter_slot_${new_node_id}');")
     or croak "timed out waiting for slot creation";
 my $end_lsn = $cluster->backup_and_init(0, $new_node_off, $new_node_id);

Original file line number	Diff line number	Diff line change
`@@ -101,8 +101,9 @@`
`101`	`101`	`{`
`102`	`102`	`$cluster->pgbench(0, ('-N', '-n', -t => '100') );`
`103`	`103`	`}`
`104`		`-# ensure monitor creates slot for new node on donor which will be used by`
`105`		`-# basebackup before proceeding`
	`104`	`+# Ensure monitor creates slot for new node on donor. We don't use it for`
	`105`	`+# basebackup anymore, but this is still a good idea (it would be even better to`
	`106`	`+# wait for logical slot creation too).`
`106`	`107`	`$cluster->poll_query_until(0, "select exists(select * from pg_replication_slots where slot_name = 'mtm_filter_slot_${new_node_id}');")`
`107`	`108`	`or croak "timed out waiting for slot creation";`
`108`	`109`	`my $end_lsn = $cluster->backup_and_init(0, $new_node_off, $new_node_id);`