If somehow the COMMIT or XID event in an event group was missing, the code in
parallel replication to handle this was not sufficient, leading to server
deadlock.