Unnecessary long sleep() in innobackupex leads to FTWRL taking too long

sub mysql_lockall() {
...
        mysql_send "FLUSH TABLES WITH READ LOCK;";
...
        mysql_send "COMMIT;";
...
        write_binlog_info;
...
        write_galera_info if $option_galera_info;
...
}

sub write_binlog_info {
...
     mysql_send 'SHOW MASTER STATUS\G';
...
}

sub write_galera_info {
...
     mysql_send "SHOW STATUS LIKE 'wsrep_local_state_uuid';";
...
      mysql_send "SHOW STATUS LIKE 'wsrep_last_committed';";
...
}

sub write_slave_info {
...
      mysql_send 'SHOW SLAVE STATUS\G';
...
}

sub mysql_unlockall {
...
     mysql_send "UNLOCK TABLES;";
...
}

sub mysql_send {
...
     mysql_ping(0);
...
}

sub mysql_ping {
...
              sleep 2;
...
}

Which means, for a PXC node, there are 7 mysql_ping calls amounting to 14 seconds of sleep time under FTWRL.

Possible fixes in the order of ascending complexity:

1. Decrease the sleep value and use usleep().
2. Make innobackupex use a real DB interface
3. Get rid of innobackupex by merging it with the xtrabackup binary.