cluster_delete fails for PXC on Xenial

[Doug Shelley]

The cluster delete scenario test is failing on Xenial for Percona XtraDB Cluster.

The cluster member nominated as the bootstrap node ends up in SHUTDOWN state but the mysql server on it isn't actually stopped. Because the guest can't stop the db server, the task manager never executes the rest of the delete logic (like nova delete etc) so the node never goes away and the task manager eventually times out.

The root causes appears to be related to this statement from the PXC installation doc [1]:
"On Debian Jessie (8.0) and Ubuntu Xenial (16.04) in order to stop or reload the node bootstrapped with service mysql bootstrap-pxc command, you’ll need to use service mysql stop-bootstrap or service mysql reload-bootstrap respectively. Also, status of the bootstrapped node should be checked with service mysql status-bootstrap."

[1] https://www.percona.com/doc/percona-xtradb-cluster/5.6/installation/apt_repo.html#apt-repo

The code currently calls "bootstrap-pxc" on the bootstrap node but will only call "stop" when trying to stop the mysql server. It has no mechanism to determine if a node was the bootstrap node and call a different form of stop.

FYI - the real solution is for Percona to create proper systemd scripts for Xenial like they did for CentOS. This issue doesn't exist on CentOS.

[Amrith Kumar (amrith)]

you probably know the most about this.