my.cnf not symlinked to /etc/mysql/percona-xtradb-cluster.cnf after upgrading to bionic

I have upgraded a percona 3-node cluster from trusty to xenial and from xenial to bionic.

The first upgrade appeared to complete successfully, while the one to bionic failed completely.

These are the symptoms I have observed:

* the post-series-upgrade hook hit bug 1838041. I worked around it by simply commenting out the subprocess call to mysql_upgrade:

441 # Execute the upgrade process
442 log("Running mysql_upgrade")
443 cmd = ['mysql_upgrade']
444 if _release == "xenial":
445 cmd.append('-p{}'.format(root_password()))
446 #subprocess.check_call(cmd)

* the update-alternatives entry for my.cnf was making /etc/mysql/my.cnf point to /etc/mysql/my.cnf.migrated, and not to /etc/mysql/percona-xtradb-cluster.cnf. As a consequence, config-changed hooks were rendering configurations that were never being picked up by mysqld

The update-alternatives setup is performed in the post-series-upgrade hook when upgrading to xenial:

458 if _release == "xenial":
459 # Point at the correct my.cnf
460 cmd = ["update-alternatives", "--set", "my.cnf",
461 "/etc/mysql/percona-xtradb-cluster.cnf"]
462 subprocess.check_call(cmd)

I do not know if this wasn't run when I transitioned from trusty to xenial, or if it got undone when upgrading to bionic, but it surely made the upgrade process very difficult to debug.

Charm version used: 291