* charm-telegraf [1] supports a mysql plugin through the mysql-root interface.
* charm-percona-cluster has a "db-admin" endpoint that uses the mysql-root interface to configure a user for the telegraf mysql plugin
* This charm (charm-mysql-innodb-cluster) should support a similar endpoint (db-admin) so telegraf can collect metrics from each mysql server.
Note: the mysql-root interface collects the "host" attribute (see [2]), which in the case of charm-percona-cluster defaults to the VIP configured (so all 3 telegraf subordinates collect metrics from a single mysql instance). For charm-mysql-innodb-cluster, the "host" attribute should be each of the "private-address" values, so telegraf will collect information about each of the instances (1x RW mode, 2x RO mode).
The Juju configuration mode would be:
# makes telegraf a subordinate, with default plugins enabled
juju relate mysql-innodb-cluster telegraf:juju-info
# enables the mysql plugin
juju relate mysql-innodb-cluster:db-admin telegraf:mysql
# exposes the 0.0.0.0:9103/metrics service in telegraf
juju relate prometheus:target telegraf
# unrelated to this bug: share a mysql dashboard with Grafana,
# so the collected metrics by prometheus are displayed
juju relate grafana:dashboards telegraf
1. https://launchpad.net/charm-telegraf/
2. https://gist.github.com/auria/c9f23772ce2ddb5797d82204d9fabd59
Sounds like a good addition to the charm, thanks. As a side-note I think you were suggesting using private-address to illustrate your point, but the implementer should remember that private-address should not be used as its deprecated and unreliable.