When the percona-cluster charm sets an access-network but the default
unit-get address is not on that network extra shared-db relations get
executed. This is specifically a problem when running upgrades and
trying to avoid API downtime.
The root cause is that the access-network is not checked until the
SharedDBContext is consulted. But then db_joined function will
change it back to the wrong ip on subsequent runs.
This change adds a check for access-network on the relation during the
db_joined function and pushes IP selection off to get_relation_ip.
Charm helpers sync to pull in changes to get_relation_ip.
Partial-bug: #1677647
Change-Id: Ifd4e975d9abbb9f7a8b0f12c8f0a8cf8f78595b6
e94f7882cb