Comment 2 for bug 1721200

<tl;dr> I really don't think this is a bug in the keystone charm.</tl;dr>

This bug can be reproduced a 100% of the time by joining keystone to lp:nrpe-charm
with juju 1. If "juju resolved --retry nrpe/0" is then run the hook again errors
but if a debug-hooks session is running on nrpe/0 and nrpe-external-master-relation-joined
is run by hand the hook completes succesfully.

The charm can also be fixed by updating charmhlpers in the nrpe charm with the following patch:

diff --git a/hooks/charmhelpers/core/hookenv.py b/hooks/charmhelpers/core/hookenv.py
index 899722f..1a1f1fc 100644
--- a/hooks/charmhelpers/core/hookenv.py
+++ b/hooks/charmhelpers/core/hookenv.py
@@ -516,7 +516,8 @@ def _metadata_unit(unit):
     joineddir = os.path.join(basedir, unitdir, 'charm', 'metadata.yaml')
     if not os.path.exists(joineddir):
         return None
- with open(joineddir) as md:
+ with open(joineddir, 'r', encoding='utf-8') as md:
+ log('_metadata_unit: Loading {} '.format(joineddir))
         return yaml.safe_load(md)

However it would need some testing, particularly where there are multiple subordinates connected to the same principle.

I am going to mark this as a bug against lp:nrpe-charm as that is where it is surfacing but
it might equally be a bug against juju 1.25.X or charm-helpers