Keystone, in a large model, gets very slow in updating endpoint information. This is because it re-fetches the same information for every iteration through the loops it uses to resolve endpoints.
It starts in "add_service_to_keystone()" which then calls "add_endpoint()" for each endpoint. This then calls "create_service_entry" which calls "manager.list_services()" (so that's a list_services() for each endpoint).
Then create_endpoint_template_vx() function then call manager.list_endpoints(). So this is multiple calls between keystone charm <-> manager.py to keystone, which results in mysql lookups, etc.
It ought to be possible to do a single call to list_services, list_endpoint()s and then use that data to decide what to do next (which is probably, just make a single call to set up the single set of endpoints that are needed). This would radically improve the performance around endpoint addition.
Fix proposed to branch: master
Review: https:/