Nova strips out the path from the URL for any glance server in the "api_servers" configuration option. This prevents the use of <host>/image as part of the location for glance to be specified.
the code in "nova.image.glance" for "get_api_servers" silently drops the path on the floor.
As you can see here:
def get_api_servers():
"""Shuffle a list of CONF.glance.api_servers and return an iterator
that will cycle through the list, looping around to the beginning
if necessary.
"""
api_servers = []
configured_servers = (['%s:%s' % (CONF.glance.host, CONF.glance.port)]
if CONF.glance.api_servers is None
else CONF.glance.api_servers)
for api_server in configured_servers:
if '//' not in api_server:
api_server = 'http://' + api_server
o = urlparse.urlparse(api_server)
port = o.port or 80
host = o.netloc.rsplit(':', 1)[0]
if host[0] == '[' and host[-1] == ']':
host = host[1:-1]
use_ssl = (o.scheme == 'https')
api_servers.append((host, port, use_ssl))
random.shuffle(api_servers)
return itertools.cycle(api_servers)
Urlparse.urlparse splits out as follows:
>>> urlparse.urlparse("http://test.com/images")
ParseResult(scheme='http', netloc='test.com', path='/images', params='', query='', fragment='')
This means that glance *must* be on the root of wherever it lives from nova's perspective. We should not prevent services from living outside of the "root" of the path / ignoring proper URL schemes.
It looks like glanceclient may make it exceedingly difficult to add in a path-prefix in the way nova is consuming it.