Microsoft didn't publish Python graph examples for CosmosDB (yet) - and I don't think that client.QueryDocuments() is the right method to execute Gremlin statements - see below:
client = document_client.DocumentClient(endpoint, masterkey)
collection_link = 'dbs/' + database + '/colls/' + collection
collection = client.ReadCollection(collection_link)
print('Collection with id \'{0}\' was found'.format(collection['id'])
gremlin = 'g.V().count()'
docs = client.QueryDocuments(collection_link, gremlin)
print(docs)
The result is not a JSON of executed Gremlin statement (the count of nodes in the graph) - it is
< pydocumentdb.query_iterable.QueryIterable object at 0x05341EFF>
Am I missing the right function in Python CosmosDB SDK, or is it simply not there?