[rdo-list] Strange API latency issue

Hello, Wondering if anyone can shed any light on the following: I'm seeing a ~2 second delay in some API responses using Mitaka trunk. E.g. $ openstack endpoint list --timing <list of endpoints> +----------------------------------------------------------------------------+-----------+ | URL | Seconds | +----------------------------------------------------------------------------+-----------+ | GET http://PUBLIC_VIP:35357/v3 | 0.009938 | | POST http://PUBLIC_VIP:35357/v3/auth/tokens | 0.257309 | | POST http://PUBLIC_VIP:35357/v3/auth/tokens | 0.171096 | | GET http://PUBLIC_VIP:35357/ | 1.741362 | | GET http://PUBLIC_VIP:35357/v3/endpoints | 0.026326 | | GET http://PUBLIC_VIP:35357/v3/services/b28a9d7986684c52905ea2f45fd6dffe | 0.023991 | | GET http://PUBLIC_VIP:35357/v3/services/1fb856a569714d44b787868151f995b6 | 0.025029 | | GET http://PUBLIC_VIP:35357/v3/services/b28a9d7986684c52905ea2f45fd6dffe | 0.028141 | | GET http://PUBLIC_VIP:35357/v3/services/a857a42350e74106881534bcb19ebaf7 | 1.995946 | | GET http://PUBLIC_VIP:35357/v3/services/442c73aabbb141e6839002665c35579f | 0.022442 | | GET http://PUBLIC_VIP:35357/v3/services/d1a85a3b47de41aeb10a4ae087048a20 | 0.037483 | | GET http://PUBLIC_VIP:35357/v3/services/25c587016a16407fb682e36fe3f62532 | 1.939549 | | GET http://PUBLIC_VIP:35357/v3/services/1fb856a569714d44b787868151f995b6 | 0.031643 | | GET http://PUBLIC_VIP:35357/v3/services/8d072c313df44f639f9d30cdd6165621 | 0.032537 | | GET http://PUBLIC_VIP:35357/v3/services/442c73aabbb141e6839002665c35579f | 0.022227 | | GET http://PUBLIC_VIP:35357/v3/services/25c587016a16407fb682e36fe3f62532 | 1.974827 | | GET http://PUBLIC_VIP:35357/v3/services/734da843f77c4e4689b0772dbd6960f3 | 2.15859 | | GET http://PUBLIC_VIP:35357/v3/services/8d072c313df44f639f9d30cdd6165621 | 0.023067 | <snip> | Total | 11.086058 | +----------------------------------------------------------------------------+-----------+ 11 seconds is pretty good, can be anything up to 90 seconds. I have another installation which works fine, all response in milliseconds. I have deployed with network isolation and predictable IPs. The only situation where this does not seem to exhibit is if I disable the external network entirely using controll-no-external.yaml. -- Chris