All, We (at ShapeBlue) did a gap analysis to figure if current set of smoke and component tests sufficiently test the VR functionality for the regressions. I have posted the analysis at [1]. I went through the test suites, and listed down all the tests that touch virtual router functionality. There is listing of general observations on grey areas.
One particular area where there were no tests, was related multiple public IP’s from different public IP ranges associated with a network. From 4.6, all the way to master only IP’s from one public IP range (eth2 on the VR) is working, any network services on the public IP’s on eth3, eth4 etc on VR are not functional. This is a common use case and is broken for last few releases. Bug in this area are reported [2] and PR is yet to be merged [3]. I will be work on the patch to get this fixed in LTS. I have also added Marvin tests [4] cover multiple public IP scenarios. Given the flexibility and rich set of network functionality is CloudStack, we could catch regression only if we have good test suite. If there are any other areas related to virtual router functionality that you see there are significant gaps, please chime in share your thoughts or add the the wiki. [1] https://cwiki.apache.org/confluence/display/CLOUDSTACK/Virtual+Router%3A+Smoke+and+Integration+tests+gap+analysis [2] https://issues.apache.org/jira/browse/CLOUDSTACK-9339 [3] https://github.com/apache/cloudstack/pull/1519 [4] https://github.com/murali-reddy/cloudstack/commit/0b6fbc29fcadb39b08d0050ca473680a614dfab4