Netflix is using Spinnaker, see techblog.netflix.com/2015/11/global-continuous-de… if you have a look in a deployed instance of Spinnaker, there are options for manual testing as well as automated testing to complete the pipeline- so they probably use both.
If you have a look at how Pivotal's cloud works, they simply monitor for errors, if the number of errors rapidly increase, they automatically rollback to the previous version.