Current pilot has regressed to not using RDS, which has the unfortunate effect of flushing envoy caches in mixerfiler on any config change. We should either resurrect RDS to preserve the caches, or share the caches between workers in envoy. The second option is more technically challenging since that requires locking discipline and can cause contention. The simpler option is to simply use RDS just like it was before the v2 change.

cc @louiscryan @rshriram