huge overhead of configuration refreshing effects local rate limit and health checker

The main thread of Envoy takes the responsibility to handle the xDS request/response, stats flushing, health checking, etc. These are super heavy works in an environment with much routes and services.

For example, the timers of local rate limit may be delayed for a long time when the main thread is occupied by a configuration refreshing. The health checker cannot find unhealthy instances or bring healthy instances back in timely.

I know it's reasonable cost/time to handle large configuration and update configuration dynamically, but it's still a problem if that effects other features.