RING_HASH lb policy doesn't adjust load balancing weights when upstream hosts become unhealthy

[cxhiano]

Title: RING_HASH lb policy doesn't adjust load balancing weights when upstream hosts become unhealthy

Description:

According to this doc, Envoy adjusts traffic weights based on overprovisioning_factor and upstream host health status. This does not seem to be the case when lb_policy is set to RING_HASH.

Repro steps:

Configure a cluster with lb_policy set to RING_HASH and a load_assignment with endpoints across at least 2 localities, traffic distribution across localities doesn't seem to change as some upstream hosts become unhealthy.

Config:

  clusters:
  - name: test_service
    lb_policy: RING_HASH
    common_lb_config:
      locality_weighted_lb_config: {}
      healthy_panic_threshold:
        value: 1
    load_assignment:
 ...

cc: @tonya11en