Currently this is what the RLS LB policy is currently doing:

• Create an RLS control channel:

  grpc-go/balancer/rls/balancer.go

  Line 362 in 7472d57

  ctrlCh, err := newControlChannel(newCfg.lookupService, newCfg.controlChannelServiceConfig, newCfg.lookupServiceTimeout, b.bopts, backToReadyFn)

• Start a goroutine to monitor the connectivity state of the channel:

  grpc-go/balancer/rls/control_channel.go

  Line 95 in 7472d57

  go ctrlCh.monitorConnectivityState()

• The goroutine waits for the channel to become READY:

  grpc-go/balancer/rls/control_channel.go

  Line 188 in 7472d57

  cc.logger.Infof("Connectivity state is READY")

• The next time it becomes READY again, it reset backoffs:

  grpc-go/balancer/rls/control_channel.go

  Line 202 in 7472d57

  cc.backToReadyFunc()

This was a wrong assumption that once we are READY and get back to READY, we should have gone through TRANSIENT_FAILURE.

What should it should actually do:

• When the state transitions to TRANSIENT_FAILURE, record that transition
• The next time it transitions to READY, reset the backoff timeouts in all cache entries. Specifically, this means that it will reset the backoff state and cancel the pending backoff timer.

We should also update this test:

to ensure that a transition from READY to IDLE to READY does not result in backoff timeouts being reset.