grpc / grpc-go

#5469 recommends an audit of existing LB policies to ensure that they update their pickers synchronously upon receipt of a configuration update.

xds_cluster_impl_experimental does not update its picker synchronously.

Configuration updates are handled here:

grpc-go/xds/internal/balancer/clusterimpl/clusterimpl.go

Line 208 in 0561c78

func (b *clusterImplBalancer) UpdateClientConnState(s balancer.ClientConnState) error {
In the above function, the LB policy
- initially checks for some error conditions
- applies new load reporting configuration, if any
- switches to a new child policy, if the child policy name has changed
- pushes the newly received load balancer configuration onto a channel here:
  
  grpc-go/xds/internal/balancer/clusterimpl/clusterimpl.go
  
  Line 253 in 0561c78
  
  b.pickerUpdateCh.Put(newConfig)
- and returns from UpdateClientConnState
The newly received configuration update is processed here asynchronously:

grpc-go/xds/internal/balancer/clusterimpl/clusterimpl.go

Line 475 in 0561c78

case *LBConfig:
And a new picker is pushed upwards here:

grpc-go/xds/internal/balancer/clusterimpl/clusterimpl.go

Line 479 in 0561c78

b.ClientConn.UpdateState(balancer.State{

This needs to be changed to return a picker synchronously upon receipt of a configuration update.

@easwars what's the need of putting the new lb config to the channel and processing it asynchronously in run method, why can't we just update it there itself and also update the picker below that?

clusterimpl: update picker synchronously upon receipt of configuration update