It appears that there are concurrency issues when the server
is sending frequent config updates, as is the case during
a rebalance. Config updates trigger the client to destroy its
socket pool after first creating a new one.
This change briefly delays the destruction of the old pool
so that any in progress operations may complete.