The logging around chunk metadata refresh is confusing and is only done after the load has completed. This means that long refresh operations (which happen under a mutex) will block other calls, but will not output anything until the call completes.

In addition they are inconsistent between mongos and the shards.