On startup time, the Balancer needs to wait for the completion of all the outstanding chunk migrations executed across the cluster before generating new requests.

In the past this was achieved by persisting in config.migrations a recovery doc for every moveChunk requested and repeating the command for every unanswered one upon a config server step down/crash.

The same approach cannot be adopted to recover moveRange commands (which semantics are slightly different) without causing undesired side effects when the cluster gets downgraded.

A more generic alternative could be built upon a new command that allows a client to wait until the contacted shard is no longer involved in any migration activity.