It could happen that the session used for executing a chunk migration get killed through the killSessions command. If this happens after the migration get committed, but before the range deletion task get scheduled (e.g. here), we will leave the chunk migration document (status: committed) on disk and most most importantly we will not mark the range deletion task as ready to be processed and thus orphan documents will not be cleaned. This situation will last until the donor shard receive an operation for the very same namespace or until the next replicaset failover of the donor shard.