In one case of addShard, we delete all of the cluster parameters on the new shard, and then set all of our cluster parameters on that shard. If t is the number of tenants, and c is the average number of set cluster parameters per tenant, this will take on the order of tc network operations, which blows up very quickly. replaceAllClusterParameters would take an object of the form {<tenant-id-1>:

, ...} and would delete all existing cluster parameters, then set all given cluster parameters. This would mean we can do all of the above in one network operation. Some thought is required in terms of max BSON size, and what happens if we overrun it.