I don't fully understand your question. Are you asking if we would override the public key value associated with the device id if one already existed in ddb? The answer is yes. But we shouldn't have multiple public keys associated with a single device

yes, sorry, could have annotated this better. basically, we have a map attribute in the users table called devices. nested within that, we have a bunch of maps attributes for each device. each of these inner maps contains a mapping from "userPublicKey" to the actual key value.

we have to use the "expression attribute name" feature of DDB to get around the fact that you can't use attribute names beginning with numbers in the actual expression.

the way we're doing it here, this is what an item might look like:

{
userID: "123",
.
.
devices {
  999: {
    userPublicKey: "a;sldjf"
    }
  }
}

if we just used USERS_TABLE_DEVICES_MAP_ATTRIBUTE_NAME in the update expression, the item would look like:

{
userID: "123",
.
.
999: {
  userPublicKey: "a;sldjf"
  }
}

this looks a little strange, since the device ID is in the outermost scope of the item without any description of what it actually is

DDB complains that the expression attribute names/values can't be empty. i.e., it can be None but it can't be Some(nothing)