We determined that reading the attributes from separate collections was too expensive, so we are going back to assuming all attributes are stored in the nodes and edges collections.
To accommodate the potential for a very large number of irrelevant attributes, I added the ability to only read in attributes that we care about using read_graph. Additionally, we can update a given attribute without accidentally removing other attributes that we didn't read but still want in the database, as would happen if we replaced the db node/edge with our local copy.
We determined that reading the attributes from separate collections was too expensive, so we are going back to assuming all attributes are stored in the nodes and edges collections.
To accommodate the potential for a very large number of irrelevant attributes, I added the ability to only read in attributes that we care about using read_graph. Additionally, we can update a given attribute without accidentally removing other attributes that we didn't read but still want in the database, as would happen if we replaced the db node/edge with our local copy.