When message batches are sent in the streaming format, in principle it should be very simple to build only specific columns, but the format for some reason is very hostile to this. The issue is that one does not a priori know which nodes and buffers one must use to build the column without iterating through them, and this cannot even be determined uniquely from the schema. So, for now, I'm not supporting this, but it's one of many reasons that building is painfully slow.
When message batches are sent in the streaming format, in principle it should be very simple to build only specific columns, but the format for some reason is very hostile to this. The issue is that one does not a priori know which
nodesandbuffersone must use to build the column without iterating through them, and this cannot even be determined uniquely from the schema. So, for now, I'm not supporting this, but it's one of many reasons that building is painfully slow.