Dataflow preprocessing reads many BigQuery logs; performance is poor and data is rapidly growing. Which change improves read performance?

Reading only necessary fields with a .fromQuery (query-based) read significantly improves performance when processing large BigQuery tables. This reduces the amount of data transferred from BigQuery to Dataflow, decreasing network I/O and Dataflow worker memory usage. Specifying the TableReference explicitly is standard practice but doesn't inherently improve read performance. Using TableSchema and TableFieldSchema is for defining the schema, not for optimizing read performance. Returning TableRow objects is the standard output format for BigQuery reads in Dataflow; it doesn't optimize the initial read operation itself.