Need (workaround) solution to improve BQ table reads in Apache beam sdk 250 java [Urgent]

[dheerajpanyam]

It takes 1min 5 sec to read 19 rows of  BQ full table scan that has around 10 columns (size is 15kb)  with Apache beam sdk 250.  On Apache beam 244 it takes just 4 seconds. Same behaviour with direct runner and also Dataflow runner. Is there a workaround solution? I cannot avoid full table scan 

I am using the below code

PCollection<TableRow> tableRows =

pipeline.apply(

"Read from BigQuery",

BigQueryIO.readTableRows()

.from(

new TableReference()

.setProjectId(params.getTableProjectName())

.setDatasetId(params.getTableDatasetName())

.setTableId(params.getTableName()))

.withMethod(Method.DIRECT_READ));