When faced with an Oracle NUMBER(*) column we sample approx. 1GB of data in the source database to estimate the correct precision/scale to use in the backend.
On BigQuery we should record these as an unsafe sample and roll the precision and scale up to the max of NUMERIC/BIGNUMERIC. We used to do that but at some point in the recent past we enabled support for copying the exact precision/scale to BigQuery, we should not do that for unsafe/sampled values.
When faced with an Oracle NUMBER(*) column we sample approx. 1GB of data in the source database to estimate the correct precision/scale to use in the backend.
On BigQuery we should record these as an unsafe sample and roll the precision and scale up to the max of NUMERIC/BIGNUMERIC. We used to do that but at some point in the recent past we enabled support for copying the exact precision/scale to BigQuery, we should not do that for unsafe/sampled values.