Which practice preserves query folding and improves performance in data transformation workflows?

• A. Filter for the data you need at the source when possible
• B. Always apply all filters after all transformations
• C. Split by position before filtering
• D. Use Delta Lake only

Answer: (A)

Filtering for the data you need at the source when possible keeps query folding alive, because the filter can be pushed down to the data store as part of the read query. When the data source applies the filter, only the matching rows are retrieved, reducing data scanned, lower I/O, and lighter downstream processing. This leads to faster overall performance since expensive transformations operate on a smaller dataset.

If you wait to apply filters after loading data or after several transformations, you miss the opportunity to fold the filter into the source query, so you pull and process a larger dataset than necessary. The other options don’t target this optimization: filtering by position before filtering isn’t related to folding, and using Delta Lake alone doesn’t automatically ensure folding for all operations. The practical takeaway is to push predicates down to the source whenever the source supports it and keep filtering as early as possible in the pipeline.