How would you debug performance bottlenecks in a data pipeline when processing on a large Cloudera-style cluster?
