An auditor asked us to reproduce a report exactly as it looked on one date three months ago. In the past, that meant rebuilding old data and rerunning pipelines for days, and still wondering if the result matched. With Delta tables in Databricks, I queried the table as it was on that date and had the same numbers in minutes. I also use the lineage view a lot. When a figure looks wrong, I can see which upstream tables feed it, instead of asking five people. Review collected by and hosted on G2.com.

Time travel only goes back as far as you keep history, so I had to change our retention settings before it covered the period the auditors cared about. Clusters left running also made our bill jump once, and now I set them to shut down on their own. Review collected by and hosted on G2.com.