Performance drop from databricks 12.2 to 14.3 LTS - solved with checkpoint(), looking for root cause

operryman — Mon, 15 Jul 2024 20:00:39 GMT

On databricks 12.2, a piece of code which has an action takes a minute to run.

On databricks 14.3, the same code unchanged, same inputs, now takes 10 minutes.

Attempting to debug using explain() shows the plan is huge (150k plus rows of output)

Replacing cache() with checkpoint() to truncate the plan also solves the performance problem.

What was different about 12.2 vs 14.3 that highlighted this problem by slowing down? Shouldn't 14.3 be faster in all cases?

topic Performance drop from databricks 12.2 to 14.3 LTS - solved with checkpoint(), looking for root cause in Data Engineering