We ran into an unnecessary obstacle when analyzing some MediaWiki API request logs, trying to use the array_frequency function documented in the latest revision. Discovering this function was not available yet, I tried the historgram function and got an error that the maximum memory usage of 40GB had been exceeded. This is surprising because the columns I was trying to aggregate were on a table of 3.5M rows but with only a small, finite (either 2 or c. 900) number of distinct elements, so this built-in histogram must be doing something unscalable behind the scenes.
This isn't a blocker for our work, but I'm filing a task to put one finger on the scale the next time an upgrade is discussed.
Thank you for maintaining Superset and Presto, my team has been loving the experience so far!