As you noted, Accumulators do not guarantee accurate results except in specific situations. I recommend never using them.
This article goes into some detail on the problems with accumulators: http://imranrashid.com/posts/Spark-Accumulators/ On Wed, Mar 1, 2017 at 7:26 AM, Charles O. Bajomo < charles.baj...@pretechconsulting.co.uk> wrote: > Hello everyone, > > I wanted to know if there is any benefit to using an acculumator over just > executing a count() on the whole RDD. There seems to be a lot of issues > with accumulator during a stage failure and also seems to be an issue > rebuilding them if the application restarts from a checkpoint. Anyone have > any suggestions no this? > > Thanks >