It's indeed very possibly caused by a "disk full" situation. The "DSS" engine of the join recipe pulls all data from all datasets to the local disk to perform the join. As much as possible, we strongly recommend that you perform joins either on Spark, Hadoop or a SQL database