This merge request fixes the problem you detected.
The last version I did to clean /dev/shm folder, correctly removed the bigmemory files if the code fails in the retrieving data step. However, the code doesn't fail so ecFlow detects the job as finished and the user gets a result (filled with zeros as @cdelgado reported).
This new version should stop the process just after cleaning the memory, so, ecFlow will show that the job has failed.
@aho, please, let me know if you have any inconveniences in installing this version in WS (R 3.6.1) and Nord3. Given that it is a problem with Compute when submitting to a cluster, this is the only way I know to test it.