I would bet the update will fix it. Huge manifests go into memory, in bulk, and in lists. The lists stay around and if Python GC is not working properly, the bulk reads do as well. During processing, there were as many as 3 copies of the manifest in memory, one bulk, one the result of re iteration, and the other in the collection list. That kind of misuse can really screw up GC. As answer to your question, changing to 'ulimit -n 10240' could help in a very cases, but I would try both individually to see which one does the job. If they are still on the 0.6-series, the 'ulimit' fix is the only one that would work. On Sun, Feb 11, 2018 at 12:34 PM, Michael Terry