The ds column was definitely in the data, no doubt about it. The same cluster was used for the experiments before, so it worked a week before and then stopped. It is working again, so I guess it was just an intermittent issue.