Was investigating the same issue because we had a surge in errors for our internal discourse instance, so I fixed it