It occurs to me now that a lot of this could have been avoided by simply restarting nginx.