Definitely an EBS issue. Our (EBS backed, Heroku hosted) DB was returning queries more and more slowly leading up to the outage.