Same here. S3 as a point of failure makes zero sense for dyno uptime, I'm very frustrated with Heroku. (the dyno was already running, no need to download a new slug in my opinion)