This sounds like something you could easily spin up on demand in AWS instead of aiming for reliability. Either an autoscaling group or a spot fleet that activates every hour, runs your job on startup, and kills itself. Micro is 1gb memory and costs $0.0104/h. Running that for 15min every hour is $1.8 a month. (+whatever storage space you need)

You could probably use a heroku task as well.