This morning I wrote about my plans to move from hosting my mastodon instance on Heroku to hosting it on AWS. Well I got some feedback, and I did some thinking, and I changed my plan, and then I actually did the whole thing! I’m now hosting on Digital Ocean (referral link) and I kind of love it.
My primary reason for moving hosts was cost. I really like Heroku, but you pay a premium for the convenience of it. That’s why I planned to switch to AWS, but turns out you can end up paying a lot for convenience on AWS, too. I found there were comparable costs for Redis and Postgres services between Heroku and AWS, the those were the two things I was expecting to save the most on.
I know that you save money when you roll up your sleeves and start administer your own database, but I didn’t realize how much. Luckily, the Mastodon developers know administrating stuff is hard, so they have support for Docker.
Docker is like a virtual machine (don’t @ me) and the Mastodon developers include what are basically recipes for Docker containers to be built from, and to run from. Combined with this blog post, I was able to get a test instance of Mastodon up within a half hour.
Which, wow, that amazed me. I’ve historically had a lot of trouble deploying stuff, and this was amazing. I couldn’t believe how quickly I had something working. But enough gushing over Docker.
I originally had a complicated migration strategy. Originally. Then, instead of doing a bunch of complicated configuration to get zero-or-near-zero downtime, I just took the instance down. The whole process took twenty minutes, and would have been shorter but I ran into problems importing the Postgres database backup.
Things went well! I’ve got SSL set up like before, and also gained IPv6 support. Somehow, I guess, I lost the Redis database? I’m not sure, but things are working well enough that I’m not going to worry about it.
So how are things working? Well, I don’t have a response time graph anymore but interactions feel a little slower. CPU utilization is peaking at 30%, and RAM is near 30% as well.
So what about the costs? Well, I’m currently running on a $40/month virtual machine, plus some backup, S3, and Mailgun costs, so I’m pegging it at $50–60, or about a third of what I was paying on Heroku. The VM has 4GB of RAM, 2 virtual CPUs, 30GB of SSD space, and 4TB of network transfer a month. Not bad!
The constellation of Heroku Dynos and Add-Ons has been reduced to one virtual computer running contains each for Sidekiq, Rails-on-Puma, Redis, and Postgres. Everything is stored to the main disk but I may take a look at Digital Ocean’s dedicated storage volumes soon.
I’m going to leave it running as-is and look at tuning or reducing costs until I have a solid baseline of how things are working. I kind of miss my Papertrail logs and my response time graphs, so I’ll probably customize the code a bit.