Leaving Fly.io: a migration checklist for a small team
Fly.io is built for apps that run close to users in many regions. Most small startups run one app, one worker and one database in one region. If that is you, a lot of what Fly asks you to look after (a Dockerfile, Machines, volumes, a Postgres cluster) is work you do not need.
This post is the checklist for moving a small Fly.io app to a server you own: what Fly does today, where each piece goes, and the steps to switch with one short maintenance window.
Disclosure: we build DollarDeploy, a control plane that deploys and operates apps on servers you own. The checklist works with any setup. Where DollarDeploy covers a step, we say so, and where it does not, we say that too.
Why teams leave
Why it matters: on Fly, the platform runs your containers, but a lot of the operations stay with you. For a team without a DevOps hire, that part lands on the CTO.
- You maintain the Dockerfile. Every app is a container image, so base images, build caches and image size are your job.
- Machines stop when idle if
auto_stop_machinesis on. That saves money, and the first request after a quiet period waits for a cold start. - Volumes belong to one Machine in one region. Data on a volume does not move with the app; you plan backups and moves yourself.
- Scheduled jobs need extra setup: a Supercronic process in your image, scheduled Machines (hourly, daily, weekly or monthly only), or a separate Cron Manager app. None of them tells you when a job starts failing.
The big picture: if you do not need many regions, the simpler setup is one server, close to your users, with the platform work done for you.
Fly.io piece by piece
Use this as the map. Left: what you use on Fly. Right: where it goes on your own server, and what DollarDeploy does for it.
fly deploy: deploys from your GitHub repo on every push to a branch you pick. In DollarDeploy you connect the repo once, and every push builds and deploys.- Dockerfile: usually not needed. DollarDeploy detects common stacks (Node.js, Next.js, Python, Go and more), fills in the build and start commands, and runs the app as a service without Docker. If your app only runs from its Dockerfile, add a short
docker-compose.ymlthat builds it: Docker Compose apps are supported too. [deploy] release_command(migrations before the new version starts): the Custom pre start shell command in the app settings.[processes]groups (web, worker): a second app from the same repo with its own start command, on the same server.- Secrets: app env vars. They are stored encrypted, and hidden from teammates without write access to that app.
- Fly Postgres or Managed Postgres, Upstash Redis: PostgreSQL and Redis installed on your server with one click, on localhost, next to the app.
- Volumes: a folder inside the app's directory on the server, included in the app's backups.
- Supercronic or scheduled Machines: the app's Cron tab, with any cron expression. Each run is a task with logs, retries and a notification when a job starts failing (paid plans).
- Certificates (
fly certs): HTTPS from Let's Encrypt for every app domain, renewed for you. fly logs: app logs and monitoring on every plan, kept for 3 days on Free and 30 days on paid plans.fly releasesand rolling back: the Builds table, where Deploy on any earlier build ships it again without rebuilding.- Org members: roles per team, project, app or server (View, Deploy, Write, Maintain, Admin), so a contractor can deploy one app without touching the server.
The server work is covered too: DollarDeploy prepares the server with a firewall and automatic security updates, and Inspect checks 13 areas of it (disk, memory, services, certificates, SSH logins and more) without changing anything.
The migration, step by step
Plan for one maintenance window of 15-60 minutes, depending on the size of your database. Everything before it can happen during the week.
- Lower your DNS TTL to 5 minutes, two days ahead. The final switch then takes minutes instead of hours.
- Rent a server in the region your Fly app runs in. One server with 2 GB of memory usually runs a small app, its worker and its database. DollarDeploy can create it for you with Hetzner, DigitalOcean, Vultr or AWS (selected regions), or connect any server you already have over SSH.
- Collect your secrets.
fly secrets listshows only names, never values. Take the values from your password manager, or print them from a running Machine withfly ssh console -C env. Keep the result out of git. - Create the app from your GitHub repo and paste the env vars. Then point
DATABASE_URLat the new database: DollarDeploy exposes the local PostgreSQL asPOSTGRES_URL, and Redis asREDIS_URL. - Deploy it on a temporary domain and run through your app with a copy of the data. Fix build differences now, not during the window.
- Add the cron jobs from your crontab or scheduled Machines, and send deploy, cron and backup notifications to the channel your team reads (Slack, Telegram, Discord, email or a webhook).
- Start the window: stop writes (scale the app to zero with
fly scale count 0, or show a maintenance page), then dump the database through a local tunnel:fly mpg proxyfor Managed Postgres,fly proxy 5432 -a <db-app>for a Postgres app you run yourself, andpg_dump -Fcagainst localhost. - Restore it on the new server with
pg_restore. If the app kept files on a volume, copy them over withfly ssh sftp. This is the one step that needs SSH, once. - Switch DNS to the new server, check the app from outside your network, and keep the Fly app and its volumes for a few days in case you need to go back.
- Take a backup and restore it once on the new server. A backup you have not restored is a guess.
Yes, but
You give up multi-region. If your users really are spread across continents and latency matters, Fly is built for that and one server is not. Most early products serve one market; check where your traffic comes from before you decide.
You now own a server. DollarDeploy keeps it updated and checked, but capacity is your call: there is no autoscaling across machines. For a small always-on app this rarely matters. For spiky traffic, size the server for the peak.
Cron runs are capped at 60 seconds. Long jobs should run in your worker instead, with the cron only starting them.
Our honest limit: DollarDeploy does not page anyone when your app is down, so add an external uptime monitor. We are also a small team without a public SLA or status page yet. If you need contractual uptime guarantees, factor that in.
By the numbers
Cost is not the reason to move, but you should know it. A small always-on app with its database on one server in the US or London: $15/mo all-in, DollarDeploy Hobby $5 plus a Vultr 1 vCPU / 2 GB server at $10.
The bottom line
Leaving Fly.io makes sense when you run in one region and want fewer moving parts. Map each piece first, rehearse on a temporary domain, and the switch itself is one short window.
Go deeper: Leaving Heroku: a migration checklist for a small team, Leaving DigitalOcean App Platform: a migration checklist for a small team, Backups for a small startup: what to keep and how to test a restore, Running production without a DevOps hire and Your first production incident: a runbook for a small team.
One region, one server, no ops to carry. DollarDeploy deploys your apps from GitHub to a server you own and runs the work that Fly left to you: builds without a Dockerfile, one-click rollback, HTTPS, monitoring and logs, PostgreSQL and Redis, cron jobs with failure alerts, backups and server updates. Nothing for you to maintain. Start free on DollarDeploy, no card needed.