Do Heroku

Dynos become services, add-ons become templates with volumes, config vars become variables, the Procfile becomes a start command. A step-by-step move.

Concept mapping#

Heroku	Skiffly	Notes
App	Project with one or more services	one service per process type
Dyno type (`web`, `worker`)	Service	`web` gets a domain; workers run without one
Pipeline stages (staging, production)	Environments	one project, `staging` and `production` environments
Review apps	PR preview environments	enable PR previews on the project
Buildpacks	Railpack	detects the same languages; or add a `Dockerfile`
`Procfile`	Start command per service	`web: gunicorn app:app` → start command `gunicorn app:app`
Config vars	Variables	`${{Postgres.DATABASE_URL}}` references instead of copied URLs
Heroku Postgres / Redis add-ons	Postgres / Redis templates with volumes	you own the instance and its data
Heroku Scheduler	Cron schedule on a service	`0 * * * *`
`heroku run bash`	`skiffly ssh`	shell in the running container
`heroku logs --tail`	`skiffly logs -f`
`heroku ps:scale web=2`	replicas	Settings → Deploy → Replicas
Dyno sizes	CPU / memory limits per service	any value within the plan
`*.herokuapp.com`	`*.skiffly.cloud`	custom domains with automatic TLS
`heroku.yml`	`.skiffly/skiffly.ts`	config as code
Monthly invoice	Prepaid balance	PayPal or crypto

Step by step#

Sign in at app.skiffly.dev with GitHub; the code must be on GitHub (Heroku's git remote is not a source).
Create the project and a service per process type: skiffly init in the repository, skiffly up for web. For a worker, create a second service from the same repository and set its start command to the worker line of the Procfile; leave it without a domain.

Databases. skiffly deploy --template postgres (and redis). Copy the data:

heroku pg:backups:capture -a my-app && heroku pg:backups:download -a my-app     # latest.dump
skiffly connect postgres --print
pg_restore -d "<url>" --no-owner --clean --if-exists latest.dump
skiffly proxy delete <port>

Config vars. heroku config -s -a my-app > vars.env, drop the add-on URLs, then:

skiffly variables set --skip-deploys $(cat vars.env | xargs)
skiffly variables set 'DATABASE_URL=${{Postgres.DATABASE_URL}}' 'REDIS_URL=${{Redis.REDIS_URL}}'
skiffly redeploy

Port. Heroku sets $PORT; so does Skiffly. Nothing to change if the app already reads it.
Release phase. There is no release phase hook. Run migrations from the start command (sh -c "npm run migrate && npm start") or as a one-off: skiffly ssh -- npm run migrate.
Scheduler. For each scheduled job create a service from the same repository with a cron schedule and the job's command as the start command.
Domain. Add the custom domain, create the CNAME and TXT records, verify, then switch DNS from Heroku once the certificate is issued.
Turn Heroku off after a few days of traffic on Skiffly.

Differences worth knowing#