Find the Fly.io resources quietly billing you every month

Every Monday at 9am, audit my Fly.io organization for resources that keep billing me when nothing is actually running, and post the findings to Slack as one message. This is an audit of idle and orphaned resources, not a scheduler that stops running Machines. Do not stop, scale, create, or delete anything at any point.

Start with Fly.io List Apps to get every app in the organization. Then, for each app, call List Machines, List Volumes, and List IP assignments for app. This needs an account-level Fly token that can see every app, because a token scoped to a single app cannot list apps or read organization-wide data. Fly rate limits are roughly one request per second per action, so page politely across the apps instead of firing every request at once. Where a list is cursor-paginated, such as the organization-wide List All Volumes, keep asking for the next page until no cursor comes back: a short page is not proof that you reached the end.

Look for three specific leaks. First, volumes attached to no Machine at all. Fly bills persistent volumes hourly whether or not a Machine is running, so a detached volume is pure waste. Second, apps where every Machine is stopped or destroyed but a volume or a dedicated IPv4 address is still assigned. These look dormant in the dashboard while still appearing on the invoice. Third, leftover pull request preview apps, which Fly's standard review-apps setup names in the pattern pr-123-myrepo.

For each app matching that preview naming pattern, pull the pull request number and the repository name out of the app name, then cross-check with GitHub List Pull Requests across all states rather than open ones only. Use Get a Pull Request when you need the precise merge or close date for a specific number. If the pull request is already merged or closed, say so in the report along with how long ago it happened, so the finding reads as "PR 412 was merged six weeks ago and this app was never torn down" rather than as a bare app name. If a preview app has no matching pull request, flag it as unmatched instead of guessing.

Estimate a monthly cost for every item using Fly's published rates: persistent volume storage is billed per GB per month against the volume's provisioned size, and a dedicated IPv4 address is $2 per month. Label these clearly as estimates and note that published rates can change. Rank every finding by how much money is at stake, largest first, so the top of the message is the most expensive thing to fix.

Post one Slack message with Send a Message, not one message per finding. Open with the total estimated monthly spend at stake across everything found, then group the findings by app, listing each item underneath its app with its estimated monthly cost and a one-line reason it was flagged. If the audit turns up nothing, still post a short all-clear so the silence is never ambiguous.

Only ever recommend, never delete. A detached volume can still hold real data, and an IP address that looks unused may still be referenced by DNS. Close the message with the exact fly commands to clean up each item so a human can copy them after checking: fly volumes destroy <volume-id> for a detached volume, fly ips release <address> --app <app-name> for a dedicated IPv4, and fly apps destroy <app-name> for a dead preview app. Put a one-line warning above the commands that destroying a volume permanently deletes its data.

Finish the message with a short standing reminder that Fly's managed add-ons, including Managed Postgres, Upstash Redis, and Tigris object storage, are billed separately and are not removed when an app is deleted, so they are worth a manual look during cleanup.