Poll a Hetzner Cloud Action to completion after triggering an asynchronous API operation

Documented steps

1. Almost every mutating call on servers, volumes, images, firewalls, networks, load balancers and IPs returns an Action object with status 'running' instead of finishing inline. Capture action.id from the response.
2. The Action object fields are id, command, status (running | success | error), started, finished (null while running), progress (percent), resources (array of {id,type}) and error (null or {code,message}).
3. Poll with the resource-scoped endpoint GET /v1/{resource}/actions/{action_id}, for example GET /v1/servers/actions/12345 . The global GET /v1/actions/{id} also works.
4. Loop until status is success or error. On error, read action.error.code and action.error.message - the HTTP status of the poll request itself will still be 200.
5. Confirm which resource an action belongs to by inspecting action.resources[] rather than assuming it from the URL you polled.
6. To fetch several actions in one request use GET /v1/actions?id=1&id=2&id=3 . The id query parameter is required on this endpoint.
7. When you created a server, also poll the entries in next_actions (for example start_server) - the create_server action reaching success does not mean the machine has booted.
8. If you use an official client library, prefer its built-in helper (Go WaitFor, Python wait_until_finished) over a hand-rolled loop.
9. Reference: https://docs.hetzner.cloud/reference/cloud