A cloud flow's HTTP action calls a partner API that occasionally returns HTTP 503 during short outages. The developer wants the action itself to retry a limited number of times, waiting longer between each attempt so the partner can recover, before the action fails. How should the developer configure the action?

Actions in Power Automate support a retry policy that applies to transient failures such as 408, 429, and 5xx responses and connectivity errors. The policy types include Default, None, Fixed interval, and Exponential interval. Exponential interval retries the action up to the configured count, with the wait between attempts growing exponentially from the specified interval within optional minimum and maximum bounds, which gives a struggling service progressively more time to recover. Fixed interval waits the same amount of time between every attempt, so it does not provide the growing delay the requirement asks for, and a very high retry count would also extend a failing run far longer than necessary. Setting the policy to None disables built-in retries; a Do until loop that resends immediately creates a tight loop of requests against an already failing service, wastes actions, and is exactly the behavior backoff exists to prevent. The Asynchronous pattern setting tells the action to follow a 202 Accepted response and poll the Location header; it does not retry failed requests or wait for an outage to end. Exam caveat: when the service returns 429 with a Retry-After header, the platform honors that value, so throttling and outages are handled by the same retry machinery. Operational check: point the action at a test endpoint that returns 503 and inspect the run history to confirm the attempts and the increasing delays.