Skip to content
Tutorials Center
My learning

Topics

Lesson 30 of 38 ยท Error Handling

Retry Policies, Timeouts and Throttling

Make flows resilient to temporary errors: the default retry policy, fixed and exponential intervals, which status codes are retried, action timeouts, HTTP limits and throttling.

Not every failure is your flow's fault. A service may be busy for ten seconds, a network packet may get lost, or an API may tell you to slow down. These transient errors vanish if you simply try again a little later. Power Automate can do that automatically with a retry policy, and it can stop an action that takes too long with a timeout. Getting both right is the difference between a flow that survives a bad minute and one that fails at 3 a.m.

This lesson explains the policy types, which errors are retried, how to configure timeouts, and how throttling works. The settings live in the designer, so the examples are JavaScript stand-ins that model the timing.

Where the settings are

For an action, open the three dots menu, choose Settings. For many actions (HTTP, most connector actions) you will find:

  • Retry Policy: Default, None, Fixed interval, or Exponential interval.
  • Timeout: how long the action may run, as an ISO 8601 duration.
  • Asynchronous Pattern, Pagination, Secure Inputs/Outputs, Concurrency control on the actions that support them.

Which failures are retried

A retry policy applies to responses that suggest the problem is temporary:

  • 408 (request timeout), 429 (too many requests) and 5xx server errors (500, 502, 503, 504);
  • connection failures and timeouts reaching the service.

Errors like 400 (bad request), 401 (unauthorized), 403 (forbidden) and 404 (not found) are not retried, because trying again will give the same answer. Fix the request, the permission or the connection instead.

The policy types

Policy Behavior Use when
Default Exponential interval, up to 4 retries Most cases; the safe starting point
None No retry; the first failure is final The action is not safe to repeat (it would send money twice)
Fixed interval Wait the same time between attempts A service recovers quickly and predictably
Exponential interval Wait longer after each attempt, with a minimum and maximum Busy or rate-limited services; avoids hammering

For the custom policies you set a Count (the number of retries, up to 90) and an Interval as ISO 8601 (for example PT10S for ten seconds, PT2M for two minutes; the minimum is five seconds). The exponential policy also takes a Minimum interval and a Maximum interval, and adds randomness so many flows do not retry at exactly the same moment.

If the service sends a Retry-After header (common with 429), Power Automate respects it where the connector supports it.

flowchart LR
    A["Call API"] -->|"200 OK"| OK["Continue"]
    A -->|"503"| W1["wait 10s"] --> B["Retry 1"]
    B -->|"503"| W2["wait 20s"] --> C["Retry 2"]
    C -->|"503"| W3["wait 40s"] --> D["Retry 3: give up<br/>after the last retry"]
    D --> F["Action failed:<br/>run after handles it"]

Example of an exponential schedule: the wait roughly doubles each time. Numbers here are illustrative.

JS Stand-in: retry schedules

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Timeouts

Two different clocks matter:

  1. The action timeout. Set it in Settings (PT30S, PT10M) to stop an action that hangs. The action then fails with a timed out status, which you can catch with run after.
  2. The flow run limit. A cloud flow run can last at most 30 days; anything still waiting after that is stopped. Long waits (approvals, delays) must fit within it.

A synchronous HTTP call also has a response limit of around two minutes (the exact figure can vary by connector). For longer work, use the asynchronous pattern: the API replies 202 Accepted with a Location header, and Power Automate polls that address until the job is done. Many connectors do this for you. For your own API, design a "start job, then check status" pair, and use a Do until loop with a delay if you must poll yourself.

Set a timeout on anything that calls the outside world. A missing timeout turns a stuck service into a stuck flow.

Throttling and 429

Services limit how many requests a client may send per minute. When you exceed that, you receive 429 Too Many Requests. Power Automate also limits actions per connection and per user (the limits depend on the license). Symptoms: sporadic 429 errors, runs that slow down, or a flow turned into "slow mode".

Ways to stay within limits:

  • Reduce calls: filter on the server (an OData $filter in "Get items") rather than fetching everything and filtering in the flow; use Select and other data operations instead of per-item calls.
  • Lower concurrency: in an Apply to each, set concurrency to a small number or leave it sequential (loops).
  • Spread the work: use a recurrence trigger with a smaller batch, or add a short Delay between calls.
  • Use batch endpoints when the API offers them.
  • Retry with exponential backoff so waiting grows when the service is busy.

Not safe to repeat? Use None or idempotency

A retry sends the same request again. If the first attempt actually reached the server but the reply was lost, a retry can create a duplicate: two invoices, two emails. Protect against it:

  • Choose None for actions that cannot be repeated safely, and handle failure yourself.
  • Send an idempotency key if the API supports one (a unique id per business operation, such as a GUID created once with guid() before the call and reused for every retry).
  • Check first ("does record 204 exist?") before creating.
JS Stand-in: why idempotency keys prevent duplicates

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. Which response does the default retry policy retry?

    408, 429 and 5xx errors are treated as transient; client errors like 400, 401, 403 and 404 are not retried.
  2. 2. Why is an exponential interval policy better for a busy API than a short fixed interval?

    Growing waits give a struggling service room to recover.
  3. 3. An action sends a payment and must never run twice. What is the safest retry setting?

    A retry after a lost reply could duplicate the payment.
  4. 4. What does HTTP status 429 mean?

    You exceeded a rate limit. Slow down, reduce calls, or back off.

Summary

  • The default retry policy retries 408, 429 and 5xx errors with exponential waits; None, Fixed and Exponential policies are configurable per action.
  • Client errors (400, 401, 403, 404) are not retried; fix the cause.
  • Set action timeouts and handle the timed-out state; runs have a 30 day maximum.
  • Reduce calls, lower concurrency and back off to avoid throttling.
  • Retries can duplicate work, so use None, idempotency keys or existence checks.

Next: Monitoring and Troubleshooting Flows