Help for owners
My Microsoft 365 integration fails with 429 Too Many Requests
Short answer
Microsoft Graph throttles an app that sends too many requests, returns HTTP 429 and puts a wait time in the Retry-After header. Microsoft advises waiting exactly that long, avoiding immediate retries, and using change tracking and notifications in place of constant polling.
Does this sound like your situation?
- My app shows 429 Too Many Requests or TooManyRequests from Microsoft.
- Mail, calendar, Teams or SharePoint syncs stop partway.
- Writes fail while reads still work.
- It fails during a bulk import or a nightly job.
- Errors continue for a long time after a restart.
Why does this happen?
A burst or a heavy job
Microsoft states that throttling limits vary by scenario, and that a large volume of writes is more likely to be throttled than reads. It also says that when a threshold is exceeded, further requests from that app are limited for some time.
Confirmed by learn.microsoft.com
Retrying immediately
Microsoft advises avoiding immediate retries because all requests count against your usage limits, and says waiting the number of seconds in the Retry-After header is the fastest way to recover. An app that retries at once keeps itself throttled.
Confirmed by learn.microsoft.com
Polling instead of listening
Microsoft states that continuously polling a resource and regularly scanning collections for changes are more likely to lead to throttling, and recommends change tracking and change notifications instead.
Confirmed by learn.microsoft.com
Batches hide the failures
Microsoft states that requests in a JSON batch are judged one by one, the batch itself can return 200, and individual requests inside it can still return 429. An app that checks only the outer status silently loses those items.
Confirmed by learn.microsoft.com
Many apps in one tenant
Microsoft lists as common causes a large number of requests across all applications in a tenant, and a large number from one application across all tenants.
Confirmed by learn.microsoft.com
What can I do about it today?
- Note the time and the jobWhat was running when the 429s started: an import, a sync, a report, a new feature.
- Do not restart over and overEach restart can start a new burst. Give the app time instead.
- Pause extra jobs that use the same appA second script or tool on the same registration adds to the same count.
- Copy a failed response if you canIf your developer shows you the error, keep the Retry-After value and the error text.
- Check for missing items afterwardsThrottled requests inside a batch can be dropped. Compare counts once things calm down.
When should I bring someone in?
- The throttling returns daily.
- Data is missing because requests were dropped.
- You need to move a very large amount of Microsoft 365 data.
How do I handle it?
- Find which calls cause itI look at which Graph requests the app makes and how often, in logs or a test run.
- Honour Retry-AfterThe app waits the time Microsoft gives and retries only then, with exponential backoff when the header is absent, as Microsoft advises.
- Replace polling with change trackingDelta queries and change notifications are used where Microsoft offers them, so the app asks only what changed.
- Check each item in a batchEvery response in a batch is read, and the throttled ones are retried in a new batch.
- Use the bulk route for bulk dataMicrosoft points to Graph Data Connect for extracting large volumes. I check whether it fits your case before building a workaround.
The price depends on what I find, so I quote in writing after a short discovery call. Tell me what is happening and I reply within 24 hours.
What do owners ask about this?
What is Retry-After?
A header in the 429 response that says how long the app should wait. Microsoft advises waiting that number of seconds before retrying.
Can the limit be raised?
Microsoft's throttling guidance focuses on sending fewer, smarter requests, and points to Graph Data Connect for bulk extraction. Redesigning the calls is the dependable fix.
Does every Graph SDK handle this?
Microsoft states its SDKs include handlers that rely on Retry-After or fall back to exponential backoff, but requests that were part of a batch are not retried automatically.
Sources
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. Found a mistake? Tell me.