Microsoft Graph API returning 503/504 errors

Question

I'm fetching messages from Office365 using Microsoft's Graph API Java client. The queries span a whole year. They're limited to 50 results per page and progress using the returned next page URL. It's done periodically every 5 minutes, but no longer than 2 minutes per job (which iterates over the next url).

Once in a while I get 503 Service Unavailable / 504 Gateway Timeout. Once this happens, the request can't progress and will keep hitting those errors.

According to Microsoft's documentation this should be treated as too many requests, and back off with a delay. There's no Retry-After header when this happens. I've noticed that making the time frame smaller and restarting the query can help sometimes. Also I've seen this raised on Stackoverflow but with no solutions.

I'm wondering is querying for a year too much, even though there's paging? Any ideas for solutions other than backing off, and for how long?

Thanks

Example query which fails:

https://graph.microsoft.com/v1.0/me/messages?$filter=IsDraft+eq+false+and+ReceivedDateTime+ge+2019-03-28T20%3a08%3a51.929Z+and+ReceivedDateTime+lt+2020-02-20T19%3a48%3a37Z&$orderby=ReceivedDateTime+desc&$expand=SingleValueExtendedProperties(%24filter%3did+eq+%27String+0x7D%27)&$select=conversationId%2cchangeKey%2csentDateTime%2creceivedDateTime%2cisRead%2chasAttachments%2cinternetMessageHeaders%2csender%2cfrom%2ctoRecipients%2cccRecipients%2cbccRecipients%2csubject%2cinternetMessageId%2cparentFolderId&$top=50&$skip=51

Edit: Waiting for an hour before sending more requests doesn't seem to help. It looks like the problem is with the large request.

Edit #2: Something which helped decrease the number of errors was to remove the filter=IsDraft+eq+false part from the query string, and filter the drafts on the client side. I still get occasional 503 errors, but much less.

Answer 1

There is specific guidance on Mailbox limits here https://learn.microsoft.com/en-us/graph/throttling#outlook-service-limits They are based on the individual user mailbox. So if you are calling more than 10,000 API requests in a 10 minute period or calling it in more than 4 concurrent processes you may be hitting this.

You are not getting a 429 response though, which would assume that you are not being throttled by the service.

So I understand your problem more, if you are running this every 5 mins, why are you having to go back a whole year? That is a much more complex query by adding that data range. You could use the Delta queries to just get the changes since the last 5 mins for the Drafts folder https://learn.microsoft.com/en-us/graph/delta-query-overview

Answer 2

Have you tried to change the order of filter statements? In your case, I would filter by the date first and then filter out all mails, that are drafts:

https://graph.microsoft.com/v1.0/me/messages?$filter=ReceivedDateTime+ge+2019-03-28T20%3a08%3a51.929Z+and+ReceivedDateTime+lt+2020-02-20T19%3a48%3a37Z+and+IsDraft+eq+false&$orderby=ReceivedDateTime+desc&$expand=SingleValueExtendedProperties(%24filter%3did+eq+%27String+0x7D%27)&$select=conversationId%2cchangeKey%2csentDateTime%2creceivedDateTime%2cisRead%2chasAttachments%2cinternetMessageHeaders%2csender%2cfrom%2ctoRecipients%2cccRecipients%2cbccRecipients%2csubject%2cinternetMessageId%2cparentFolderId&$top=50&$skip=51

I have had some throttling issues with MS Graph, which I was able to fix by changing the order of the filter statements. In a ten year old mailbox, there are probably more mails, that are not drafts, than there are mails from the last two years. I always try to apply the narrower filter first.