Best practices for working with Microsoft Graph
This article describes best practices that you can apply to help your applications get the most out of Microsoft Graph - whether that involves learning about Microsoft Graph, improving app performance, or making your application more reliable for end users.
Use Graph Explorer to get to know the API
The easiest way to start exploring the data available through Microsoft Graph is to use Graph Explorer. Graph Explorer lets you craft REST requests (with full CRUD support), adapt the HTTP request headers, and see the data responses. To help you get started, Graph Explorer also provides a set of sample queries.
Experiment with new APIs before you integrate them into your application.
To access the data in Microsoft Graph, your application will need to acquire an OAuth 2.0 access token, and present it to Microsoft Graph in either of the following:
- The HTTP Authorization request header, as a Bearer token
- The graph client constructor, when using a Microsoft Graph client library
Use the Microsoft Authentication Library API, MSAL to acquire the access token to Microsoft Graph.
Consent and authorization
Apply the following best practices for consent and authorization in your app:
Use least privilege. Only request permissions that are absolutely necessary, and only when you need them. For the APIs your application calls, check the permissions section in the method topics (for example, see creating a user, and choose the least privileged permissions. For a full list of permissions, see permissions reference.
Use the correct permission type based on scenarios. If you're building an interactive application where a signed in user is present, your application should use delegated permissions, where the application is delegated permission to act as the signed-in user when making calls to Microsoft Graph. If, however, your application runs without a signed-in user, such as a background service or daemon, your application should use application permissions.
Note: Using application permissions for interactive scenarios can put your application at compliance and security risk. It can inadvertantly elevate a user's privileges to access data, circumnavigating policies configured by an administrator.
Be thoughtful when configuring your app. This will directly affect end user and admin experiences, along with application adoption and security. For example:
- Consider who will be consenting to your application - either end users or administrators - and configure your application to request permissions appropriately.
- Ensure that you understand the difference between static, dynamic and incremental consent.
Consider multi-tenant applications. Expect customers to have various application and consent controls in different states. For example:
- Tenant administrators can disable the ability for end users to consent to applications. In this case, an administrator would need to consent on behalf of their users.
- Tenant administrators can set custom authorization policies such as blocking users from reading other user's profiles, or limiting self-service group creation to a limited set of users. In this case, your application should expect to handle 403 error response when acting on behalf of a user.
Handle responses effectively
Depending on the requests you make to Microsoft Graph, your applications should be prepared to handle different types of responses. The following are some of the most important practices to follow to ensure that your application behaves reliably and predictably for your end users.
When querying a resource collection, you should expect that Microsoft Graph will the return result set in multiple pages, due to server-side page size limits. When a result set spans multiple pages, Microsoft Graph returns an
@odata.nextLink property in the response that contains a URL to the next page of results.
For example, listing the signed-in users messages:
Would return a response containing an
@odata.nextLink property, if the result set exceeds the server-side page size limit.
Note: Your application should always handle the possibility that the responses are paged in nature, and use the
@odata.nextLinkproperty to obtain the next paged set of results, until all pages of the result set have been read. The final page will not contain an
@odata.nextLinkproperty. You should include the entire URL in the
@odata:nextLinkproperty in your request for the next page of results, treating the entire URL as an opaque string.
For more details, see paging.
Handling expected errors
While your application should handle all error responses (in the 400 and 500 ranges), pay special attention to certain expected errors and responses, listed in the following table.
|Topic||HTTP error code||Best practice|
|User does not have access||403||If your application is up and running, it could encounter this error even if it has been granted the necessary permissions through a consent experience. In this case, it's most likely that the signed-in user does not have privileges to access the resource requested. Your application should provide a generic "Access denied" error back to the signed-in user.|
|Not found||404||In certain cases, a requested resource might not be found. For example a resource might not exist, because it has not yet been provisioned (like a user's photo) or because it has been deleted. Some deleted resources might be fully restored within 30 days of deletion - such as user, group and application resources, so your application should also take this into account.|
|Throttling||429||APIs might throttle at any time for a variety of reasons, so your application must always be prepared to handle 429 responses. This error response includes the Retry-After field in the HTTP response header. Backing off requests using the Retry-After delay is the fastest way to recover from throttling. For more information, see throttling.|
|Service unavailable||503||This is likely because the services is busy. You should employ a back-off strategy similar to 429. Additionally, you should always make new retry requests over a new HTTP connection.|
Client applications can be broken by the addition of members to an existing enum. For some newer enums in Microsoft Graph, a mechanism is available to allow for adding new members without incurring a breaking change. On these newer enums, you'll see a common sentinel member called
unknownFutureValue that demarcates known and unknown enum members. Known members will have a number less than the sentinel member, while unknown members will be greater in value.
By default, unknown members are not returned by Microsoft Graph. If, however, your application is written to handle the appearance of unknown members, it can opt-in to receive unknown enum members, using an HTTP Prefer request header.
Note: If your application is prepared to handle unknown enum members, it should opt-in by using an HTTP prefer request header:
Storing data locally
In general, for performance and even security or privacy reasons, you should only get the data your application really needs, and nothing more.
Choose only the properties your application really needs and no more, because this saves unnecessary network traffic and data processing in your application (and in the service).
Note: Use the
$selectquery parameter to limit the properties returned by a query to those needed by your application.
For example, when retrieving the messages of the signed-in user, you can specify that only the from and subject properties be returned:
Getting minimal responses
For some operations, such as PUT and PATCH (and in some cases POST), if your application doesn't need to make use of a response payload, you can ask the API to return minimal data. Note that some services already return a 204 No Content response for PUT and PATCH operations.
Note: Request minimal representation responses using an HTTP request header where appropriate: Prefer: return=minimal. Note that for creation operations, this might not be appropriate because your application may expect to get the service generated
idfor the newly created object in the response.
Track changes: delta query and webhook notifications
If your application needs to know about changes to data, you can get a webhook notification whenever data of interest has changed. This is more efficient than simply polling on a regular basis.
Use webhook notifications to get push notifications when data changes.
If your application is required to cache or store Microsoft Graph data locally, and keep that data up to date, or track changes to data for any other reasons, you should use delta query. This will avoid excessive computation by your application to retrieve data your application already has, minimize network traffic, and reduce the likelihood of reaching a throttling threshold.
Use delta query to efficiently keep data up to date.
Using webhooks and delta query together
Webhooks and delta query are often used better together, because if you use delta query alone, you need to figure out the right polling interval - too short and this might lead to empty responses which wastes resources, too long and you might end up with stale data. If you use webhook notifications as the trigger to make delta query calls, you get the best of both worlds.
Use webhook notifications as the trigger to make delta query calls. You should also ensure that your application has a backstop polling threshold, in case no notifications are triggered.
JSON batching allows you to optimize your application by combining multiple requests into a single JSON object. Combining individual requests into a single batch request can save the application significant network latency and can conserve connection resources.
Use batching where significant network latency can have a big impact on the performance.
Reliability and support
To ensure reliability and facilitate support for your application:
- Honor DNS TTL and set connection TTL to match it. This ensures availability in case of failovers.
- Open connections to all advertised DNS answers.
- Generate a unique GUID and send it on each Microsoft Graph REST request. This will help Microsoft investigate any errors more easily if you need to report an issue with Microsoft Graph.
- On every request to Microsoft Graph, generate a unique GUID, send it in the
client-request-idHTTP request header, and also log it in your application's logs.
- Always log the
x-ms-ags-diagnosticfrom the HTTP response headers. These, together with the
client-request-id, are required when reporting issues in Stack Overflow or to Microsoft Support.
- On every request to Microsoft Graph, generate a unique GUID, send it in the
We'd love to hear your thoughts. Choose the type you'd like to provide:
Our feedback system is built on GitHub Issues. Read more on our blog.