Επεξεργασία

Κοινή χρήση μέσω


Service protection API limits

To ensure consistent availability and performance for everyone we apply some limits to how APIs are used. These limits are designed to detect when client applications are making extraordinary demands on server resources.

The limits should not affect normal users of interactive clients. Only client applications that perform extraordinary API requests should be affected. The limits provide a level of protection from random and unexpected surges in request volumes that threaten the availability and performance characteristics of the Microsoft Dataverse platform.

When a client application makes extraordinarily demanding requests, the Dataverse follows the common pattern for online services. We return an error indicating that too many requests have been made.

Impact on client applications

It is the responsibility of client applications to manage service protection API limit errors. Exactly how to manage this error depends on the nature of the application.

Interactive client applications

The service protection limits are high enough that it should be rare for an individual using an interactive client application to encounter them during normal usage. However, it is possible if the client application allows for bulk operations. Client application developers should be aware of how service protection API limits are enforced and design the UI to reduce the potential for users to send extremely demanding requests to the server. But they should still expect that service protection API limit errors can occur and be prepared to handle them.

Client application developers should not simply throw the error to display the message to the user. The error message is not intended for end users. See Retry operations for specific strategies.

Data integration applications

Applications designed to load data into Dataverse or perform bulk updates must also be able to manage service protection API limit errors. These applications prioritize throughput so they can complete their work in the minimum amount of time. These applications must have a strategy to retry operations. There are several strategies that they can apply to get the maximum throughput. More information: How to maximize throughput.

Portal applications

Portal applications typically send requests from anonymous users through a service principal account. Because the service protection API limits are based on a per user basis, portal applications can hit service protection API limits based on the amount of traffic the portal experiences. Like interactive client applications, it isn't expected that the service protection API limits errors should be displayed to the portal end user. It is expected that the UI should disable further requests and display a message that the server is busy. The message may include the time when the application can begin accepting new requests.

Impact on plug-ins and custom workflow activities

Plug-ins and custom workflow activities apply business logic triggered by incoming requests. Service protection limits are not applied to plug-ins and custom workflow activities. Plug-ins and custom workflow activities are uploaded and run within the isolated sandbox service. Dataverse operations invoked on the sandbox service do not use the public API endpoints.

If your application performs operations that trigger custom logic, the number of requests sent by plug-ins or custom workflow activities will not be counted towards service protection API limits. However, the additional computation time that these operations contribute will be added to the initial request that triggered them. This computation time is part of the service protection API limits. More information: How service protection API limits are enforced

Retry operations

When a service protection API limit error occurs, it will provide a value indicating the duration before any new requests from the user can be processed.

The Retry-After duration

The Retry-After duration will depend on the nature of the operations that have been sent in the preceding 5 minute period. The more demanding the requests are, the longer it will take for the server to recover.

Today, because of the way the limits are evaluated, you can expect to exceed the number of requests and execution time limits for a 5 minute period before the service protection API limits will take effect. However, exceeding the number of concurrent requests will immediately return an error. If the application continues to send such demanding requests, the duration will be extended to minimize the impact on shared resources. This will cause the individual retry-after duration period to be longer, which means your application will see longer periods of inactivity while it is waiting. This behavior may change in the future.

When possible, we recommend trying to achieve a consistent rate by starting with a lower number of requests and gradually increasing until you start hitting the service protection API limits. After that, let the server tell you how many requests it can handle within a 5 minute period. Keeping your maximum number of requests limited within this 5 minute period and gradually increasing will keep the retry-after duration low, optimizing your total throughput and minimizing server resource spikes.

Interactive application re-try

If the client is an interactive application, you should display a message that the server is busy while you re-try the request the user made. You may want to provide an option for the user to cancel the operation. Don't allow users to submit more requests until the previous request you sent has completed.

Non-interactive application re-try

If the client is not interactive, the common practice is to simply wait for the duration to pass before sending the request again. This is commonly done by pausing the execution of the current task using Task.Delay or equivalent methods.

How to re-try

The following describes how to retry .NET applications using the Dataverse SDK for .NET or Web API:

If you are using the SDK for .NET, we recommend that you use the Microsoft.Xrm.Tooling.Connector.CrmServiceClient or ServiceClient classes. Those classes implement the IOrganizationService methods and can manage any service protection API limit errors that are returned.

Since Xrm.Tooling.Connector version 9.0.2.16, it will automatically pause and re-send the request after the Retry-After duration period.

If your application is currently using the low-level Microsoft.Xrm.Sdk.Client.OrganizationServiceProxy or Microsoft.Xrm.Sdk.WebServiceClient.OrganizationWebProxyClient classes. You should be able to replace those with the CrmServiceClient or ServiceClient class. The OrganizationServiceProxy is deprecated.

More information:

How Service Protection API Limits are enforced

Two of the service protection API limits are evaluated within a 5 minute (300 second) sliding window. If either limits are exceeded within the preceding 300 seconds, a service protection API Limit error will be returned on subsequent requests to protect the service until the Retry-After duration has ended.

The service protection API limits are evaluated per user. Each authenticated user is limited independently. Only those users accounts which are making extraordinary demands will be limited. Other users will not be impacted.

Service protection API limits are enforced based on three facets:

  • The number of requests sent by a user.
  • The combined execution time required to process requests sent by a user.
  • The number of concurrent requests sent by a user.

If the only limit was on the number of requests sent by a user, it would be possible to bypass it. The other facets were added to counter these attempts. For example:

  • You could send fewer requests by bundling them in batch operations.
    • The combined execution time limit will counter this.
  • Rather than sending requests individually in succession, you could send a large number of concurrent requests before service protection API limits are enforced.
    • The concurrent request limit will counter this.

Each web server available to your environment will enforce these limits independently. Most environments will have more than one web server. Trial environments are allocated only a single web server. The actual number of web servers that are available to your environment depends on multiple factors that are part of the managed service we provide. One of the factors is how many user licenses you have purchased.

The following table describes the default service protection API limits enforced per web server:

Measure Description Limit per web server
Number of requests The cumulative number of requests made by the user. 6000 within the 5 minute sliding window
Execution time The combined execution time of all requests made by the user. 20 minutes (1200 seconds) within the 5 minute sliding window
Number of concurrent requests The number of concurrent requests made by the user 52 or higher

Important

These limits are subject to change and may vary between different environments. These numbers represent default values and are provided to give you some idea of what values you can expect.

Service Protection API Limit Errors returned

This section describes the three types of service protection API limit errors that can be returned as well as factors that cause these errors and possible mitigation strategies.

  • The Error code is the numerical error value returned by the SDK for .NET OrganizationServiceFault.ErrorDetails.
  • The Hex code is the hexadecimal error value returned by the Web API.

Number of requests

This limit counts the total number of requests during the preceding 300 second period.

Error code Hex code Message
-2147015902 0x80072322 Number of requests exceeded the limit of 6000 over time window of 300 seconds.

It is not expected that a typical user of an interactive application will be able to send 1,200 requests per minute to exceed this limit unless the application enables users to perform bulk operations.

For example, if a list view enables selection of 250 records at a time and allows a user to perform some operation on all these records, the user would need to perform this operation 24 times in a span of 300 seconds. The user would need to complete the operation on each list within 12.5 seconds.

If your application provides this capability, you should consider some of the following strategies:

  • Decreasing the total number of records that can be selected in a list. If the number of items displayed in a list is reduced to 50, the user would need to perform this operation 120 times within 300 seconds. The user would have to complete the operation on each list within 2.5 seconds.
  • Combine the selected operations into a batch. A batch can contain up to 1000 operations and will avoid the number of requests limit. However, you will need to be prepared for the execution time limit.

Execution time

This limit tracks the combined execution time of incoming requests during the preceding 300 second period.

Error code Hex code Message
-2147015903 0x80072321 Combined execution time of incoming requests exceeded limit of 1,200,000 milliseconds over time window of 300 seconds. Decrease number of concurrent requests or reduce the duration of requests and try again later.

Some operations require more resources than others. Batch operations, importing solutions, and highly complex queries can be very demanding. These operations may also be performed simultaneously in concurrent requests. Therefore, within the 5 minute window it is possible to request operations that will require more than 20 minutes of combined computation time.

This limit can be encountered when strategies using batch operations and concurrent requests are used to avoid the number of requests limit.

Concurrent requests

This limit tracks the number of concurrent requests.

Error code Hex code Message
-2147015898 0x80072326 Number of concurrent requests exceeded the limit of 52.

Client applications are not limited to sending individual requests sequentially. The client may apply parallel programming patterns or various methods to send multiple requests simultaneously. The server can detect when it is responding to multiple requests from the same user simultaneously. If this number of concurrent requests is exceeded, this error will be thrown. The limit may be higher than 52 in some cases.

Sending concurrent requests can be a key part of a strategy to maximize throughput, but it is important to keep it under control. When using Parallel Programming in .NET the default degree of parallelism depends on the number of CPU cores on the server running the code. It should not exceed the limit. The ParallelOptions.MaxDegreeOfParallelism Property can be set to define a maximum number of concurrent tasks.

More information: Send parallel requests

How to maximize throughput

When you have an application that must prioritize throughput to move the most data in the shortest period, there are some strategies you can apply.

Let the server tell you how much it can handle

Don't try to calculate how many requests to send at a time. Each environment can be different. Gradually increase the rate you send requests until you begin to hit limits and then depend on the service protection API Limit Retry-After value to tell you when to send more. This value will keep your total throughput at the highest possible level.

Use multiple threads

The higher limit on number of concurrent threads is something your application can use to have a significant improvement in performance. This is true if your individual operations are relatively quick. Depending on the nature of the data you are processing, you may need to adjust the number of threads to get optimum throughput. More information: Send parallel requests

Avoid large batches

Batching refers to sending multiple operations in a single request.

Most scenarios will be fastest sending single requests with a high degree of parallelism. If you feel batch size might improve performance, it is best to start with a small batch size of 10 and increase concurrency until you start getting service protection API limit errors that you will retry.

With the SDK for .NET this means using ExecuteMultipleRequest, which typically allows sending up to 1000 operations in a request. The main benefit this provides is that it reduces the total amount of XML payload that must be sent over the wire. This provides some performance benefit when network latency is an issue. For service protection limits it increases the total execution time per request. Larger sized batches increase the chance you will encounter execution time limits rather than limits on the number of requests.

In the past, ExecuteMultiple operations were limited to just 2 at a time because of the impact on performance that this could have. This is no longer the case, because service protection execution time API limits have made that limit redundant.

When using the Web API, the smaller JSON payload sent over the wire for individual requests means that network latency is not an issue. More information: Execute batch operations using the Web API

Note

Batch operations are not a valid strategy to bypass entitlement limits. Service protection API limits and Entitlement limits are evaluated separately. Entitlement limits are based on CRUD operations and accrue whether or not they are included in a batch operation. More information: Entitlement limits

Strategies to manage Service Protection API limits

This section describes ways that you can design your clients and systems to avoid service protection API limit errors. You may also want to consider how you manage your licenses to reduce the impact.

Update your client application

Service Protection API limits have been applied to Dataverse since 2018, but there are many client applications written before these limits existed. These clients didn't expect these errors and can't handle the errors correctly. You should update these applications and apply the patterns to Retry operations described above.

Move towards real-time integration

Remember that the main point of service protection API limits is to smooth out the impact of highly demanding requests occurring over a short period of time. If your current business processes depend on large periodic nightly, weekly, or monthly jobs which attempt to process large amounts of data in a short period of time, consider how you might enable a real-time data integration strategy. If you can move away from processes that require highly demanding operations, you can reduce the impact service protection limits will have.

Frequently asked questions

This section includes frequently asked questions. If you have questions that are not answered, please post them using the Feedback button at the bottom of this page to submit feedback about this page.

I'm using an ETL application I licensed. How do I get optimum throughput?

Work with the ETL application vendor to learn which settings to apply. Make sure you are using a version of the product that supports the Retry-After behavior.

No. Dataverse native search is a different API (api/search rather than api/data) and has different rules. When using the Dataverse search API, there is a throttling limit of one request per second for each user.

More information: Dataverse Search Service Protection Limits

How do these limits apply to how many requests a user is entitled to each day?

These limits are not related to entitlement limits. More information: Entitlement limits

Are limits applied differently for application users?

No. The limits are applied to all users in the same way.

See also

Administer Power Platform / Licensing and license management / Requests limits and allocations
Dataverse API limits overview
Use Dataverse Web API
Use Dataverse SDK for .NET