Request throttling for the Amazon EC2 API

Amazon EC2 throttles EC2 API requests for each AWS account on a per-Region basis. We do this to help the performance of the service, and to ensure fair usage for all Amazon EC2 customers. Throttling ensures that requests to the Amazon EC2 API do not exceed the maximum allowed API request limits. API requests are subject to the request limits whether they originate from:

If you exceed an API throttling limit, you get the RequestLimitExceeded error code.

How throttling is applied

Amazon EC2 uses the token bucket algorithm to implement API throttling. With this algorithm, your account has a bucket that holds a specific number of tokens. The number of tokens in the bucket represents your throttling limit at any given second.

Amazon EC2 implements two types of API throttling:

Request rate limiting

With request rate limiting, each API is evaluated individually, and you are throttled on the number of requests you make on a per-API basis. Each request that you make removes one token from the API's bucket. For example, the token bucket size for DescribeHosts, a non-mutating API action, is 100 tokens. You can make up to 100 DescribeHosts requests in one second. If you exceed 100 requests in a second, you are throttled on that API and the remaining requests within that second fail, however, requests for other API are not affected.

Buckets automatically refill at a set rate. If the bucket is below its maximum capacity, a set number of tokens is added back to it every second until it reaches its maximum capacity. If the bucket is full when refill tokens arrive, they are discarded. The bucket can't hold more than its maximum number of tokens. For example, the bucket size for DescribeHosts, a non-mutating API action, is 100 tokens and the refill rate is 20 tokens per second. If you make 100 DescribeHosts requests in one second, the bucket is reduced to zero (0) tokens. The bucket is then refilled by 20 tokens every second, until it reaches its maximum capacity of 100 tokens. This means that an empty bucket reaches its maximum capacity after 5 seconds if no requests are made during that time.

You do not need to wait for the bucket to be completely full before you can make API requests. You can use refill tokens as they are added to the bucket. If you immediately use the refill tokens, the bucket does not reach its maximum capacity. For example, the bucket size for DescribeHosts, a non-mutating API action, is 100 tokens and the refill rate is 20 tokens per second. If you deplete the bucket by making 100 API requests in a second, you can continue to make 20 API requests per second by using the refill tokens as they are added to the bucket. The bucket can refill to the maximum capacity only if you make fewer than 20 API requests per second.

Resource rate limiting

Some API actions, such as RunInstances and TerminateInstances, as described in the table that follows, use resource rate limiting in addition to request rate limiting. These API actions have a separate resource token bucket that depletes based on the number of resources that are impacted by the request. Like request token buckets, resource token buckets have a bucket maximum that allows you to burst, and a refill rate that allows you to sustain a steady rate of requests for as long as needed. If you exceed a specific bucket limit for an API, including when a bucket has not yet refilled to support the next API request, the action of the API is limited even though you have not reached the total API throttle limit.

For example, the resource token bucket size for RunInstances is 1000 tokens, and the refill rate is two tokens per second. Therefore, you can immediately launch 1000 instances, using any number of API requests, such as one request for 1000 instances or four requests for 250 instances. After the resource token bucket is empty, you can launch up to two instances every second, using either one request for two instances or two requests for one instance.

For more information, see Resource token bucket sizes and refill rates.

Throttling limits

The following sections describe the request token bucket and resource token bucket sizes and refill rates.

Request token bucket sizes and refill rates

For request rate limiting purposes, API actions are grouped into the following categories:

The following table shows the request token bucket sizes and refill rates for all AWS Regions.

Resource token bucket sizes and refill rates

The following table lists the resource token bucket sizes and refill rates for API actions that use resource rate limiting.

Monitor API throttling

You can use Amazon CloudWatch to monitor your Amazon EC2 API requests and to collect and track metrics around API throttling. You can also create an alarm to warn you when you are close to reaching the API throttling limits. For more information, see Monitor Amazon EC2 API requests using Amazon CloudWatch.

Retries and exponential backoff

Your application might need to retry an API request. For example:

However, for a client error (4xx), you must revise the request to correct the problem before trying the request again.

Resource status changes

Before you start polling to check for status updates, give the request time to potentially complete. For example, wait a few minutes before checking whether your instance is active. When you begin polling, use an appropriate sleep interval between successive requests to lower the rate of API requests. For best results, use an increasing or variable sleep interval.

Alternatively, you can use Amazon EventBridge to notify you of the status of some resources. For example, you can use the EC2 Instance State-change Notification event to notify you of a state change for an instance. For more information, see Automate Amazon EC2 using EventBridge.

Retries

When you need to poll or retry an API request, we recommend using an exponential backoff algorithm to calculate the sleep interval between API requests. The idea behind exponential backoff is to use progressively longer waits between retries for consecutive error responses. You should implement a maximum delay interval, as well as a maximum number of retries. You can also use jitter (randomized delay) to prevent successive collisions. For more information, see Timeouts, retries, and backoff with jitter.

Each AWS SDK implements automatic retry logic. For more information, see Retry behavior in the AWS SDKs and Tools Reference Guide.

Request a limit increase

You can request an increase for API throttling limits for your AWS account.