CloudConsole

ThrottlingException: Rate exceeded

Updated 16 November 2026 · How we write and check guides

Short answer

An AWS API is rejecting calls because your account is sending them faster than that API's per-account, per-region rate limit. Find which caller and operation are being throttled (CloudTrail records the errorCode), then reduce the call rate: cache results, batch requests, replace polling with events, and use the SDK's standard or adaptive retry mode. Raise the quota in Service Quotas only where the limit is adjustable.

Fix it, in 5 steps: 1. Let the SDK back off properly; 2. Cache; 3. Batch; 4. Replace polling with events; 5. Raise the…
The diagnosis and fix steps from this page, in the order to try them; commands and details are below.

What it means

Most AWS control-plane APIs (CloudWatch, Systems Manager, Secrets Manager, CloudFormation, ECS, IAM and many more) enforce request-rate limits per account and region. Go over the limit and the call fails with HTTP 400:

An error occurred (ThrottlingException) when calling the GetParameter operation (reached max retries: 2): Rate exceeded

The SDK has usually already retried a few times by the time you see it, as the "reached max retries" part shows. Because the limit is shared across the whole account in that region, the code that fails isn't necessarily the code that's causing it; a noisy neighbour in the same account can throttle your service.

Amazon Bedrock uses the same exception for model quotas, with messages such as Too many requests, please wait before trying again or Too many tokens. There the limits are per model and region, counted in requests and tokens per minute.

Common causes

PatternExample
Fetching config on every invocationEach Lambda call runs GetParameter or GetSecretValue; a traffic spike multiplies it
Polling loopsScripts or dashboards calling DescribeStacks, DescribeServices or GetMetricData every few seconds
Fan-out jobsA Step Functions Map state or a batch job running hundreds of identical API calls at once
Tooling at scaleTerraform plans over large states, or inventory tools sweeping every resource
Bedrock burstsLong prompts or high concurrency exceeding tokens-per-minute for the model

Find who is being throttled

If CloudTrail is delivered to CloudWatch Logs, this Logs Insights query ranks throttled calls by operation and caller:

filter errorCode like /Throttl|RequestLimitExceeded/
| stats count(*) as throttled by eventSource, eventName, userIdentity.arn
| sort throttled desc
| limit 20

A single role dominating the list is your culprit. Spread across many roles points to a shared limit you need to raise or a shared cache you need to add.

Fix it

  1. Let the SDK back off properly. Standard retry mode uses exponential backoff with jitter; adaptive mode also slows the client down when it sees throttling.
    [profile batch]
    retry_mode = adaptive
    max_attempts = 10
    The same can be set with AWS_RETRY_MODE and AWS_MAX_ATTEMPTS.
  2. Cache. Read parameters and secrets once per execution environment rather than per request; for Lambda, the AWS Parameters and Secrets Lambda Extension does this for you.
  3. Batch. Use GetParameters or GetParametersByPath instead of many single calls; ask GetMetricData for several metrics at once.
  4. Replace polling with events, such as EventBridge rules for ECS task state, CloudFormation stack status or EC2 state changes.
  5. Raise the limit where possible. Check Service Quotas for an adjustable rate quota; some API limits are only changed through AWS Support, and some can't be changed.
    aws service-quotas list-service-quotas --service-code ssm \
      --query 'Quotas[?contains(QuotaName, `rate`) || contains(QuotaName, `TPS`)].[QuotaName,Value,Adjustable]' \
      --output table

Bedrock specifics

For Bedrock, check the model's tokens-per-minute and requests-per-minute quotas in Service Quotas for the region you call. Reducing max_tokens, trimming prompts and limiting client concurrency all help. Cross-region inference profiles can spread load over several regions, and provisioned throughput gives reserved capacity for steady high volume. Retrying without backoff only makes Bedrock throttling worse.

Questions

What does ThrottlingException: Rate exceeded mean in AWS?

Your account sent requests to an API faster than its per-account, per-region rate limit. The request was rejected without being processed and can be retried after a short, increasing delay.

How do I fix ThrottlingException in Bedrock?

Lower your request and token rate, add backoff with jitter, and check the model's per-minute quotas in Service Quotas. Request a quota increase or use cross-region inference if you consistently need more.

Does retrying make throttling worse?

Immediate retries from many clients do. Exponential backoff with jitter, which the SDK's standard and adaptive retry modes provide, spreads the retries out so the limit can recover.

CloudConsole home screen showing what needs attention on a phone, for following up on ThrottlingException away from your desk
When this happens away from your desk

Throttling usually starts during a traffic spike, when you are least likely to be at a desk. CloudConsole, a phone app for AWS, is built to show the alarm and the throttled calls on your phone so you can see which service is the noisy one.

Did it start after a deploy? What changed? on the app’s alarm screen lists the deploys and configuration changes from the hour before the alarm fired.

About the app → · Download on the App Store

Sources