# Implement best practices with CloudWatch Alarms


Add CloudWatch alarms for when the following metrics exceed the buffering limit (a maximum of 15 minutes).
+ `DeliveryToS3.DataFreshness`
+ `DeliveryToIceberg.DataFreshness`
+ `DeliveryToSplunk.DataFreshness`
+ `DeliveryToAmazonOpenSearchService.DataFreshness`
+ `DeliveryToAmazonOpenSearchServerless.DataFreshness`
+ `DeliveryToHttpEndpoint.DataFreshness`

Also, create alarms based on the following metric math expressions.
+ `IncomingBytes (Sum per 5 Minutes) / 300` approaches a percentage of `BytesPerSecondLimit`.
+ `IncomingRecords (Sum per 5 Minutes) / 300` approaches a percentage of `RecordsPerSecondLimit`.
+ `IncomingPutRequests (Sum per 5 Minutes) / 300` approaches a percentage of `PutRequestsPerSecondLimit`.

Another metric for which we recommend an alarm is `ThrottledRecords`.

For information about troubleshooting when alarms go to the `ALARM` state, see [Troubleshoot errors in Amazon Data Firehose](troubleshooting.md).