Why is AWS::Lambda::EventInvokeConfig's limit for MaximumRetryAttempts 2?

When you invoke a function asynchronously, the event is added to an internal queue and then the Lambda service gets the messages from the queue and invokes the function. It will try to invoke the function up to a maximum of 3 times (2 retries - configurable) and if it fails, it will send the event to a DLQ (if configured).

For different reasons, it may be that the event will remain in the queue and will not be handled immediately. In that case you can configure the max event age (up to the max of 6 hours, which is the default). If the event remains in the queue for longer that specified, it wil be dropped.

Thanks, I appreciate the help. Sadly this is the answer (throttling) I was dreading and it means I can not do an easy exponential backoff [1] [2] for myself and the AMQP replacement with EventBride -> Lambda (w/Destinations) system I am building.

If Destinations allowed me to have MaximumRetryAttempts more than 2 I could get something similar (up to 6 hours) to the service-level behavior for my client code. Building code like (https://bitesizedserverless.com/bite/async-lambda-function-retries-with-backoff-and-jitter/) is not a solution. All we need is more attempts.

system errors (500-series), Lambda returns the event to the queue and attempts to run the function again for up to 6 hours

Just to make sure, when you say "system" you mean AWS service, not my client side 500 errors right?

I know it's confusing because there are two types of retries. Uri correctly pointed out that the age limit in the queue and retry limit are for different reasons, it's not about those retries "using that limit". As in https://docs.aws.amazon.com/lambda/latest/dg/invocation-async.html, retries (max 2) are for when your function returns an error. The age limit is for this situation where many more than 2 retries may occur:

If the function doesn't have enough concurrency available to process all events, additional requests are throttled. For throttling errors (429) and system errors (500-series), Lambda returns the event to the queue and attempts to run the function again for up to 6 hours. The retry interval increases exponentially from 1 second after the first attempt to a maximum of 5 minutes. If the queue contains many entries, Lambda increases the retry interval and reduces the rate at which it reads events from the queue.