When you have a lambda that is driven by a SQS queue, like this, your lambda can receive up to ten messages per batch. Your handler can look like this, handling all of the messages in a single event in a for loop.

def lambda_handler(event: dict, context: dict) -> None:
    for msg in event["Records"]:

If you do nothing and all messages are handled without exceptions, the messages will be deleted from the SQS queue automatically for you. Logically, if there is an error, the messages will not be deleted. They will be put back to the queue, or, depending on the arrangement, they will be sent to the dead letter queue. But if your handler function raises an exception, the whole batch will be failed, including the messages that have been processed already. This is typically not the behaviour you want, and to solve this you have to delete or put messages to the queue, keeping track of the failures and succesfully handled messages yourself. This is not very obvious, and you can find some questions on how to handle this properly here and here.

AWS has introduced a new possibility for handling this, pretty recently, in December 2021. If you include the failed messages in a lambda response called batchItemFailures, only those will be reposted to the queue (or the dead letter queue). In python, this looks like this.

def lambda_handler(event: dict, context: dict) -> dict:
    batch_item_failures = []  # list of things that failed

    for msg in event["Records"]:
        except Exception as e:  # more specific is better

    return {
        "batchItemFailures": batch_item_failures