What strategies help you design SQS based systems for exactly once processing semantics despite the underlying at least once delivery guarantee?

Since Standard SQS queues guarantee at least once delivery, meaning a message could theoretically be delivered more than once, applications should design their message processing logic to be idempotent, meaning processing the same message multiple times produces the same end result, often achieved by tracking already processed message identifiers in a database or using conditional writes, rather than relying purely on the queue itself to guarantee exactly once processing.

// Idempotent processing pattern
if not database.exists(message.id):
    process(message)
    database.mark_processed(message.id)