message queues best practices for handling failures, like how to deal with things when they go wrong

Hey everyone, I've been thinking a lot about using message queues, like Kafka or RabbitMQ, for our asynchronous tasks. We're looking at decoupling services and improving resilience. A big concern for me is how to properly handle failures when messages are being processed. What are your go-to strategies for ensuring messages aren't lost if a consumer crashes or fails to process a message? Are you using dead-letter queues, retries with backoff, or something else entirely? I'm really curious to hear what's worked well in production.