message queues, durability versus latency, trade-offs?

i been wrestling with message queue configs lately, esp durability vs latency. when building microservices that need event-driven comms, we often need messages to be saved reliably. but, super strict durability can sometimes add noticeable latency. so, what are your strategies for balancing these two? any specific patterns or configs uve found effective for high-throughput, low-latency stuff, while still making sure messages get delivered tho