message queues durability vs latency what's your trade-off

hey everyone, been thinking about message queue implementations lately. when building systems that rely heavily on queues, especially for critical data, durability is obviously key. but sometimes, the configurations that ensure maximum durability (like synchronous writes to disk, strict replication) can introduce noticeable latency. on the flip side, prioritizing low latency might mean accepting a slightly higher risk of message loss under certain failure scenarios. what are your experiences balancing these two? when have you had to make a tough call between guaranteed delivery and lightning-fast processing, and what factors influenced your decision?