Idempotency in distributed systems, what do you usually do

hey all, ive been wrestling with ensuring idempotency in a few microservices lately, especially when dealing with external event triggers. weve implemented a few strategies like unique request ids and checking state before processing, but it feels like theres always an edge case. what patterns or techniques do you find most effective and reliable for guaranteeing that an operation only happens once, even if its retried multiple times due to network hiccups or service failures? curious to hear your war stories and preferred solutions.