Idempotency in API Design: Is it a Must-Have or Too Much Trouble?

hey folks, ive been thinking a lot about idempotency lately, especially when designing APIs that deal with financial transactions or critical state changes. making sure a request can be safely retried without causing duplicate operations seems crucial. but implementing and testing for idempotency can add a layer of complexity. what are your thoughts? is it always worth the overhead, or are there scenarios where you'd accept the risk of non-idempotent operations? how do you typically handle unique request identifiers and server-side checks?