Problem
The message queue feeding preprocessing delivers at-least-once, so roughly 0.5% of the time the exact same event arrives twice. A worker restart made it obvious the current fix wasn't real: dedup state that only lived in the worker's own memory vanished on restart, and revenue-per-user came out inflated for every user whose event got redelivered right after a deploy. Make duplicate handling survive the thing it's supposed to survive.
Required components
Click to add a component to the canvas.
Additional components
Browse by category. Click or drag a component onto the canvas.