> 2. A PGPSQL function that attempts to INSERT, catches the duplicate key exception and instead UPDATEs (also handling the case that the row vanishes), which is

> 2. A PGPSQL function that attempts to INSERT, catches the duplicate key exception and instead UPDATEs (also handling the case that the row vanishes), which is slower than hell That's not true, at least not when you're operating on a single row at a time. According to my benchmarks the function approach was usually within 10% of the native upsert -- even outperforming it by around 5% depending on data distribution. For insert-or-select (instead of update) the function consistently outperformed INSERT ... IGNORE.