Right, you just need to avoid having the btree index on the UUID field. Similarly, don't pose queries to sort by the randomized UUID field either. This is where

Right, you just need to avoid having the btree index on the UUID field. Similarly, don't pose queries to sort by the randomized UUID field either. This is where the often maligned hash index type could be useful, to allow lookup of individual rows by UUID without all the expense of a btree index maintenance. Use other fields for ordering that have better write locality. What this means in practice, of course, is that you shouldn't expect to do application driven pagination with UUID keys either. You would need to expose some other boundary marker with a total order that works well with btrees. And this could bring you back to "leaking" predictable key material that you were trying to hide by adopting UUIDs...