I'm in the opposite camp. Transactions can fail, however retrying them is not the solution. It hides the symptom (which is not a problem with proper monitoring), but more importantly, it can lead to overwhelming the db server. Sometimes failure happens because of the load, in which case retrying the query is counterproductive. And even in cases where retrying would be the correct approach, it is the responsibility of the app logic, not of the db connection layer, to retry the query. Imho of course. :)
