A long while back I created this issue to support a way for Gremlin Server to retry failed transactions automatically. Aside from the difficulties discussed with Dylan in the issue itself, I'm realizing now that the complexity is more that Gremlin Server should probably bear.
Barring objection I will assume lazy consensus after 72 hours (Thursday, September 22, 2016, 7am EST) and simply close this one: https://issues.apache.org/jira/browse/TINKERPOP-739