We run in serializable mode to make sure that our writes are actually atomic and zedtokens and the snapshots they represent are accurate.