process.on('unhandledRejection') catches them, but the process state may already be compromised. Some people crash and restart, others log and continue. What's the least dangerous policy for a long-running service?