We shipped a couple improvements related to Node.js crashes for Fluid compute lately: Generally, node.js exits the process when you have an unhandled exception or unhandled rejected promise in your code. While that is a good default, it can be surprising that any such error may interrupt all concurrent requests being handled by a node process. You can override the crashing for your own code, but it isn't safe in general. Really, a well-behaved program should never crash in this way. That is why we shipped a logs filter for fatal crashes, so you can easily check whether you have any such issues in your code and address them. Additionally, we changed Fluid compute auch that when such an error occurs, the function immediately denies further requests from the load balancer, but instead of crashing right away it gives all in-flight requests handled by the same node process time to finish. Only once they are done, the process is exited. The latter seems like a good default for node.js web servers. I'm not sure it is "best practice", but I think it should be. https://t.co/rF7bC0aYhM