When an app is OOM (out of memory) killed, there is downtime in between when it's killed and when the server finishes rebooting. I would like render to not kill the process until another instance has booted (just like it works in zero-downtime deployments), and use swap in the meantime to stay within the memory limits. This might give somewhat degraded performance while the replacement server is booting, but it would be far superior to downtime.