@wwilliams Thank you - this is very useful. Unfortunately I have limited control over the starting and stopping of the services as they are all part of the Amazon Machine Image, which is copied anew onto each server instance as it spins up. I could provide patches or make a custom image, but working out exactly what to change is hit-and-miss and could take a long time.
It is quite possibly the same pathology you were seeing though: HTTP requests being sent to Tomcat and then to Lucee before Lucee has finished getting started.
How did you work out what was happening and do you know any way to increase the debug output during startup?
[Update]
I don’t think it’s early HTTP requests: even if no requests come in for several minutes, the application still doesn’t start cleanly. It feels like Lucee is taking a long time to start and Tomcat can’t cope for some reason.