So... Here's a fun one. New customer, retrofitted the old app to use Lucee 5.x in a container (official image). Passes all tests in dev, push it to test in ECS on AWS all tests pass, everything is fine. Build a new ECS cluster for prod, launch the exact same image and half the app fails. No rhyme or reason. It's like the extensions didn't load -- but they are baked into the container image. Since it's on ECS with Fargate, no access to the OS to poke around the logs (other than the ones I transported to /dev/stdout). Push new versions, restart the task/service, all the same thing. Futz around with it for nearly a day... Finally, say f-it, and build /another/ new cluster, same configs, everything comes up clean and all tests pass....
Ticket with AWS, and their only solution is to "add more RAM to the task."