Thinking of nginx as the server in front of your server is right for most production layouts: terminate TLS here, decide what is cached versus passed through, and keep app processes off the public ports. I keep static and cacheable responses at the edge when I can, and I only proxy dynamic paths with Host, X-Forwarded-For, and X-Forwarded-Proto set so the app sees the public scheme and hostname.
Ops checks that catch most outages: reload instead of restart after config changes, watch error_log for connect() failed versus upstream timed out, and size worker_connections for both client and upstream sockets. After a change I hit a cached asset and a dynamic path separately — a mis-scoped proxy_pass or missing trailing-slash rule often looks like an app bug when it is still an edge config issue.