Spent half a day on a production issue where a service was reachable externally but health checks timed out internally. Turned out two rules in the FORWARD chain were in the wrong order. iptables matches top-down and stops at first hit — no second chances for the same packet.
Memory crept up to 2 GB within three days of launch. Restarts bought time but weren't a fix. Enabled the pprof endpoint and watched goroutine count climb ~60 per minute. Root cause: an unclosed channel left receivers blocked forever.
Every time I set up a reverse proxy I end up looking up the same things. Writing them down: upstream health checks, header forwarding, TLS termination, buffer tuning. Mostly for myself.
Three usual culprits: wrong base image, no layer caching strategy, build artifacts left behind. Multi-stage builds and an Alpine base got us most of the way there. Notes on what worked and what didn't.
What do cost, rows, actual time, and loops actually mean? Is a Seq Scan always bad? When does an index make things worse? A few real query examples to make the numbers less abstract.