Why systems fail catastrophically under overload instead of degrading gracefully, and the patterns - load shedding, backpressure, circuit breakers - that prevent cascading failure.
Production traffic will test your system whether you do it first or not. Load testing before launch finds the capacity limits, bottlenecks, and failure modes that only appear under real conditions.
What memory leaks actually are in garbage-collected languages, the common patterns that cause them, and how to use heap profiling to track down and fix them.
How to read a Postgres EXPLAIN ANALYZE output, what sequential scans and index scans mean, and how to identify and fix slow queries before they become incidents.
Why guessing at performance bottlenecks wastes time, and how to use profiling tools to find what actually needs fixing before you touch the code.
Caches speed things up until they serve stale data. Understanding the tradeoffs at each caching layer - and how invalidation actually works - is what keeps them from becoming bugs.
