golang-performance

Follow the iterative cycle: define a metric, write an atomic benchmark, measure a baseline with -benchmem and -count=6, apply one optimization at a time, then compare with benchstat. Use the decision tree to map your pprof signal to the right fix.

Use fgprof, which captures both on-CPU and off-CPU time such as I/O wait and blocked goroutines. Standard pprof CPU profiles only sample on-CPU time, so external bottlenecks like database calls appear invisible.

Reuse slice backing arrays with append(s[:0], ...), preallocate maps with size hints, use sync.Pool for short-lived objects under 32KB, and avoid interface boxing with generics. Verify with go test -bench -benchmem to track allocs/op.

Yes. Go adds padding to satisfy alignment requirements, so ordering fields from largest to smallest reduces struct size. Run the fieldalignment tool to detect structs with wasted padding automatically.

Avoid pooling objects larger than 32KB, infrequently allocated objects, and buffers returned to callers who may retain them. Always reset state before Put and return copies rather than pooled memory.

Go maps never release bucket memory when entries are deleted, so a map that held millions of keys retains its peak allocation. Replace the map with a fresh make() to let the old bucket array be garbage collected.