Go Runtime Performance

Measurement playbook

1. Reproduce
2. Profile (CPU, allocs, trace)
3. Classify (CPU/alloc/lock/IO)
4. Hypothesize
5. Verify
6. Guard with benchmarks

Tooling commands

• go test -bench . -benchmem
• go tool pprof -http=:0 cpu.out
• go test -trace trace.out + go tool trace
• GODEBUG=gctrace=1 for GC lines

Allocations

• Track allocs/op and bytes/op
• Preallocate slices/maps
• Avoid string/byte conversions

GC

• Cost scales with live heap
• Reduce long‑lived objects
• Prefer compact data
• Tune with GOGC / GOMEMLIMIT

Goroutines + scheduler

• Coarse‑grained goroutines
• Batch work; avoid lock contention
• M/P/G model; cgo and syscalls can pin threads

Memory model

• Synchronization establishes happens‑before
• No data races (ever)
• Use channels/atomics for safe publication

Generics

• Measure generic vs specialized
• Avoid interface boxing in hot paths
• Constrained types help inlining

Zero‑alloc patterns

• Caller‑provided buffers
• No fmt on hot paths
• Short‑buffer errors