Profiling with pprof

The profiling workflow

Never optimise without data. The Go toolchain ships with pprof, a profiling tool that identifies exactly which functions consume the most CPU time or allocate the most memory.

The workflow is always:

1. Collect a profile (CPU or memory)
2. Analyse with go tool pprof
3. Identify the hottest functions
4. Optimise and re-measure

CPU profiling via go test

The easiest way to collect a CPU profile is through the test runner:

Inside pprof:

Memory profiling

Key pprof commands for memory:

• top — functions by allocated bytes
• list funcName — show allocations per line
• -alloc_objects vs -inuse_objects — total allocated vs currently live

net/http/pprof — always-on profiling in production

Import the pprof HTTP handler as a side effect in your main.go:

Then expose a debug server (on a separate port, internal-only):

Now you can collect live profiles from a running production server:

Security: Never expose :6060 publicly. Bind to localhost or protect with a firewall rule.

Reading a flame graph

A flame graph shows call stacks as nested rectangles:

• Width = proportion of CPU time (or allocations)
• The widest frames at the top are the hottest call sites
• Click a frame to zoom in

runtime.MemStats — inline memory inspection

For quick in-process checks without pprof:

• Alloc — bytes currently allocated on the heap
• TotalAlloc — bytes allocated over the entire lifetime (monotonically increasing)
• NumGC — number of GC cycles completed

Common findings and fixes