PeterBooker/profile

CPU and Memory Profiling

Profile Go code to identify CPU hotspots and memory allocators using pprof.

Usage

• /profile cpu ./internal/index/ - CPU profiling on index package
• /profile memory ./internal/repo/ - Memory profiling on repo package
• /profile all ./... - Both CPU and memory on all packages

Steps

1. Parse arguments

   • First argument: Profile type (cpu, memory, or all)
   • Second argument: Package path (defaults to ./...)

2. Create profile output directory

   mkdir -p .profiles
   

3. Run profiling benchmarks

   For CPU profiling:

   go test -cpuprofile=.profiles/cpu.prof -bench=. $PACKAGE 2>&1
   

   For memory profiling:

   go test -memprofile=.profiles/mem.prof -bench=. $PACKAGE 2>&1
   

4. Analyze CPU profile

   go tool pprof -top -cum .profiles/cpu.prof 2>&1 | head -30
   

   Identify:

   • Top 10 CPU consumers by cumulative time
   • Functions with high self time (computation hotspots)
   • Unexpected entries (potential optimization targets)

5. Analyze memory profile

   go tool pprof -top -alloc_space .profiles/mem.prof 2>&1 | head -30
   

   Identify:

   • Top allocators by total bytes
   • Functions with high allocation counts
   • Potential sources of GC pressure

6. Generate flamegraph data (if requested)

   go tool pprof -raw .profiles/cpu.prof > .profiles/cpu.raw
   

7. Report findings

   Structure the report as:

   CPU Hotspots

   | Function | Self% | Cum% | Observation | |----------|-------|------|-------------|

   Memory Allocators

   | Function | Bytes | Allocs | Observation | |----------|-------|--------|-------------|

   Optimization Suggestions

   • List specific, actionable recommendations
   • Reference line numbers where applicable
   • Note any patterns (e.g., repeated allocations in loops)

Interpreting Results

CPU Profile Indicators

• High self%: Direct computation hotspot
• High cum% but low self%: Calls expensive functions
• runtime.*: GC or scheduler overhead

Memory Profile Indicators

• High alloc_space: Total memory pressure
• High alloc_objects: GC pressure from many small allocations
• Repeated patterns: Loop allocations, string concatenation

Common Hotspots in Veloria

Watch for issues in:

• (*Index).Search - Regex compilation, line reading
• (*Repository).Load - Index file mapping
• (*IndexedExtension).Update - Hot-swap operations
• HTTP handlers - JSON marshaling, response writing

Cleanup

Profile files are stored in .profiles/. Add to .gitignore if not already present.