Stage 3 · Build

Performance, Memory & Debugging

GC Tuning

GOGC, GOMEMLIMIT, heap goals, pacer behavior, and latency tradeoffs.

8 min readGo for SREBuild

GC Overview

Go uses a concurrent, tri-color mark-and-sweep garbage collector. It runs concurrently with application code, minimizing pause times. The GC targets low latency over high throughput.

GC related flags


# Enable GC tracing
GODEBUG=gctrace=1 ./myapp

# Set GC target percentage
GOGC=100 ./myapp

# Set memory limit
GOMEMLIMIT=1GiB ./myapp

# Disable GC (dangerous)
GOGC=off ./myapp

# View GC stats programmatically
go tool pprof http://localhost:6060/debug/pprof/heap?gc=1

GODEBUG=gctrace=1 prints GC timing and heap statistics. GOGC controls the GC target. GOMEMLIMIT sets a soft memory limit. These are the primary knobs for GC tuning.

GOGC Tuning

GOGC controls the tradeoff between memory usage and CPU overhead. It is the percentage of live heap that triggers a GC cycle. The default is 100 — GC triggers when the heap doubles.

Gogogc-examples.go

27 linesLn 1, Col 1Go

GOGC=100 means GC runs when the heap grows by 100% over live data. Lower GOGC = more frequent GC = less memory = more CPU. Higher GOGC = less frequent GC = more memory = less CPU. Profile to find the right balance.

GOMEMLIMIT

GOMEMLIMIT sets a soft memory limit. The GC becomes more aggressive as the limit approaches. This prevents OOM kills in containers with fixed memory limits.

Gogomemlimit-usage.go

22 linesLn 1, Col 1Go

GOMEMLIMIT provides a safety net against OOM kills. The GC adjusts its aggressiveness based on memory pressure. At 50% of limit, GC starts to increase frequency. At 90%, GC becomes very aggressive. Always set this in containers.

Heap Goals

The GC sets heap goals based on GOGC and live heap size. Understanding these goals helps predict memory usage and GC behavior.

Goheap-goal-calculations.go

21 linesLn 1, Col 1Go

m.NextGC is the heap size that triggers the next GC. m.HeapAlloc is the current live heap. m.HeapObjects is the number of live objects. These metrics help you understand GC pressure and tune GOGC/GOMEMLIMIT.

GC Pacer

The GC pacer coordinates the concurrent GC with application goroutines. It adjusts the assist ratio — how much GC work each goroutine must do — to meet the heap goal.

Gopacer-behavior.go

31 linesLn 1, Col 1Go

The gctrace output shows detailed GC timing. The assist ratio controls how much application goroutines help with GC. High assist ratio means the GC is catching up and application throughput drops. Low assist ratio means the GC is ahead.

Monitoring GC

Gogc-monitoring.go

30 linesLn 1, Col 1Go

Monitor GC pauses, heap size, and GC cycle count. Histogram of pause times shows the distribution. Track heap growth rate to predict memory needs. Alert on sudden increases in GC frequency or pause duration.

Start with GOMEMLIMIT

For containerized services, start by setting GOMEMLIMIT to 80% of the container memory limit. This prevents OOM kills without tuning GOGC. Only tune GOGC if you need to optimize the memory-CPU tradeoff further.

Mark this lesson complete to store local progress and unlock a cleaner resume path the next time you visit.