The Go Runtime

Overview

Before writing more code, it’s critical to understand how Go actually executes it. This chapter covers the hidden mechanisms of the Go runtime: memory management, the garbage collector, and performance tuning.

Understanding these concepts transforms you from writing “random Go code” to writing code that cooperates with the runtime for optimal performance.

Stack vs Heap (Escape Analysis)

Imagine every function in Go is a small room. When you create a variable, Go has to decide:

“Can this thing stay inside the room, or do I need to put it outside where everyone can reach it?”

Memory Locations

Stack (The Room)
Heap (The Outside)

If Go is 100% sure the variable is used only inside that room, it keeps it on the stack.Characteristics:

When you leave the room (function returns), the variable is gone
Fast allocation and deallocation
No garbage collection overhead
Limited in size

func calculate() int {
    x := 42  // Stays on stack
    y := x * 2
    return y
}

If Go is not sure—even a little bit unsure—it moves the variable to the heap so it doesn’t disappear too early.Characteristics:

Persists beyond function scope
Requires garbage collection
Slower allocation
Can grow dynamically

func createUser() *User {
    u := &User{Name: "Alice"}  // Escapes to heap
    return u
}

Escape Analysis in Action

Escape Analysis is Go’s compile-time decision process. To see this in action:

go build -gcflags="-m" .

Go will literally tell you what it decided and why:

./main.go:5:6: can inline calculate
./main.go:11:6: cannot inline createUser
./main.go:12:7: &User{...} escapes to heap

If you don’t look at escape analysis output, you’re guessing about performance. Excessive heap allocations can significantly slow down your program.

Common Escape Scenarios

When Variables Escape to Heap

Returning Pointers

func newCounter() *int {
    count := 0
    return &count  // ❌ Escapes
}

Interface Conversions

func log(v interface{}) {
    fmt.Println(v)
}

x := 42
log(x)  // ❌ x escapes (interface{} can hold anything)

Closures

func counter() func() int {
    count := 0
    return func() int {  // ❌ count escapes (outlives function)
        count++
        return count
    }
}

Slices Growing Beyond Known Size

func makeSlice(n int) []int {
    s := make([]int, n)  // ❌ May escape if n is not constant
    return s
}

Garbage Collector (GC)

Imagine the Heap is a playground:

Objects: Kids running around
Garbage: Forgotten toys lying around

How the GC Works

The Garbage Collector is a cleaning robot that walks around while kids are still playing. It checks:

“Is anyone still holding this toy?”

Yes → Leave it
No → Pick it up and throw it away

Mark Phase

The GC traces from root objects (goroutine stacks, global variables) and marks all reachable objects.

Sweep Phase

The GC reclaims memory from unmarked objects that are no longer reachable.

Concurrent Collection

Go’s GC runs concurrently with your application, but sometimes it needs brief “Stop-The-World” pauses.

The Cost of Garbage Collection

The GC gets tired if you keep throwing new toys on the ground every second. Lots of tiny heap allocations slow programs down, even if memory looks “free”.

Performance Impact:

// ❌ Bad: Creates garbage every iteration
for i := 0; i < 1000000; i++ {
    data := fmt.Sprintf("item-%d", i)  // Allocates on heap
    process(data)
}

// ✅ Better: Reuse buffers
var buf bytes.Buffer
for i := 0; i < 1000000; i++ {
    buf.Reset()
    fmt.Fprintf(&buf, "item-%d", i)
    process(buf.String())
}

sync.Pool

Imagine instead of throwing toys away, you have a shared toy box. When a kid finishes playing, they put it back in the box instead of throwing it away. The next kid reuses it.

sync.Pool is NOT a Cache

A cache is like a fridge: you put food in and expect it to be there later.sync.Pool is a Recycling Bin:

You throw something in, maybe someone reuses it
The GC is allowed to empty the bin whenever it wants

The Golden Rule

Your program must behave exactly the same if the pool were always empty.

var bufferPool = sync.Pool{
    New: func() interface{} {
        return new(bytes.Buffer)
    },
}

func processData(data []byte) {
    buf := bufferPool.Get().(*bytes.Buffer)
    defer bufferPool.Put(buf)
    
    buf.Reset()  // Always reset before use!
    buf.Write(data)
    // ... process buffer ...
}

When to Use sync.Pool

Good use cases:

Temporary buffers (bytes.Buffer, strings.Builder)
Encoding/decoding scratch space
Frequently allocated, short-lived objects

Bad use cases:

Database connections
User sessions
Configuration objects
Anything that must persist

unsafe Package

Normally, Go acts like a strict parent, checking types and memory to make sure you don’t hurt yourself. unsafe is Go saying:

“Fine. You know what you’re doing. I won’t stop you.”

Using unsafe is like giving scissors to a kid. You can do useful things (like direct memory manipulation for performance), but one wrong move can silently corrupt your program in ways you won’t notice until much later.

Common Uses of unsafe

Type Conversion
Memory Layout
Zero-Copy Casting

import "unsafe"

func stringToBytes(s string) []byte {
    // ⚠️ Dangerous: No copy, but []byte should not be modified!
    return unsafe.Slice(unsafe.StringData(s), len(s))
}

type MyStruct struct {
    a int32
    b int64
}

func main() {
    var s MyStruct
    // Get memory offset of field b
    offset := unsafe.Offsetof(s.b)
    fmt.Println(offset)  // Likely 8 (due to alignment)
}

// Cast slice of one type to another (same size)
func int32ToBytes(nums []int32) []byte {
    return unsafe.Slice(
        (*byte)(unsafe.Pointer(&nums[0])),
        len(nums)*4,
    )
}

Rules for Safe Use of unsafe

Read the documentation - The unsafe package docs are critical
Understand alignment - Different architectures have different requirements
Keep it localized - Isolate unsafe code in well-tested packages
Add comments - Explain why unsafe is necessary
Benchmark first - Make sure the performance gain justifies the risk

pprof: The Truth About Performance

Imagine your program is a city, and it feels slow. You think traffic is the problem, but you’re guessing. pprof is a helicopter camera. It shows you:

Where time is actually being spent
Where memory is actually being allocated

Enabling Profiling

import (
    "os"
    "runtime/pprof"
)

func main() {
    f, _ := os.Create("cpu.prof")
    pprof.StartCPUProfile(f)
    defer pprof.StopCPUProfile()
    
    // Your code here
}

Analyzing Profiles

# CPU profile
go tool pprof cpu.prof

# Memory profile  
go tool pprof mem.prof

# Live profiling from HTTP server
go tool pprof http://localhost:6060/debug/pprof/heap

# Generate visual graph
go tool pprof -http=:8080 cpu.prof

Without pprof, performance tuning is just “vibes”. With it, you can pinpoint the exact line causing 90% of the garbage or CPU time.

The Unified Theory

Go is constantly trying to keep things simple, fast, and safe:

Escape Analysis decides where things live

Stack (fast) vs Heap (flexible but slower)

Heap Objects create work for the GC

Minimize allocations to reduce GC pressure

sync.Pool reduces garbage creation

Reuse temporary objects instead of allocating new ones

unsafe lets you break rules when you must

But only when you fully understand the consequences

pprof tells you the truth when intuition lies

Always measure before optimizing

Best Practices

Profile first, optimize second - Don’t guess at performance bottlenecks
Prefer stack allocation - Design APIs to avoid unnecessary heap escapes
Reuse buffers - Use sync.Pool for temporary objects
Minimize allocations - Preallocate slices and maps when sizes are known
Avoid unsafe until necessary - Readable, safe code is usually fast enough
Monitor GC metrics - Use GODEBUG=gctrace=1 to see GC behavior

Next Steps

Methods - Learn how to attach behavior to types
Interfaces - Master polymorphism in Go
Concurrency - Goroutines and channels with runtime awareness

Get Started

Fundamentals

Advanced Concepts

Web Development

Projects

Overview

Stack vs Heap (Escape Analysis)

Memory Locations

Escape Analysis in Action

Common Escape Scenarios

Garbage Collector (GC)

How the GC Works

The Cost of Garbage Collection

sync.Pool

sync.Pool is NOT a Cache

The Golden Rule

When to Use sync.Pool

unsafe Package

Common Uses of unsafe

Rules for Safe Use of unsafe

pprof: The Truth About Performance

Enabling Profiling

Analyzing Profiles

The Unified Theory

Best Practices

Next Steps

Get Started

Fundamentals

Advanced Concepts

Web Development

Projects

​Overview

​Stack vs Heap (Escape Analysis)

​Memory Locations

​Escape Analysis in Action

​Common Escape Scenarios

​Garbage Collector (GC)

​How the GC Works

​The Cost of Garbage Collection

​sync.Pool

​sync.Pool is NOT a Cache

​The Golden Rule

​When to Use sync.Pool

​unsafe Package

​Common Uses of unsafe

​Rules for Safe Use of unsafe

​pprof: The Truth About Performance

​Enabling Profiling

​Analyzing Profiles

​The Unified Theory

​Best Practices

​Next Steps

Overview

Stack vs Heap (Escape Analysis)

Memory Locations

Escape Analysis in Action

Common Escape Scenarios

Garbage Collector (GC)

How the GC Works

The Cost of Garbage Collection

sync.Pool

sync.Pool is NOT a Cache

The Golden Rule

When to Use sync.Pool

unsafe Package

Common Uses of unsafe

Rules for Safe Use of unsafe

pprof: The Truth About Performance

Enabling Profiling

Analyzing Profiles

The Unified Theory

Best Practices

Next Steps