Summary - Runtime

Combine GOMAXPROCS, Gosched and Stack to write a monitor.

Runtime summary

In this group you learned to talk to the runtime: observing with NumCPU, NumGoroutine and runtime/metrics; tuning parallelism with GOMAXPROCS; yielding with Gosched; diagnosing with Stack and Caller; ending goroutines with Goexit.

GOTRACEBACK=all ./app   // all goroutines on a crash
GODEBUG=gctrace=1,schedtrace=1000 ./app

debug.SetTraceback and limits

The runtime/debug package offers extra knobs: SetTraceback("all") widens crash output, SetMaxStack sets a single goroutine's stack limit (default 1 GB on 64-bit), SetMaxThreads changes the M limit (10,000), and FreeOSMemory forces returning free memory to the operating system.

Exercise Leak detector

Take once := runtime.NumGoroutine(). Start 5 goroutines that all wait on a dur channel. Call close(dur), wait with time.Sleep(20 * time.Millisecond) and print whether the count returned to the old value (runtime.NumGoroutine() == once).

Show example solution

package main

import (
    "fmt"
    "runtime"
    "time"
)

func main() {
    once := runtime.NumGoroutine()
    dur := make(chan struct{})
    for i := 0; i < 5; i++ {
        go func() { <-dur }()
    }
    close(dur)
    time.Sleep(20 * time.Millisecond)
    fmt.Println(runtime.NumGoroutine() == once)
}

Test yourself

If you suspect a goroutine leak in a service, what do you watch first?
- Whether the goroutine count grows over time
- The NumCPU value
- The GOOS value
- The build time
Show answer

Whether the goroutine count grows over time — A steadily rising NumGoroutine (or the /sched/goroutines metric) signals a leak; the goroutine profile shows where they are stuck.
Crash detail
```
____=all ./app
```
Show answer
```
GOTRACEBACK=all ./app
```
Group summary

NumGoroutine

Leak monitoring
GOMAXPROCS

Parallelism
runtime.Caller

Line information
Goexit

Exit with defers

Next lesson: WebAssembly

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies