What causes "fatal error: all goroutines are asleep - deadlock!" and when does Go fail to detect a deadlock?

The runtime's checkdead triggers when every goroutine is blocked on something that no other goroutine can unblock (channel ops, select{}, sync primitives), and there are no timers or I/O that could wake anything up. Classic causes:

func main() {
	ch := make(chan int)
	ch <- 1          // no receiver exists: deadlock
	fmt.Println(<-ch)
}

Others include ranging over a channel nobody closes, a WaitGroup.Wait() whose counter never reaches zero, and locking a non-reentrant mutex twice.

When it is NOT detected (the senior-level part):

• A partial deadlock, where some goroutines are stuck while others still run (an HTTP server, a ticker). The stuck ones just leak silently.
• Any goroutine blocked in a network poll or a syscall, or a pending timer, stops the check from firing.
• Programs using cgo, or goroutines in time.Sleep.

So in production services, deadlocks show up as goroutine leaks and hangs, not crashes. Diagnose them with pprof goroutine dumps (/debug/pprof/goroutine?debug=2), SIGQUIT, or go.uber.org/goleak in tests. Go 1.26 adds an experimental goroutine-leak profile built on GC reachability.