The root cause is that SIGURG (preemption signal) occupies the buf of chan whose size is only 10. I can reproduce it in the following way on my computer.

go test -c os/signal
./signal.test -test.v=1 -test.run='^TestSignal$' -test.count=20000

And add some test log into signal.go

func process(sig os.Signal) {
...
	for c, h := range handlers.m {
		if h.want(n) {
			// send but do not block for it
			select {
			case c <- sig:
				println("send ", n," ok! chan=", c)
			default:
				if n == 28 {
					println("send SIGWINCH failed! chan=", c)
				}
			}
		}
	}
...
}

=== RUN   TestSignal
    signal_test.go:136: sighup...
send  1  ok! chan= 0xc000280f60
    signal_test.go:149: sigwinch...
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send  23  ok! chan= 0xc0000945a0
send SIGWINCH failed! chan= 0xc0000945a0
...

I will try to fix this.

Change https://golang.org/cl/366514 mentions this issue: os/signal: fix TestSignal frequent timeouts