wgengine/magicsock: increase legacy ping timeout again

I based my estimation of the required timeout based on locally
observed behavior. But CI machines are worse than my local machine.
16s was enough to reduce flakiness but not eliminate it. Bump it up again.

Signed-off-by: Josh Bleecher Snyder <josh@tailscale.com>

util/cibuild/cibuild.go

@@ -0,0 +1,13 @@
+// Copyright (c) 2021 Tailscale Inc & AUTHORS All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package cibuild reports runtime CI information.
+package cibuild
+
+import "os"
+
+// On reports whether the current binary is executing on a CI system.
+func On() bool {
+	return os.Getenv("GITHUB_ACTIONS") != ""
+}

wgengine/magicsock/magicsock_test.go

@@ -44,6 +44,7 @@ import (
 	"tailscale.com/types/netmap"
 	"tailscale.com/types/nettype"
 	"tailscale.com/types/wgkey"
+	"tailscale.com/util/cibuild"
 	"tailscale.com/wgengine/filter"
 	"tailscale.com/wgengine/tstun"
 	"tailscale.com/wgengine/wgcfg"
@@ -929,10 +930,14 @@ func testTwoDevicePing(t *testing.T, d *devices) {
 
 	// In the normal case, pings succeed immediately.
 	// However, in the case of a handshake race, we need to retry.
-	// Typical retries take 5s. With very bad luck, we can need to retry
+	// With very bad luck, we can need to retry multiple times.
-	// multiple times. Give ourselves enough time for three retries
+	allowedRetries := 3
-	// plus a bit of processing time.
+	if cibuild.On() {
-	const pingTimeout = 16 * time.Second
+		// Allow extra retries on small/flaky/loaded CI machines.
+		allowedRetries *= 2
+	}
+	// Retries take 5s each. Add 1s for some processing time.
+	pingTimeout := 5*time.Second*time.Duration(allowedRetries) + time.Second
 
 	ping1 := func(t *testing.T) {
 		msg2to1 := tuntest.Ping(net.ParseIP("1.0.0.1"), net.ParseIP("1.0.0.2"))