Socket bind helper package

Package bind provides a convenient syntax for binding to sockets, as well as a
fair bit of magic to automatically Do The Right Thing in both development and
production.

bind/bind.go

@@ -0,0 +1,115 @@
+/*
+Package bind provides a convenient way to bind to sockets. It exposes a flag in
+the default flag set named "bind" which provides syntax to bind TCP and UNIX
+sockets. It also supports binding to arbitrary file descriptors passed by a
+parent (for instance, systemd), and for binding to Einhorn sockets (including
+Einhorn ACK support).
+
+If the value passed to bind contains a colon, as in ":8000" or "127.0.0.1:9001",
+it will be treated as a TCP address. If it begins with a "/" or a ".", it will
+be treated as a path to a UNIX socket. If it begins with the string "fd@", as in
+"fd@3", it will be treated as a file descriptor (useful for use with systemd,
+for instance). If it begins with the string "einhorn@", as in "einhorn@0", the
+corresponding einhorn socket will be used.
+
+If an option is not explicitly passed, the implementation will automatically
+select between using "einhorn@0", "fd@3", and ":8000", depending on whether
+Einhorn or systemd (or neither) is detected.
+
+This package is a teensy bit magical, and goes out of its way to Do The Right
+Thing in many situations, including in both development and production. If
+you're looking for something less magical, you'd probably be better off just
+calling net.Listen() the old-fashioned way.
+*/
+package bind
+
+import (
+	"flag"
+	"fmt"
+	"log"
+	"net"
+	"os"
+	"strconv"
+	"strings"
+	"sync"
+)
+
+var bind string
+
+func init() {
+	einhornInit()
+	systemdInit()
+
+	defaultBind := ":8000"
+	if usingEinhorn() {
+		defaultBind = "einhorn@0"
+	} else if usingSystemd() {
+		defaultBind = "fd@3"
+	}
+	flag.StringVar(&bind, "bind", defaultBind,
+		`Address to bind on. If this value has a colon, as in ":8000" or
+		"127.0.0.1:9001", it will be treated as a TCP address. If it
+		begins with a "/" or a ".", it will be treated as a path to a
+		UNIX socket. If it begins with the string "fd@", as in "fd@3",
+		it will be treated as a file descriptor (useful for use with
+		systemd, for instance). If it begins with the string "einhorn@",
+		as in "einhorn@0", the corresponding einhorn socket will be
+		used. If an option is not explicitly passed, the implementation
+		will automatically select among "einhorn@0" (Einhorn), "fd@3"
+		(systemd), and ":8000" (fallback) based on its environment.`)
+}
+
+func listenTo(bind string) (net.Listener, error) {
+	if strings.Contains(bind, ":") {
+		return net.Listen("tcp", bind)
+	} else if strings.HasPrefix(bind, ".") || strings.HasPrefix(bind, "/") {
+		return net.Listen("unix", bind)
+	} else if strings.HasPrefix(bind, "fd@") {
+		fd, err := strconv.Atoi(bind[3:])
+		if err != nil {
+			return nil, fmt.Errorf("Error while parsing fd %v: %v",
+				bind, err)
+		}
+		f := os.NewFile(uintptr(fd), bind)
+		return net.FileListener(f)
+	} else if strings.HasPrefix(bind, "einhorn@") {
+		fd, err := strconv.Atoi(bind[8:])
+		if err != nil {
+			return nil, fmt.Errorf(
+				"Error while parsing einhorn %v: %v", bind, err)
+		}
+		return einhornBind(fd)
+	}
+
+	return nil, fmt.Errorf("Error while parsing bind arg %v", bind)
+}
+
+// Parse and bind to the specified address. If Socket encounters an error while
+// parsing or binding to the given socket it will exit by calling log.Fatal.
+func Socket(bind string) net.Listener {
+	l, err := listenTo(bind)
+	if err != nil {
+		log.Fatal(err)
+	}
+	return l
+}
+
+// Parse and bind to the default socket as given to us by the flag module. If
+// there was an error parsing or binding to that socket, Default will exit by
+// calling `log.Fatal`.
+func Default() net.Listener {
+	return Socket(bind)
+}
+
+// I'm not sure why you'd ever want to call Ready() more than once, but we may
+// as well be safe against it...
+var ready sync.Once
+
+// Notify the environment (for now, just Einhorn) that the process is ready to
+// receive traffic. Should be called at the last possible moment to maximize the
+// chances that a faulty process exits before signaling that it's ready.
+func Ready() {
+	ready.Do(func() {
+		einhornAck()
+	})
+}

bind/einhorn.go

@@ -0,0 +1,88 @@
+package bind
+
+import (
+	"fmt"
+	"log"
+	"net"
+	"os"
+	"strconv"
+	"syscall"
+)
+
+const tooBigErr = "bind: einhorn@%d not found (einhorn only passed %d fds)"
+const bindErr = "bind: could not bind einhorn@%d: not running under einhorn"
+const einhornErr = "bind: einhorn environment initialization error"
+const ackErr = "bind: error ACKing to einhorn: %v"
+
+var einhornNumFds int
+
+func envInt(val string) (int, error) {
+	return strconv.Atoi(os.Getenv(val))
+}
+
+// Unfortunately this can't be a normal init function, because their execution
+// order is undefined, and we need to run before the init() in bind.go.
+func einhornInit() {
+	mpid, err := envInt("EINHORN_MASTER_PID")
+	if err != nil || mpid != os.Getppid() {
+		return
+	}
+
+	einhornNumFds, err = envInt("EINHORN_FD_COUNT")
+	if err != nil {
+		einhornNumFds = 0
+		return
+	}
+
+	// Prevent einhorn's fds from leaking to our children
+	for i := 0; i < einhornNumFds; i++ {
+		fd := int(einhornFd(i).Fd())
+		syscall.CloseOnExec(fd)
+	}
+}
+
+func usingEinhorn() bool {
+	return einhornNumFds > 0
+}
+
+func einhornFd(n int) *os.File {
+	name := fmt.Sprintf("EINHORN_FD_%d", n)
+	fno, err := envInt(name)
+	if err != nil {
+		log.Fatal(einhornErr)
+	}
+	return os.NewFile(uintptr(fno), name)
+}
+
+func einhornBind(n int) (net.Listener, error) {
+	if !usingEinhorn() {
+		return nil, fmt.Errorf(bindErr, n)
+	}
+	if n >= einhornNumFds || n < 0 {
+		return nil, fmt.Errorf(tooBigErr, n, einhornNumFds)
+	}
+
+	f := einhornFd(n)
+	return net.FileListener(f)
+}
+
+// Fun story: this is actually YAML, not JSON.
+const ackMsg = `{"command":"worker:ack","pid":%d}` + "\n"
+
+func einhornAck() {
+	if !usingEinhorn() {
+		return
+	}
+	log.Print("bind: ACKing to einhorn")
+
+	ctl, err := net.Dial("unix", os.Getenv("EINHORN_SOCK_PATH"))
+	if err != nil {
+		log.Fatalf(ackErr, err)
+	}
+	defer ctl.Close()
+
+	_, err = fmt.Fprintf(ctl, ackMsg, os.Getpid())
+	if err != nil {
+		log.Fatalf(ackErr, err)
+	}
+}

bind/systemd.go

@@ -0,0 +1,34 @@
+package bind
+
+import (
+	"os"
+	"syscall"
+)
+
+const systemdMinFd = 3
+
+var systemdNumFds int
+
+// Unfortunately this can't be a normal init function, because their execution
+// order is undefined, and we need to run before the init() in bind.go.
+func systemdInit() {
+	pid, err := envInt("LISTEN_PID")
+	if err != nil || pid != os.Getpid() {
+		return
+	}
+
+	systemdNumFds, err = envInt("LISTEN_FDS")
+	if err != nil {
+		systemdNumFds = 0
+		return
+	}
+
+	// Prevent fds from leaking to our children
+	for i := 0; i < systemdNumFds; i++ {
+		syscall.CloseOnExec(systemdMinFd + i)
+	}
+}
+
+func usingSystemd() bool {
+	return systemdNumFds > 0
+}