Make the docker proxy a standalone binary not a re-exec

This reduces memory usage with a lot of docker proxy processes.
On Docker for Mac we are currently carrying a patch to replace
the binary as we modify it to forward ports to the Mac rather
than the Linux VM, this allows us to simply replace this binary
in our packaging with one that has a compatible interface. This
patch does not provide an easy way to substitute a binary as
the interface is complex and there are few use cases, but where
needed this can be done.

Signed-off-by: Justin Cormack <justin.cormack@docker.com>
This commit is contained in:
Justin Cormack 2016-06-03 17:28:14 +01:00
parent 656c66dd50
commit 0682468431
14 changed files with 36 additions and 549 deletions

View file

@ -131,6 +131,9 @@ clean() {
findArgs+=( -path "vendor/src/$import" )
done
# The docker proxy command is built from libnetwork
findArgs+=( -or -path vendor/src/github.com/docker/libnetwork/cmd/proxy )
local IFS=$'\n'
local prune=( $($find vendor -depth -type d -not '(' "${findArgs[@]}" ')') )
unset IFS

View file

@ -2,3 +2,4 @@
DOCKER_CLIENT_BINARY_NAME='docker'
DOCKER_DAEMON_BINARY_NAME='dockerd'
DOCKER_PROXY_BINARY_NAME='docker-proxy'

View file

@ -22,6 +22,7 @@ override_dh_auto_install:
mkdir -p debian/docker-engine/usr/bin
cp -aT "$$(readlink -f bundles/$(VERSION)/dynbinary-client/docker)" debian/docker-engine/usr/bin/docker
cp -aT "$$(readlink -f bundles/$(VERSION)/dynbinary-daemon/dockerd)" debian/docker-engine/usr/bin/dockerd
cp -aT "$$(readlink -f bundles/$(VERSION)/dynbinary-daemon/docker-proxy)" debian/docker-engine/usr/bin/docker-proxy
cp -aT /usr/local/bin/containerd debian/docker-engine/usr/bin/docker-containerd
cp -aT /usr/local/bin/containerd-shim debian/docker-engine/usr/bin/docker-containerd-shim
cp -aT /usr/local/bin/ctr debian/docker-engine/usr/bin/docker-containerd-ctr

View file

@ -126,6 +126,7 @@ export DOCKER_GITCOMMIT=%{_gitcommit}
install -d $RPM_BUILD_ROOT/%{_bindir}
install -p -m 755 bundles/%{_origversion}/dynbinary-client/docker-%{_origversion} $RPM_BUILD_ROOT/%{_bindir}/docker
install -p -m 755 bundles/%{_origversion}/dynbinary-daemon/dockerd-%{_origversion} $RPM_BUILD_ROOT/%{_bindir}/dockerd
install -p -m 755 bundles/%{_origversion}/dynbinary-daemon/docker-proxy-%{_origversion} $RPM_BUILD_ROOT/%{_bindir}/docker-proxy
# install containerd
install -p -m 755 /usr/local/bin/containerd $RPM_BUILD_ROOT/%{_bindir}/docker-containerd

View file

@ -9,5 +9,8 @@ set -e
export BINARY_SHORT_NAME="$DOCKER_DAEMON_BINARY_NAME"
export SOURCE_PATH='./cmd/dockerd'
source "${MAKEDIR}/.binary"
export BINARY_SHORT_NAME="$DOCKER_PROXY_BINARY_NAME"
export SOURCE_PATH='./vendor/src/github.com/docker/libnetwork/cmd/proxy'
source "${MAKEDIR}/.binary"
copy_containerd "$DEST" 'hash'
)

View file

@ -9,4 +9,7 @@ set -e
export BUILDFLAGS=( "${BUILDFLAGS[@]/netgo /}" ) # disable netgo, since we don't need it for a dynamic binary
export BUILDFLAGS=( "${BUILDFLAGS[@]/static_build /}" ) # we're not building a "static" binary here
source "${MAKEDIR}/.binary"
export BINARY_SHORT_NAME='docker-proxy'
export SOURCE_PATH='./vendor/src/github.com/docker/libnetwork/cmd/proxy'
source "${MAKEDIR}/.binary"
)

View file

@ -5,6 +5,9 @@ BINARY_NAME="dockerd-$VERSION"
BINARY_EXTENSION="$(binary_extension)"
BINARY_FULLNAME="$BINARY_NAME$BINARY_EXTENSION"
PROXY_NAME="docker-proxy-$VERSION"
PROXY_FULLNAME="$PROXY_NAME$BINARY_EXTENSION"
CLIENTBIN_NAME="docker-$VERSION"
CLIENTBIN_FULLNAME="$CLIENTBIN_NAME$BINARY_EXTENSION"
@ -29,6 +32,21 @@ go build -compiler=gccgo \
echo "Created binary: $DEST/$BINARY_FULLNAME"
ln -sf "$BINARY_FULLNAME" "$DEST/dockerd$BINARY_EXTENSION"
go build -compiler=gccgo \
-o "$DEST/$PROXY_FULLNAME" \
"${BUILDFLAGS[@]}" \
-gccgoflags "
-g
$EXTLDFLAGS_STATIC
-Wl,--no-export-dynamic
-ldl
-pthread
" \
./vendor/src/github.com/docker/libnetwork/cmd/proxy
echo "Created binary: $DEST/$PROXY_FULLNAME"
ln -sf "$PROXY_FULLNAME" "$DEST/docker-proxy$BINARY_EXTENSION"
copy_containerd "$DEST" "hash"
hash_files "$DEST/$BINARY_FULLNAME"

View file

@ -7,4 +7,5 @@ rm -rf "$DEST"
DEST="$(dirname $DEST)/binary-daemon"
source "${MAKEDIR}/.binary-setup"
install_binary "${DEST}/${DOCKER_DAEMON_BINARY_NAME}"
install_binary "${DEST}/${DOCKER_PROXY_BINARY_NAME}"
)

View file

@ -18,6 +18,7 @@ for d in "$CROSS/"*/*; do
BINARY_NAME="${DOCKER_CLIENT_BINARY_NAME}-$VERSION"
DAEMON_BINARY_NAME="${DOCKER_DAEMON_BINARY_NAME}-$VERSION"
PROXY_BINARY_NAME="${DOCKER_PROXY_BINARY_NAME}-$VERSION"
BINARY_EXTENSION="$(export GOOS && binary_extension)"
if [ "$GOOS" = 'windows' ]; then
# if windows use a zip, not tgz
@ -29,6 +30,7 @@ for d in "$CROSS/"*/*; do
fi
BINARY_FULLNAME="$BINARY_NAME$BINARY_EXTENSION"
DAEMON_BINARY_FULLNAME="$DAEMON_BINARY_NAME$BINARY_EXTENSION"
PROXY_BINARY_FULLNAME="$PROXY_BINARY_NAME$BINARY_EXTENSION"
mkdir -p "$DEST/$GOOS/$GOARCH"
TGZ="$DEST/$GOOS/$GOARCH/$BINARY_NAME$BUNDLE_EXTENSION"
@ -47,6 +49,9 @@ for d in "$CROSS/"*/*; do
if [ -f "$d/$DAEMON_BINARY_FULLNAME" ]; then
cp -L "$d/$DAEMON_BINARY_FULLNAME" "$TAR_PATH/${DOCKER_DAEMON_BINARY_NAME}${BINARY_EXTENSION}"
fi
if [ -f "$d/$PROXY_BINARY_FULLNAME" ]; then
cp -L "$d/$PROXY_BINARY_FULLNAME" "$TAR_PATH/${DOCKER_PROXY_BINARY_NAME}${BINARY_EXTENSION}"
fi
# copy over all the containerd binaries
copy_containerd $TAR_PATH

View file

@ -1,216 +0,0 @@
package proxy
import (
"bytes"
"fmt"
"io"
"net"
"strings"
"testing"
"time"
)
var testBuf = []byte("Buffalo buffalo Buffalo buffalo buffalo buffalo Buffalo buffalo")
var testBufSize = len(testBuf)
type EchoServer interface {
Run()
Close()
LocalAddr() net.Addr
}
type TCPEchoServer struct {
listener net.Listener
testCtx *testing.T
}
type UDPEchoServer struct {
conn net.PacketConn
testCtx *testing.T
}
func NewEchoServer(t *testing.T, proto, address string) EchoServer {
var server EchoServer
if strings.HasPrefix(proto, "tcp") {
listener, err := net.Listen(proto, address)
if err != nil {
t.Fatal(err)
}
server = &TCPEchoServer{listener: listener, testCtx: t}
} else {
socket, err := net.ListenPacket(proto, address)
if err != nil {
t.Fatal(err)
}
server = &UDPEchoServer{conn: socket, testCtx: t}
}
return server
}
func (server *TCPEchoServer) Run() {
go func() {
for {
client, err := server.listener.Accept()
if err != nil {
return
}
go func(client net.Conn) {
if _, err := io.Copy(client, client); err != nil {
server.testCtx.Logf("can't echo to the client: %v\n", err.Error())
}
client.Close()
}(client)
}
}()
}
func (server *TCPEchoServer) LocalAddr() net.Addr { return server.listener.Addr() }
func (server *TCPEchoServer) Close() { server.listener.Close() }
func (server *UDPEchoServer) Run() {
go func() {
readBuf := make([]byte, 1024)
for {
read, from, err := server.conn.ReadFrom(readBuf)
if err != nil {
return
}
for i := 0; i != read; {
written, err := server.conn.WriteTo(readBuf[i:read], from)
if err != nil {
break
}
i += written
}
}
}()
}
func (server *UDPEchoServer) LocalAddr() net.Addr { return server.conn.LocalAddr() }
func (server *UDPEchoServer) Close() { server.conn.Close() }
func testProxyAt(t *testing.T, proto string, proxy Proxy, addr string) {
defer proxy.Close()
go proxy.Run()
client, err := net.Dial(proto, addr)
if err != nil {
t.Fatalf("Can't connect to the proxy: %v", err)
}
defer client.Close()
client.SetDeadline(time.Now().Add(10 * time.Second))
if _, err = client.Write(testBuf); err != nil {
t.Fatal(err)
}
recvBuf := make([]byte, testBufSize)
if _, err = client.Read(recvBuf); err != nil {
t.Fatal(err)
}
if !bytes.Equal(testBuf, recvBuf) {
t.Fatal(fmt.Errorf("Expected [%v] but got [%v]", testBuf, recvBuf))
}
}
func testProxy(t *testing.T, proto string, proxy Proxy) {
testProxyAt(t, proto, proxy, proxy.FrontendAddr().String())
}
func TestTCP4Proxy(t *testing.T) {
backend := NewEchoServer(t, "tcp", "127.0.0.1:0")
defer backend.Close()
backend.Run()
frontendAddr := &net.TCPAddr{IP: net.IPv4(127, 0, 0, 1), Port: 0}
proxy, err := NewProxy(frontendAddr, backend.LocalAddr())
if err != nil {
t.Fatal(err)
}
testProxy(t, "tcp", proxy)
}
func TestTCP6Proxy(t *testing.T) {
backend := NewEchoServer(t, "tcp", "[::1]:0")
defer backend.Close()
backend.Run()
frontendAddr := &net.TCPAddr{IP: net.IPv6loopback, Port: 0}
proxy, err := NewProxy(frontendAddr, backend.LocalAddr())
if err != nil {
t.Fatal(err)
}
testProxy(t, "tcp", proxy)
}
func TestTCPDualStackProxy(t *testing.T) {
// If I understand `godoc -src net favoriteAddrFamily` (used by the
// net.Listen* functions) correctly this should work, but it doesn't.
t.Skip("No support for dual stack yet")
backend := NewEchoServer(t, "tcp", "[::1]:0")
defer backend.Close()
backend.Run()
frontendAddr := &net.TCPAddr{IP: net.IPv6loopback, Port: 0}
proxy, err := NewProxy(frontendAddr, backend.LocalAddr())
if err != nil {
t.Fatal(err)
}
ipv4ProxyAddr := &net.TCPAddr{
IP: net.IPv4(127, 0, 0, 1),
Port: proxy.FrontendAddr().(*net.TCPAddr).Port,
}
testProxyAt(t, "tcp", proxy, ipv4ProxyAddr.String())
}
func TestUDP4Proxy(t *testing.T) {
backend := NewEchoServer(t, "udp", "127.0.0.1:0")
defer backend.Close()
backend.Run()
frontendAddr := &net.UDPAddr{IP: net.IPv4(127, 0, 0, 1), Port: 0}
proxy, err := NewProxy(frontendAddr, backend.LocalAddr())
if err != nil {
t.Fatal(err)
}
testProxy(t, "udp", proxy)
}
func TestUDP6Proxy(t *testing.T) {
backend := NewEchoServer(t, "udp", "[::1]:0")
defer backend.Close()
backend.Run()
frontendAddr := &net.UDPAddr{IP: net.IPv6loopback, Port: 0}
proxy, err := NewProxy(frontendAddr, backend.LocalAddr())
if err != nil {
t.Fatal(err)
}
testProxy(t, "udp", proxy)
}
func TestUDPWriteError(t *testing.T) {
frontendAddr := &net.UDPAddr{IP: net.IPv4(127, 0, 0, 1), Port: 0}
// Hopefully, this port will be free: */
backendAddr := &net.UDPAddr{IP: net.IPv4(127, 0, 0, 1), Port: 25587}
proxy, err := NewProxy(frontendAddr, backendAddr)
if err != nil {
t.Fatal(err)
}
defer proxy.Close()
go proxy.Run()
client, err := net.Dial("udp", "127.0.0.1:25587")
if err != nil {
t.Fatalf("Can't connect to the proxy: %v", err)
}
defer client.Close()
// Make sure the proxy doesn't stop when there is no actual backend:
client.Write(testBuf)
client.Write(testBuf)
backend := NewEchoServer(t, "udp", "127.0.0.1:25587")
defer backend.Close()
backend.Run()
client.SetDeadline(time.Now().Add(10 * time.Second))
if _, err = client.Write(testBuf); err != nil {
t.Fatal(err)
}
recvBuf := make([]byte, testBufSize)
if _, err = client.Read(recvBuf); err != nil {
t.Fatal(err)
}
if !bytes.Equal(testBuf, recvBuf) {
t.Fatal(fmt.Errorf("Expected [%v] but got [%v]", testBuf, recvBuf))
}
}

View file

@ -1,37 +0,0 @@
// Package proxy provides a network Proxy interface and implementations for TCP
// and UDP.
package proxy
import (
"fmt"
"net"
)
// Proxy defines the behavior of a proxy. It forwards traffic back and forth
// between two endpoints : the frontend and the backend.
// It can be used to do software port-mapping between two addresses.
// e.g. forward all traffic between the frontend (host) 127.0.0.1:3000
// to the backend (container) at 172.17.42.108:4000.
type Proxy interface {
// Run starts forwarding traffic back and forth between the front
// and back-end addresses.
Run()
// Close stops forwarding traffic and close both ends of the Proxy.
Close()
// FrontendAddr returns the address on which the proxy is listening.
FrontendAddr() net.Addr
// BackendAddr returns the proxied address.
BackendAddr() net.Addr
}
// NewProxy creates a Proxy according to the specified frontendAddr and backendAddr.
func NewProxy(frontendAddr, backendAddr net.Addr) (Proxy, error) {
switch frontendAddr.(type) {
case *net.UDPAddr:
return NewUDPProxy(frontendAddr.(*net.UDPAddr), backendAddr.(*net.UDPAddr))
case *net.TCPAddr:
return NewTCPProxy(frontendAddr.(*net.TCPAddr), backendAddr.(*net.TCPAddr))
default:
panic(fmt.Errorf("Unsupported protocol"))
}
}

View file

@ -1,31 +0,0 @@
package proxy
import (
"net"
)
// StubProxy is a proxy that is a stub (does nothing).
type StubProxy struct {
frontendAddr net.Addr
backendAddr net.Addr
}
// Run does nothing.
func (p *StubProxy) Run() {}
// Close does nothing.
func (p *StubProxy) Close() {}
// FrontendAddr returns the frontend address.
func (p *StubProxy) FrontendAddr() net.Addr { return p.frontendAddr }
// BackendAddr returns the backend address.
func (p *StubProxy) BackendAddr() net.Addr { return p.backendAddr }
// NewStubProxy creates a new StubProxy
func NewStubProxy(frontendAddr, backendAddr net.Addr) (Proxy, error) {
return &StubProxy{
frontendAddr: frontendAddr,
backendAddr: backendAddr,
}, nil
}

View file

@ -1,96 +0,0 @@
package proxy
import (
"io"
"net"
"sync"
"syscall"
"github.com/Sirupsen/logrus"
)
// TCPProxy is a proxy for TCP connections. It implements the Proxy interface to
// handle TCP traffic forwarding between the frontend and backend addresses.
type TCPProxy struct {
listener *net.TCPListener
frontendAddr *net.TCPAddr
backendAddr *net.TCPAddr
}
// NewTCPProxy creates a new TCPProxy.
func NewTCPProxy(frontendAddr, backendAddr *net.TCPAddr) (*TCPProxy, error) {
listener, err := net.ListenTCP("tcp", frontendAddr)
if err != nil {
return nil, err
}
// If the port in frontendAddr was 0 then ListenTCP will have a picked
// a port to listen on, hence the call to Addr to get that actual port:
return &TCPProxy{
listener: listener,
frontendAddr: listener.Addr().(*net.TCPAddr),
backendAddr: backendAddr,
}, nil
}
func (proxy *TCPProxy) clientLoop(client *net.TCPConn, quit chan bool) {
backend, err := net.DialTCP("tcp", nil, proxy.backendAddr)
if err != nil {
logrus.Printf("Can't forward traffic to backend tcp/%v: %s\n", proxy.backendAddr, err)
client.Close()
return
}
var wg sync.WaitGroup
var broker = func(to, from *net.TCPConn) {
if _, err := io.Copy(to, from); err != nil {
// If the socket we are writing to is shutdown with
// SHUT_WR, forward it to the other end of the pipe:
if err, ok := err.(*net.OpError); ok && err.Err == syscall.EPIPE {
from.CloseWrite()
}
}
to.CloseRead()
wg.Done()
}
wg.Add(2)
go broker(client, backend)
go broker(backend, client)
finish := make(chan struct{})
go func() {
wg.Wait()
close(finish)
}()
select {
case <-quit:
case <-finish:
}
client.Close()
backend.Close()
<-finish
}
// Run starts forwarding the traffic using TCP.
func (proxy *TCPProxy) Run() {
quit := make(chan bool)
defer close(quit)
for {
client, err := proxy.listener.Accept()
if err != nil {
logrus.Printf("Stopping proxy on tcp/%v for tcp/%v (%s)", proxy.frontendAddr, proxy.backendAddr, err)
return
}
go proxy.clientLoop(client.(*net.TCPConn), quit)
}
}
// Close stops forwarding the traffic.
func (proxy *TCPProxy) Close() { proxy.listener.Close() }
// FrontendAddr returns the TCP address on which the proxy is listening.
func (proxy *TCPProxy) FrontendAddr() net.Addr { return proxy.frontendAddr }
// BackendAddr returns the TCP proxied address.
func (proxy *TCPProxy) BackendAddr() net.Addr { return proxy.backendAddr }

View file

@ -1,169 +0,0 @@
package proxy
import (
"encoding/binary"
"net"
"strings"
"sync"
"syscall"
"time"
"github.com/Sirupsen/logrus"
)
const (
// UDPConnTrackTimeout is the timeout used for UDP connection tracking
UDPConnTrackTimeout = 90 * time.Second
// UDPBufSize is the buffer size for the UDP proxy
UDPBufSize = 65507
)
// A net.Addr where the IP is split into two fields so you can use it as a key
// in a map:
type connTrackKey struct {
IPHigh uint64
IPLow uint64
Port int
}
func newConnTrackKey(addr *net.UDPAddr) *connTrackKey {
if len(addr.IP) == net.IPv4len {
return &connTrackKey{
IPHigh: 0,
IPLow: uint64(binary.BigEndian.Uint32(addr.IP)),
Port: addr.Port,
}
}
return &connTrackKey{
IPHigh: binary.BigEndian.Uint64(addr.IP[:8]),
IPLow: binary.BigEndian.Uint64(addr.IP[8:]),
Port: addr.Port,
}
}
type connTrackMap map[connTrackKey]*net.UDPConn
// UDPProxy is proxy for which handles UDP datagrams. It implements the Proxy
// interface to handle UDP traffic forwarding between the frontend and backend
// addresses.
type UDPProxy struct {
listener *net.UDPConn
frontendAddr *net.UDPAddr
backendAddr *net.UDPAddr
connTrackTable connTrackMap
connTrackLock sync.Mutex
}
// NewUDPProxy creates a new UDPProxy.
func NewUDPProxy(frontendAddr, backendAddr *net.UDPAddr) (*UDPProxy, error) {
listener, err := net.ListenUDP("udp", frontendAddr)
if err != nil {
return nil, err
}
return &UDPProxy{
listener: listener,
frontendAddr: listener.LocalAddr().(*net.UDPAddr),
backendAddr: backendAddr,
connTrackTable: make(connTrackMap),
}, nil
}
func (proxy *UDPProxy) replyLoop(proxyConn *net.UDPConn, clientAddr *net.UDPAddr, clientKey *connTrackKey) {
defer func() {
proxy.connTrackLock.Lock()
delete(proxy.connTrackTable, *clientKey)
proxy.connTrackLock.Unlock()
proxyConn.Close()
}()
readBuf := make([]byte, UDPBufSize)
for {
proxyConn.SetReadDeadline(time.Now().Add(UDPConnTrackTimeout))
again:
read, err := proxyConn.Read(readBuf)
if err != nil {
if err, ok := err.(*net.OpError); ok && err.Err == syscall.ECONNREFUSED {
// This will happen if the last write failed
// (e.g: nothing is actually listening on the
// proxied port on the container), ignore it
// and continue until UDPConnTrackTimeout
// expires:
goto again
}
return
}
for i := 0; i != read; {
written, err := proxy.listener.WriteToUDP(readBuf[i:read], clientAddr)
if err != nil {
return
}
i += written
}
}
}
// Run starts forwarding the traffic using UDP.
func (proxy *UDPProxy) Run() {
readBuf := make([]byte, UDPBufSize)
for {
read, from, err := proxy.listener.ReadFromUDP(readBuf)
if err != nil {
// NOTE: Apparently ReadFrom doesn't return
// ECONNREFUSED like Read do (see comment in
// UDPProxy.replyLoop)
if !isClosedError(err) {
logrus.Printf("Stopping proxy on udp/%v for udp/%v (%s)", proxy.frontendAddr, proxy.backendAddr, err)
}
break
}
fromKey := newConnTrackKey(from)
proxy.connTrackLock.Lock()
proxyConn, hit := proxy.connTrackTable[*fromKey]
if !hit {
proxyConn, err = net.DialUDP("udp", nil, proxy.backendAddr)
if err != nil {
logrus.Printf("Can't proxy a datagram to udp/%s: %s\n", proxy.backendAddr, err)
proxy.connTrackLock.Unlock()
continue
}
proxy.connTrackTable[*fromKey] = proxyConn
go proxy.replyLoop(proxyConn, from, fromKey)
}
proxy.connTrackLock.Unlock()
for i := 0; i != read; {
written, err := proxyConn.Write(readBuf[i:read])
if err != nil {
logrus.Printf("Can't proxy a datagram to udp/%s: %s\n", proxy.backendAddr, err)
break
}
i += written
}
}
}
// Close stops forwarding the traffic.
func (proxy *UDPProxy) Close() {
proxy.listener.Close()
proxy.connTrackLock.Lock()
defer proxy.connTrackLock.Unlock()
for _, conn := range proxy.connTrackTable {
conn.Close()
}
}
// FrontendAddr returns the UDP address on which the proxy is listening.
func (proxy *UDPProxy) FrontendAddr() net.Addr { return proxy.frontendAddr }
// BackendAddr returns the proxied UDP address.
func (proxy *UDPProxy) BackendAddr() net.Addr { return proxy.backendAddr }
func isClosedError(err error) bool {
/* This comparison is ugly, but unfortunately, net.go doesn't export errClosing.
* See:
* http://golang.org/src/pkg/net/net.go
* https://code.google.com/p/go/issues/detail?id=4337
* https://groups.google.com/forum/#!msg/golang-nuts/0_aaCvBmOcM/SptmDyX1XJMJ
*/
return strings.HasSuffix(err.Error(), "use of closed network connection")
}