control/controlclient: introduce eventbus messages instead of callbacks (#16956)

This is a small introduction of the eventbus into controlclient that
communicates with mainly ipnlocal. While ipnlocal is a complicated part
of the codebase, the subscribers here are from the perspective of
ipnlocal already called async.

Updates #15160

Signed-off-by: Claus Lensbøl <claus@tailscale.com>
This commit is contained in:
Claus Lensbøl
2025-09-15 10:36:17 -04:00
committed by GitHub
parent 782c16c513
commit b816fd7117
15 changed files with 293 additions and 143 deletions

View File

@@ -205,7 +205,6 @@ func NewNoStart(opts Options) (_ *Auto, err error) {
}
})
return c, nil
}
// SetPaused controls whether HTTP activity should be paused.
@@ -424,6 +423,11 @@ func (c *Auto) unpausedChanLocked() <-chan bool {
return unpaused
}
// ClientID returns the ClientID of the direct controlClient
func (c *Auto) ClientID() int64 {
return c.direct.ClientID()
}
// mapRoutineState is the state of Auto.mapRoutine while it's running.
type mapRoutineState struct {
c *Auto

View File

@@ -81,6 +81,9 @@ type Client interface {
// in a separate http request. It has nothing to do with the rest of
// the state machine.
UpdateEndpoints(endpoints []tailcfg.Endpoint)
// ClientID returns the ClientID of a client. This ID is meant to
// distinguish one client from another.
ClientID() int64
}
// UserVisibleError is an error that should be shown to users.

View File

@@ -35,6 +35,7 @@ import (
"tailscale.com/types/logger"
"tailscale.com/types/netmap"
"tailscale.com/types/persist"
"tailscale.com/util/eventbus/eventbustest"
)
func fieldsOf(t reflect.Type) (fields []string) {
@@ -218,6 +219,8 @@ func TestDirectProxyManual(t *testing.T) {
t.Skip("skipping without --live-network-test")
}
bus := eventbustest.NewBus(t)
dialer := &tsdial.Dialer{}
dialer.SetNetMon(netmon.NewStatic())
@@ -239,6 +242,7 @@ func TestDirectProxyManual(t *testing.T) {
},
Dialer: dialer,
ControlKnobs: &controlknobs.Knobs{},
Bus: bus,
}
d, err := NewDirect(opts)
if err != nil {
@@ -263,6 +267,8 @@ func TestHTTPSWithProxy(t *testing.T) { testHTTPS(t, true) }
func testHTTPS(t *testing.T, withProxy bool) {
bakedroots.ResetForTest(t, tlstest.TestRootCA())
bus := eventbustest.NewBus(t)
controlLn, err := tls.Listen("tcp", "127.0.0.1:0", tlstest.ControlPlane.ServerTLSConfig())
if err != nil {
t.Fatal(err)
@@ -327,6 +333,7 @@ func testHTTPS(t *testing.T, withProxy bool) {
t.Logf("PopBrowserURL: %q", url)
},
Dialer: dialer,
Bus: bus,
}
d, err := NewDirect(opts)
if err != nil {

View File

@@ -14,6 +14,7 @@ import (
"fmt"
"io"
"log"
"math/rand/v2"
"net"
"net/http"
"net/netip"
@@ -52,6 +53,7 @@ import (
"tailscale.com/types/ptr"
"tailscale.com/types/tkatype"
"tailscale.com/util/clientmetric"
"tailscale.com/util/eventbus"
"tailscale.com/util/multierr"
"tailscale.com/util/singleflight"
"tailscale.com/util/syspolicy/pkey"
@@ -63,30 +65,31 @@ import (
// Direct is the client that connects to a tailcontrol server for a node.
type Direct struct {
httpc *http.Client // HTTP client used to talk to tailcontrol
interceptedDial *atomic.Bool // if non-nil, pointer to bool whether ScreenTime intercepted our dial
dialer *tsdial.Dialer
dnsCache *dnscache.Resolver
controlKnobs *controlknobs.Knobs // always non-nil
serverURL string // URL of the tailcontrol server
clock tstime.Clock
logf logger.Logf
netMon *netmon.Monitor // non-nil
health *health.Tracker
discoPubKey key.DiscoPublic
getMachinePrivKey func() (key.MachinePrivate, error)
debugFlags []string
skipIPForwardingCheck bool
pinger Pinger
polc policyclient.Client // always non-nil
popBrowser func(url string) // or nil
c2nHandler http.Handler // or nil
onClientVersion func(*tailcfg.ClientVersion) // or nil
onControlTime func(time.Time) // or nil
onTailnetDefaultAutoUpdate func(bool) // or nil
panicOnUse bool // if true, panic if client is used (for testing)
closedCtx context.Context // alive until Direct.Close is called
closeCtx context.CancelFunc // cancels closedCtx
httpc *http.Client // HTTP client used to talk to tailcontrol
interceptedDial *atomic.Bool // if non-nil, pointer to bool whether ScreenTime intercepted our dial
dialer *tsdial.Dialer
dnsCache *dnscache.Resolver
controlKnobs *controlknobs.Knobs // always non-nil
serverURL string // URL of the tailcontrol server
clock tstime.Clock
logf logger.Logf
netMon *netmon.Monitor // non-nil
health *health.Tracker
discoPubKey key.DiscoPublic
busClient *eventbus.Client
clientVersionPub *eventbus.Publisher[tailcfg.ClientVersion]
autoUpdatePub *eventbus.Publisher[AutoUpdate]
controlTimePub *eventbus.Publisher[ControlTime]
getMachinePrivKey func() (key.MachinePrivate, error)
debugFlags []string
skipIPForwardingCheck bool
pinger Pinger
popBrowser func(url string) // or nil
polc policyclient.Client // always non-nil
c2nHandler http.Handler // or nil
panicOnUse bool // if true, panic if client is used (for testing)
closedCtx context.Context // alive until Direct.Close is called
closeCtx context.CancelFunc // cancels closedCtx
dialPlan ControlDialPlanner // can be nil
@@ -107,6 +110,8 @@ type Direct struct {
tkaHead string
lastPingURL string // last PingRequest.URL received, for dup suppression
connectionHandleForTest string // sent in MapRequest.ConnectionHandleForTest
controlClientID int64 // Random ID used to differentiate clients for consumers of messages.
}
// Observer is implemented by users of the control client (such as LocalBackend)
@@ -120,26 +125,24 @@ type Observer interface {
}
type Options struct {
Persist persist.Persist // initial persistent data
GetMachinePrivateKey func() (key.MachinePrivate, error) // returns the machine key to use
ServerURL string // URL of the tailcontrol server
AuthKey string // optional node auth key for auto registration
Clock tstime.Clock
Hostinfo *tailcfg.Hostinfo // non-nil passes ownership, nil means to use default using os.Hostname, etc
DiscoPublicKey key.DiscoPublic
PolicyClient policyclient.Client // or nil for none
Logf logger.Logf
HTTPTestClient *http.Client // optional HTTP client to use (for tests only)
NoiseTestClient *http.Client // optional HTTP client to use for noise RPCs (tests only)
DebugFlags []string // debug settings to send to control
HealthTracker *health.Tracker
PopBrowserURL func(url string) // optional func to open browser
OnClientVersion func(*tailcfg.ClientVersion) // optional func to inform GUI of client version status
OnControlTime func(time.Time) // optional func to notify callers of new time from control
OnTailnetDefaultAutoUpdate func(bool) // optional func to inform GUI of default auto-update setting for the tailnet
Dialer *tsdial.Dialer // non-nil
C2NHandler http.Handler // or nil
ControlKnobs *controlknobs.Knobs // or nil to ignore
Persist persist.Persist // initial persistent data
GetMachinePrivateKey func() (key.MachinePrivate, error) // returns the machine key to use
ServerURL string // URL of the tailcontrol server
AuthKey string // optional node auth key for auto registration
Clock tstime.Clock
Hostinfo *tailcfg.Hostinfo // non-nil passes ownership, nil means to use default using os.Hostname, etc
DiscoPublicKey key.DiscoPublic
PolicyClient policyclient.Client // or nil for none
Logf logger.Logf
HTTPTestClient *http.Client // optional HTTP client to use (for tests only)
NoiseTestClient *http.Client // optional HTTP client to use for noise RPCs (tests only)
DebugFlags []string // debug settings to send to control
HealthTracker *health.Tracker
PopBrowserURL func(url string) // optional func to open browser
Dialer *tsdial.Dialer // non-nil
C2NHandler http.Handler // or nil
ControlKnobs *controlknobs.Knobs // or nil to ignore
Bus *eventbus.Bus
// Observer is called when there's a change in status to report
// from the control client.
@@ -287,33 +290,32 @@ func NewDirect(opts Options) (*Direct, error) {
}
c := &Direct{
httpc: httpc,
interceptedDial: interceptedDial,
controlKnobs: opts.ControlKnobs,
getMachinePrivKey: opts.GetMachinePrivateKey,
serverURL: opts.ServerURL,
clock: opts.Clock,
logf: opts.Logf,
persist: opts.Persist.View(),
authKey: opts.AuthKey,
discoPubKey: opts.DiscoPublicKey,
debugFlags: opts.DebugFlags,
netMon: netMon,
health: opts.HealthTracker,
skipIPForwardingCheck: opts.SkipIPForwardingCheck,
pinger: opts.Pinger,
polc: cmp.Or(opts.PolicyClient, policyclient.Client(policyclient.NoPolicyClient{})),
popBrowser: opts.PopBrowserURL,
onClientVersion: opts.OnClientVersion,
onTailnetDefaultAutoUpdate: opts.OnTailnetDefaultAutoUpdate,
onControlTime: opts.OnControlTime,
c2nHandler: opts.C2NHandler,
dialer: opts.Dialer,
dnsCache: dnsCache,
dialPlan: opts.DialPlan,
httpc: httpc,
interceptedDial: interceptedDial,
controlKnobs: opts.ControlKnobs,
getMachinePrivKey: opts.GetMachinePrivateKey,
serverURL: opts.ServerURL,
clock: opts.Clock,
logf: opts.Logf,
persist: opts.Persist.View(),
authKey: opts.AuthKey,
discoPubKey: opts.DiscoPublicKey,
debugFlags: opts.DebugFlags,
netMon: netMon,
health: opts.HealthTracker,
skipIPForwardingCheck: opts.SkipIPForwardingCheck,
pinger: opts.Pinger,
polc: cmp.Or(opts.PolicyClient, policyclient.Client(policyclient.NoPolicyClient{})),
popBrowser: opts.PopBrowserURL,
c2nHandler: opts.C2NHandler,
dialer: opts.Dialer,
dnsCache: dnsCache,
dialPlan: opts.DialPlan,
}
c.closedCtx, c.closeCtx = context.WithCancel(context.Background())
c.controlClientID = rand.Int64()
if opts.Hostinfo == nil {
c.SetHostinfo(hostinfo.New())
} else {
@@ -331,6 +333,12 @@ func NewDirect(opts Options) (*Direct, error) {
if strings.Contains(opts.ServerURL, "controlplane.tailscale.com") && envknob.Bool("TS_PANIC_IF_HIT_MAIN_CONTROL") {
c.panicOnUse = true
}
c.busClient = opts.Bus.Client("controlClient.direct")
c.clientVersionPub = eventbus.Publish[tailcfg.ClientVersion](c.busClient)
c.autoUpdatePub = eventbus.Publish[AutoUpdate](c.busClient)
c.controlTimePub = eventbus.Publish[ControlTime](c.busClient)
return c, nil
}
@@ -340,6 +348,7 @@ func (c *Direct) Close() error {
c.mu.Lock()
defer c.mu.Unlock()
c.busClient.Close()
if c.noiseClient != nil {
if err := c.noiseClient.Close(); err != nil {
return err
@@ -826,6 +835,23 @@ func (c *Direct) SendUpdate(ctx context.Context) error {
return c.sendMapRequest(ctx, false, nil)
}
// ClientID returns the ControlClientID of the controlClient
func (c *Direct) ClientID() int64 {
return c.controlClientID
}
// AutoUpdate wraps a bool for naming on the eventbus
type AutoUpdate struct {
ClientID int64 // The ID field is used for consumers to differentiate instances of Direct
Value bool
}
// ControlTime wraps a [time.Time] for naming on the eventbus
type ControlTime struct {
ClientID int64 // The ID field is used for consumers to differentiate instances of Direct
Value time.Time
}
// If we go more than watchdogTimeout without hearing from the server,
// end the long poll. We should be receiving a keep alive ping
// every minute.
@@ -1085,14 +1111,12 @@ func (c *Direct) sendMapRequest(ctx context.Context, isStreaming bool, nu Netmap
c.logf("netmap: control says to open URL %v; no popBrowser func", u)
}
}
if resp.ClientVersion != nil && c.onClientVersion != nil {
c.onClientVersion(resp.ClientVersion)
if resp.ClientVersion != nil {
c.clientVersionPub.Publish(*resp.ClientVersion)
}
if resp.ControlTime != nil && !resp.ControlTime.IsZero() {
c.logf.JSON(1, "controltime", resp.ControlTime.UTC())
if c.onControlTime != nil {
c.onControlTime(*resp.ControlTime)
}
c.controlTimePub.Publish(ControlTime{c.controlClientID, *resp.ControlTime})
}
if resp.KeepAlive {
vlogf("netmap: got keep-alive")
@@ -1112,9 +1136,7 @@ func (c *Direct) sendMapRequest(ctx context.Context, isStreaming bool, nu Netmap
continue
}
if au, ok := resp.DefaultAutoUpdate.Get(); ok {
if c.onTailnetDefaultAutoUpdate != nil {
c.onTailnetDefaultAutoUpdate(au)
}
c.autoUpdatePub.Publish(AutoUpdate{c.controlClientID, au})
}
metricMapResponseMap.Add(1)

View File

@@ -17,12 +17,14 @@ import (
"tailscale.com/net/tsdial"
"tailscale.com/tailcfg"
"tailscale.com/types/key"
"tailscale.com/util/eventbus/eventbustest"
)
func TestNewDirect(t *testing.T) {
hi := hostinfo.New()
ni := tailcfg.NetInfo{LinkType: "wired"}
hi.NetInfo = &ni
bus := eventbustest.NewBus(t)
k := key.NewMachine()
opts := Options{
@@ -32,6 +34,7 @@ func TestNewDirect(t *testing.T) {
return k, nil
},
Dialer: tsdial.NewDialer(netmon.NewStatic()),
Bus: bus,
}
c, err := NewDirect(opts)
if err != nil {
@@ -99,6 +102,7 @@ func TestTsmpPing(t *testing.T) {
hi := hostinfo.New()
ni := tailcfg.NetInfo{LinkType: "wired"}
hi.NetInfo = &ni
bus := eventbustest.NewBus(t)
k := key.NewMachine()
opts := Options{
@@ -108,6 +112,7 @@ func TestTsmpPing(t *testing.T) {
return k, nil
},
Dialer: tsdial.NewDialer(netmon.NewStatic()),
Bus: bus,
}
c, err := NewDirect(opts)