feat: use tailnet v2 API for coordination (#11638)

This one is huge, and I'm sorry.

The problem is that once I change `tailnet.Conn` to start doing v2 behavior, I kind of have to change it everywhere, including in CoderSDK (CLI), the agent, wsproxy, and ServerTailnet.

There is still a bit more cleanup to do, and I need to add code so that when we lose connection to the Coordinator, we mark all peers as LOST, but that will be in a separate PR since this is big enough!
This commit is contained in:
Spike Curtis
2024-01-22 11:07:50 +04:00
committed by GitHub
parent 5a2cf7cd14
commit f01cab9894
31 changed files with 1192 additions and 1114 deletions
+26 -3
View File
@@ -33,6 +33,7 @@ import (
"github.com/coder/coder/v2/codersdk"
"github.com/coder/coder/v2/provisioner/echo"
"github.com/coder/coder/v2/tailnet"
tailnetproto "github.com/coder/coder/v2/tailnet/proto"
"github.com/coder/coder/v2/testutil"
)
@@ -98,14 +99,32 @@ func TestDERP(t *testing.T) {
w2Ready := make(chan struct{})
w2ReadyOnce := sync.Once{}
w1ID := uuid.New()
w1.SetNodeCallback(func(node *tailnet.Node) {
w2.UpdateNodes([]*tailnet.Node{node}, false)
pn, err := tailnet.NodeToProto(node)
if !assert.NoError(t, err) {
return
}
w2.UpdatePeers([]*tailnetproto.CoordinateResponse_PeerUpdate{{
Id: w1ID[:],
Node: pn,
Kind: tailnetproto.CoordinateResponse_PeerUpdate_NODE,
}})
w2ReadyOnce.Do(func() {
close(w2Ready)
})
})
w2ID := uuid.New()
w2.SetNodeCallback(func(node *tailnet.Node) {
w1.UpdateNodes([]*tailnet.Node{node}, false)
pn, err := tailnet.NodeToProto(node)
if !assert.NoError(t, err) {
return
}
w1.UpdatePeers([]*tailnetproto.CoordinateResponse_PeerUpdate{{
Id: w2ID[:],
Node: pn,
Kind: tailnetproto.CoordinateResponse_PeerUpdate_NODE,
}})
})
conn := make(chan struct{})
@@ -199,7 +218,11 @@ func TestDERPForceWebSockets(t *testing.T) {
defer cancel()
resources := coderdtest.AwaitWorkspaceAgents(t, client, workspace.ID)
conn, err := client.DialWorkspaceAgent(ctx, resources[0].Agents[0].ID, nil)
conn, err := client.DialWorkspaceAgent(ctx, resources[0].Agents[0].ID,
&codersdk.DialWorkspaceAgentOptions{
Logger: slogtest.Make(t, nil).Leveled(slog.LevelDebug).Named("client"),
},
)
require.NoError(t, err)
defer func() {
_ = conn.Close()
+17 -18
View File
@@ -121,12 +121,23 @@ func NewServerTailnet(
}
tn.agentConn.Store(&agentConn)
err = tn.getAgentConn().UpdateSelf(conn.Node())
pn, err := tailnet.NodeToProto(conn.Node())
if err != nil {
tn.logger.Warn(context.Background(), "server tailnet update self", slog.Error(err))
tn.logger.Critical(context.Background(), "failed to convert node", slog.Error(err))
} else {
err = tn.getAgentConn().UpdateSelf(pn)
if err != nil {
tn.logger.Warn(context.Background(), "server tailnet update self", slog.Error(err))
}
}
conn.SetNodeCallback(func(node *tailnet.Node) {
err := tn.getAgentConn().UpdateSelf(node)
pn, err := tailnet.NodeToProto(node)
if err != nil {
tn.logger.Critical(context.Background(), "failed to convert node", slog.Error(err))
return
}
err = tn.getAgentConn().UpdateSelf(pn)
if err != nil {
tn.logger.Warn(context.Background(), "broadcast server node to agents", slog.Error(err))
}
@@ -191,21 +202,9 @@ func (s *ServerTailnet) doExpireOldAgents(cutoff time.Duration) {
// If no one has connected since the cutoff and there are no active
// connections, remove the agent.
if time.Since(lastConnection) > cutoff && len(s.agentTickets[agentID]) == 0 {
deleted, err := s.conn.RemovePeer(tailnet.PeerSelector{
ID: tailnet.NodeID(agentID),
IP: netip.PrefixFrom(tailnet.IPFromUUID(agentID), 128),
})
if err != nil {
s.logger.Warn(ctx, "failed to remove peer from server tailnet", slog.Error(err))
continue
}
if !deleted {
s.logger.Warn(ctx, "peer didn't exist in tailnet", slog.Error(err))
}
deletedCount++
delete(s.agentConnectionTimes, agentID)
err = agentConn.UnsubscribeAgent(agentID)
err := agentConn.UnsubscribeAgent(agentID)
if err != nil {
s.logger.Error(ctx, "unsubscribe expired agent", slog.Error(err), slog.F("agent_id", agentID))
}
@@ -221,7 +220,7 @@ func (s *ServerTailnet) doExpireOldAgents(cutoff time.Duration) {
func (s *ServerTailnet) watchAgentUpdates() {
for {
conn := s.getAgentConn()
nodes, ok := conn.NextUpdate(s.ctx)
resp, ok := conn.NextUpdate(s.ctx)
if !ok {
if conn.IsClosed() && s.ctx.Err() == nil {
s.logger.Warn(s.ctx, "multiagent closed, reinitializing")
@@ -231,7 +230,7 @@ func (s *ServerTailnet) watchAgentUpdates() {
return
}
err := s.conn.UpdateNodes(nodes, false)
err := s.conn.UpdatePeers(resp.GetPeerUpdates())
if err != nil {
if xerrors.Is(err, tailnet.ErrConnClosed) {
s.logger.Warn(context.Background(), "tailnet conn closed, exiting watchAgentUpdates", slog.Error(err))
+10 -13
View File
@@ -3,7 +3,6 @@ package coderd_test
import (
"context"
"fmt"
"net"
"net/http"
"net/http/httptest"
"net/netip"
@@ -204,22 +203,20 @@ func setupAgent(t *testing.T, agentAddresses []netip.Prefix) (uuid.UUID, agent.A
Logger: logger.Named("client"),
})
require.NoError(t, err)
clientConn, serverConn := net.Pipe()
serveClientDone := make(chan struct{})
t.Cleanup(func() {
_ = clientConn.Close()
_ = serverConn.Close()
_ = conn.Close()
<-serveClientDone
})
go func() {
defer close(serveClientDone)
coord.ServeClient(serverConn, uuid.New(), manifest.AgentID)
}()
sendNode, _ := tailnet.ServeCoordinator(clientConn, func(node []*tailnet.Node) error {
return conn.UpdateNodes(node, false)
clientID := uuid.New()
testCtx, testCtxCancel := context.WithCancel(context.Background())
t.Cleanup(testCtxCancel)
coordination := tailnet.NewInMemoryCoordination(
testCtx, logger,
clientID, manifest.AgentID,
coord, conn,
)
t.Cleanup(func() {
_ = coordination.Close()
})
conn.SetNodeCallback(sendNode)
return codersdk.NewWorkspaceAgentConn(conn, codersdk.WorkspaceAgentConnOptions{
AgentID: manifest.AgentID,
AgentIP: codersdk.WorkspaceAgentIP,
+4 -4
View File
@@ -30,6 +30,10 @@ func (v *APIVersion) WithBackwardCompat(majs ...int) *APIVersion {
return v
}
func (v *APIVersion) String() string {
return fmt.Sprintf("%d.%d", v.supportedMajor, v.supportedMinor)
}
// Validate validates the given version against the given constraints:
// A given major.minor version is valid iff:
// 1. The requested major version is contained within v.supportedMajors
@@ -42,10 +46,6 @@ func (v *APIVersion) WithBackwardCompat(majs ...int) *APIVersion {
// - 1.x is supported,
// - 2.0, 2.1, and 2.2 are supported,
// - 2.3+ is not supported.
func (v *APIVersion) String() string {
return fmt.Sprintf("%d.%d", v.supportedMajor, v.supportedMinor)
}
func (v *APIVersion) Validate(version string) error {
major, minor, err := Parse(version)
if err != nil {
+5 -23
View File
@@ -857,8 +857,6 @@ func (api *API) workspaceAgentListeningPorts(rw http.ResponseWriter, r *http.Req
// Deprecated: use api.tailnet.AgentConn instead.
// See: https://github.com/coder/coder/issues/8218
func (api *API) _dialWorkspaceAgentTailnet(agentID uuid.UUID) (*codersdk.WorkspaceAgentConn, error) {
clientConn, serverConn := net.Pipe()
derpMap := api.DERPMap()
conn, err := tailnet.NewConn(&tailnet.Options{
Addresses: []netip.Prefix{netip.PrefixFrom(tailnet.IP(), 128)},
@@ -868,8 +866,6 @@ func (api *API) _dialWorkspaceAgentTailnet(agentID uuid.UUID) (*codersdk.Workspa
BlockEndpoints: api.DeploymentValues.DERP.Config.BlockDirect.Value(),
})
if err != nil {
_ = clientConn.Close()
_ = serverConn.Close()
return nil, xerrors.Errorf("create tailnet conn: %w", err)
}
ctx, cancel := context.WithCancel(api.ctx)
@@ -887,10 +883,10 @@ func (api *API) _dialWorkspaceAgentTailnet(agentID uuid.UUID) (*codersdk.Workspa
return left
})
sendNodes, _ := tailnet.ServeCoordinator(clientConn, func(nodes []*tailnet.Node) error {
return conn.UpdateNodes(nodes, true)
})
conn.SetNodeCallback(sendNodes)
clientID := uuid.New()
coordination := tailnet.NewInMemoryCoordination(ctx, api.Logger,
clientID, agentID,
*(api.TailnetCoordinator.Load()), conn)
// Check for updated DERP map every 5 seconds.
go func() {
@@ -920,27 +916,13 @@ func (api *API) _dialWorkspaceAgentTailnet(agentID uuid.UUID) (*codersdk.Workspa
AgentID: agentID,
AgentIP: codersdk.WorkspaceAgentIP,
CloseFunc: func() error {
_ = coordination.Close()
cancel()
_ = clientConn.Close()
_ = serverConn.Close()
return nil
},
})
go func() {
err := (*api.TailnetCoordinator.Load()).ServeClient(serverConn, uuid.New(), agentID)
if err != nil {
// Sometimes, we get benign closed pipe errors when the server is
// shutting down.
if api.ctx.Err() == nil {
api.Logger.Warn(ctx, "tailnet coordinator client error", slog.Error(err))
}
_ = agentConn.Close()
}
}()
if !agentConn.AwaitReachable(ctx) {
_ = agentConn.Close()
_ = serverConn.Close()
_ = clientConn.Close()
cancel()
return nil, xerrors.Errorf("agent not reachable")
}
-1
View File
@@ -535,7 +535,6 @@ func TestWorkspaceAgentTailnetDirectDisabled(t *testing.T) {
})
require.NoError(t, err)
defer conn.Close()
require.True(t, conn.BlockEndpoints())
require.True(t, conn.AwaitReachable(ctx))
_, p2p, _, err := conn.Ping(ctx)
+78 -26
View File
@@ -12,14 +12,19 @@ import (
"net/url"
"strings"
"sync"
"sync/atomic"
"testing"
"time"
"github.com/google/uuid"
"github.com/stretchr/testify/assert"
"github.com/stretchr/testify/require"
"go.uber.org/atomic"
"go.uber.org/goleak"
"golang.org/x/xerrors"
"storj.io/drpc"
"storj.io/drpc/drpcmux"
"storj.io/drpc/drpcserver"
"tailscale.com/tailcfg"
"cdr.dev/slog"
"cdr.dev/slog/sloggers/slogtest"
@@ -27,7 +32,9 @@ import (
"github.com/coder/coder/v2/coderd/wsconncache"
"github.com/coder/coder/v2/codersdk"
"github.com/coder/coder/v2/codersdk/agentsdk"
drpcsdk "github.com/coder/coder/v2/codersdk/drpc"
"github.com/coder/coder/v2/tailnet"
"github.com/coder/coder/v2/tailnet/proto"
"github.com/coder/coder/v2/tailnet/tailnettest"
"github.com/coder/coder/v2/testutil"
)
@@ -41,7 +48,7 @@ func TestCache(t *testing.T) {
t.Run("Same", func(t *testing.T) {
t.Parallel()
cache := wsconncache.New(func(id uuid.UUID) (*codersdk.WorkspaceAgentConn, error) {
return setupAgent(t, agentsdk.Manifest{}, 0), nil
return setupAgent(t, agentsdk.Manifest{}, 0)
}, 0)
defer func() {
_ = cache.Close()
@@ -54,10 +61,10 @@ func TestCache(t *testing.T) {
})
t.Run("Expire", func(t *testing.T) {
t.Parallel()
called := atomic.NewInt32(0)
called := int32(0)
cache := wsconncache.New(func(id uuid.UUID) (*codersdk.WorkspaceAgentConn, error) {
called.Add(1)
return setupAgent(t, agentsdk.Manifest{}, 0), nil
atomic.AddInt32(&called, 1)
return setupAgent(t, agentsdk.Manifest{}, 0)
}, time.Microsecond)
defer func() {
_ = cache.Close()
@@ -70,12 +77,12 @@ func TestCache(t *testing.T) {
require.NoError(t, err)
release()
<-conn.Closed()
require.Equal(t, int32(2), called.Load())
require.Equal(t, int32(2), called)
})
t.Run("NoExpireWhenLocked", func(t *testing.T) {
t.Parallel()
cache := wsconncache.New(func(id uuid.UUID) (*codersdk.WorkspaceAgentConn, error) {
return setupAgent(t, agentsdk.Manifest{}, 0), nil
return setupAgent(t, agentsdk.Manifest{}, 0)
}, time.Microsecond)
defer func() {
_ = cache.Close()
@@ -108,7 +115,7 @@ func TestCache(t *testing.T) {
go server.Serve(random)
cache := wsconncache.New(func(id uuid.UUID) (*codersdk.WorkspaceAgentConn, error) {
return setupAgent(t, agentsdk.Manifest{}, 0), nil
return setupAgent(t, agentsdk.Manifest{}, 0)
}, time.Microsecond)
defer func() {
_ = cache.Close()
@@ -154,7 +161,7 @@ func TestCache(t *testing.T) {
})
}
func setupAgent(t *testing.T, manifest agentsdk.Manifest, ptyTimeout time.Duration) *codersdk.WorkspaceAgentConn {
func setupAgent(t *testing.T, manifest agentsdk.Manifest, ptyTimeout time.Duration) (*codersdk.WorkspaceAgentConn, error) {
t.Helper()
logger := slogtest.Make(t, nil).Leveled(slog.LevelDebug)
manifest.DERPMap, _ = tailnettest.RunDERPAndSTUN(t)
@@ -184,18 +191,25 @@ func setupAgent(t *testing.T, manifest agentsdk.Manifest, ptyTimeout time.Durati
DERPForceWebSockets: manifest.DERPForceWebSockets,
Logger: slogtest.Make(t, nil).Named("tailnet").Leveled(slog.LevelDebug),
})
require.NoError(t, err)
clientConn, serverConn := net.Pipe()
// setupAgent is called by wsconncache Dialer, so we can't use require here as it will end the
// test, which in turn closes the wsconncache, which in turn waits for the Dialer and deadlocks.
if !assert.NoError(t, err) {
return nil, err
}
t.Cleanup(func() {
_ = clientConn.Close()
_ = serverConn.Close()
_ = conn.Close()
})
go coordinator.ServeClient(serverConn, uuid.New(), manifest.AgentID)
sendNode, _ := tailnet.ServeCoordinator(clientConn, func(nodes []*tailnet.Node) error {
return conn.UpdateNodes(nodes, false)
clientID := uuid.New()
testCtx, testCtxCancel := context.WithCancel(context.Background())
t.Cleanup(testCtxCancel)
coordination := tailnet.NewInMemoryCoordination(
testCtx, logger,
clientID, manifest.AgentID,
coordinator, conn,
)
t.Cleanup(func() {
_ = coordination.Close()
})
conn.SetNodeCallback(sendNode)
agentConn := codersdk.NewWorkspaceAgentConn(conn, codersdk.WorkspaceAgentConnOptions{
AgentID: manifest.AgentID,
AgentIP: codersdk.WorkspaceAgentIP,
@@ -206,16 +220,20 @@ func setupAgent(t *testing.T, manifest agentsdk.Manifest, ptyTimeout time.Durati
ctx, cancel := context.WithTimeout(context.Background(), testutil.WaitMedium)
defer cancel()
if !agentConn.AwaitReachable(ctx) {
t.Fatal("agent not reachable")
// setupAgent is called by wsconncache Dialer, so we can't use t.Fatal here as it will end
// the test, which in turn closes the wsconncache, which in turn waits for the Dialer and
// deadlocks.
t.Error("agent not reachable")
return nil, xerrors.New("agent not reachable")
}
return agentConn
return agentConn, nil
}
type client struct {
t *testing.T
agentID uuid.UUID
manifest agentsdk.Manifest
coordinator tailnet.CoordinatorV1
coordinator tailnet.Coordinator
}
func (c *client) Manifest(_ context.Context) (agentsdk.Manifest, error) {
@@ -240,19 +258,53 @@ func (*client) DERPMapUpdates(_ context.Context) (<-chan agentsdk.DERPMapUpdate,
}, nil
}
func (c *client) Listen(_ context.Context) (net.Conn, error) {
clientConn, serverConn := net.Pipe()
func (c *client) Listen(_ context.Context) (drpc.Conn, error) {
logger := slogtest.Make(c.t, nil).Leveled(slog.LevelDebug).Named("drpc")
conn, lis := drpcsdk.MemTransportPipe()
closed := make(chan struct{})
c.t.Cleanup(func() {
_ = serverConn.Close()
_ = clientConn.Close()
_ = conn.Close()
_ = lis.Close()
<-closed
})
coordPtr := atomic.Pointer[tailnet.Coordinator]{}
coordPtr.Store(&c.coordinator)
mux := drpcmux.New()
drpcService := &tailnet.DRPCService{
CoordPtr: &coordPtr,
Logger: logger,
// TODO: handle DERPMap too!
DerpMapUpdateFrequency: time.Hour,
DerpMapFn: func() *tailcfg.DERPMap { panic("not implemented") },
}
err := proto.DRPCRegisterTailnet(mux, drpcService)
if err != nil {
return nil, xerrors.Errorf("register DRPC service: %w", err)
}
server := drpcserver.NewWithOptions(mux, drpcserver.Options{
Log: func(err error) {
if xerrors.Is(err, io.EOF) ||
xerrors.Is(err, context.Canceled) ||
xerrors.Is(err, context.DeadlineExceeded) {
return
}
logger.Debug(context.Background(), "drpc server error", slog.Error(err))
},
})
serveCtx, cancel := context.WithCancel(context.Background())
c.t.Cleanup(cancel)
auth := tailnet.AgentTunnelAuth{}
streamID := tailnet.StreamID{
Name: "wsconncache_test-agent",
ID: c.agentID,
Auth: auth,
}
serveCtx = tailnet.WithStreamID(serveCtx, streamID)
go func() {
_ = c.coordinator.ServeAgent(serverConn, c.agentID, "")
server.Serve(serveCtx, lis)
close(closed)
}()
return clientConn, nil
return conn, nil
}
func (*client) ReportStats(_ context.Context, _ slog.Logger, _ <-chan *agentsdk.Stats, _ func(time.Duration)) (io.Closer, error) {