chore: refactor sending telemetry (#15345)

Implements a tailnet API Telemetry controller by refactoring from `workspacesdk`.

chore re: #14729
This commit is contained in:
Spike Curtis
2024-11-06 20:23:23 +04:00
committed by GitHub
parent 9126cd78a6
commit 335e4ab6bf
4 changed files with 254 additions and 186 deletions
+83
View File
@@ -4,11 +4,14 @@ import (
"context"
"fmt"
"io"
"strings"
"sync"
"time"
"github.com/google/uuid"
"golang.org/x/xerrors"
"storj.io/drpc"
"storj.io/drpc/drpcerr"
"tailscale.com/tailcfg"
"cdr.dev/slog"
@@ -440,3 +443,83 @@ func (l *derpSetLoop) recvLoop() {
l.setter.SetDERPMap(dm)
}
}
type BasicTelemetryController struct {
logger slog.Logger
sync.Mutex
client TelemetryClient
unavailable bool
}
func (b *BasicTelemetryController) New(client TelemetryClient) {
b.Lock()
defer b.Unlock()
b.client = client
b.unavailable = false
b.logger.Debug(context.Background(), "new telemetry client connected to controller")
}
func (b *BasicTelemetryController) SendTelemetryEvent(event *proto.TelemetryEvent) {
b.Lock()
if b.client == nil {
b.Unlock()
b.logger.Debug(context.Background(),
"telemetry event dropped; no client", slog.F("event", event))
return
}
if b.unavailable {
b.Unlock()
b.logger.Debug(context.Background(),
"telemetry event dropped; unavailable", slog.F("event", event))
return
}
client := b.client
b.Unlock()
unavailable := sendTelemetry(b.logger, client, event)
if unavailable {
b.Lock()
defer b.Unlock()
if b.client == client {
b.unavailable = true
}
}
}
func NewBasicTelemetryController(logger slog.Logger) *BasicTelemetryController {
return &BasicTelemetryController{logger: logger}
}
var (
_ TelemetrySink = &BasicTelemetryController{}
_ TelemetryController = &BasicTelemetryController{}
)
func sendTelemetry(
logger slog.Logger, client TelemetryClient, event *proto.TelemetryEvent,
) (
unavailable bool,
) {
ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
_, err := client.PostTelemetry(ctx, &proto.TelemetryRequest{
Events: []*proto.TelemetryEvent{event},
})
if drpcerr.Code(err) == drpcerr.Unimplemented ||
drpc.ProtocolError.Has(err) &&
strings.Contains(err.Error(), "unknown rpc: ") {
logger.Debug(
context.Background(),
"attempted to send telemetry to a server that doesn't support it",
slog.Error(err),
)
return true
} else if err != nil {
logger.Warn(
context.Background(),
"failed to post telemetry event",
slog.F("event", event), slog.Error(err),
)
}
return false
}
+156
View File
@@ -13,6 +13,8 @@ import (
"github.com/stretchr/testify/require"
"go.uber.org/mock/gomock"
"golang.org/x/xerrors"
"storj.io/drpc"
"storj.io/drpc/drpcerr"
"tailscale.com/tailcfg"
"tailscale.com/types/key"
@@ -351,3 +353,157 @@ func (f fakeDERPClient) Recv() (*tailcfg.DERPMap, error) {
}
return nil, io.EOF
}
func TestBasicTelemetryController_Success(t *testing.T) {
t.Parallel()
ctx := testutil.Context(t, testutil.WaitShort)
logger := slogtest.Make(t, nil).Leveled(slog.LevelDebug)
uut := tailnet.NewBasicTelemetryController(logger)
ft := newFakeTelemetryClient()
uut.New(ft)
sendDone := make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{
Id: []byte("test event"),
})
}()
call := testutil.RequireRecvCtx(ctx, t, ft.calls)
require.Len(t, call.req.GetEvents(), 1)
require.Equal(t, call.req.GetEvents()[0].GetId(), []byte("test event"))
testutil.RequireSendCtx(ctx, t, call.errCh, nil)
testutil.RequireRecvCtx(ctx, t, sendDone)
}
func TestBasicTelemetryController_Unimplemented(t *testing.T) {
t.Parallel()
ctx := testutil.Context(t, testutil.WaitShort)
logger := slogtest.Make(t, nil).Leveled(slog.LevelDebug)
ft := newFakeTelemetryClient()
uut := tailnet.NewBasicTelemetryController(logger)
uut.New(ft)
// bad code, doesn't count
telemetryError := drpcerr.WithCode(xerrors.New("Unimplemented"), 0)
sendDone := make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{})
}()
call := testutil.RequireRecvCtx(ctx, t, ft.calls)
testutil.RequireSendCtx(ctx, t, call.errCh, telemetryError)
testutil.RequireRecvCtx(ctx, t, sendDone)
sendDone = make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{})
}()
// we get another call since it wasn't really the Unimplemented error
call = testutil.RequireRecvCtx(ctx, t, ft.calls)
// for real this time
telemetryError = drpcerr.WithCode(xerrors.New("Unimplemented"), drpcerr.Unimplemented)
testutil.RequireSendCtx(ctx, t, call.errCh, telemetryError)
testutil.RequireRecvCtx(ctx, t, sendDone)
// now this returns immediately without a call, because unimplemented error disables calling
sendDone = make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{})
}()
testutil.RequireRecvCtx(ctx, t, sendDone)
// getting a "new" client resets
uut.New(ft)
sendDone = make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{})
}()
call = testutil.RequireRecvCtx(ctx, t, ft.calls)
testutil.RequireSendCtx(ctx, t, call.errCh, nil)
testutil.RequireRecvCtx(ctx, t, sendDone)
}
func TestBasicTelemetryController_NotRecognised(t *testing.T) {
t.Parallel()
ctx := testutil.Context(t, testutil.WaitShort)
logger := slogtest.Make(t, nil).Leveled(slog.LevelDebug)
ft := newFakeTelemetryClient()
uut := tailnet.NewBasicTelemetryController(logger)
uut.New(ft)
sendDone := make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{})
}()
// returning generic protocol error doesn't trigger unknown rpc logic
call := testutil.RequireRecvCtx(ctx, t, ft.calls)
testutil.RequireSendCtx(ctx, t, call.errCh, drpc.ProtocolError.New("Protocol Error"))
testutil.RequireRecvCtx(ctx, t, sendDone)
sendDone = make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{})
}()
call = testutil.RequireRecvCtx(ctx, t, ft.calls)
// return the expected protocol error this time
testutil.RequireSendCtx(ctx, t, call.errCh,
drpc.ProtocolError.New("unknown rpc: /coder.tailnet.v2.Tailnet/PostTelemetry"))
testutil.RequireRecvCtx(ctx, t, sendDone)
// now this returns immediately without a call, because unimplemented error disables calling
sendDone = make(chan struct{})
go func() {
defer close(sendDone)
uut.SendTelemetryEvent(&proto.TelemetryEvent{})
}()
testutil.RequireRecvCtx(ctx, t, sendDone)
}
type fakeTelemetryClient struct {
calls chan *fakeTelemetryCall
}
var _ tailnet.TelemetryClient = &fakeTelemetryClient{}
func newFakeTelemetryClient() *fakeTelemetryClient {
return &fakeTelemetryClient{
calls: make(chan *fakeTelemetryCall),
}
}
// PostTelemetry implements tailnet.TelemetryClient
func (f *fakeTelemetryClient) PostTelemetry(ctx context.Context, req *proto.TelemetryRequest) (*proto.TelemetryResponse, error) {
fr := &fakeTelemetryCall{req: req, errCh: make(chan error)}
select {
case <-ctx.Done():
return nil, ctx.Err()
case f.calls <- fr:
// OK
}
select {
case <-ctx.Done():
return nil, ctx.Err()
case err := <-fr.errCh:
return &proto.TelemetryResponse{}, err
}
}
type fakeTelemetryCall struct {
req *proto.TelemetryRequest
errCh chan error
}