[client, relay] Expose relay transport and connection errors in status and metrics (#6342)

shared/relay/client/picker.go

@@ -40,6 +40,7 @@ func (sp *ServerPicker) PickServer(parentCtx context.Context) (*Client, error) {
 
 	connResultChan := make(chan connResult, totalServers)
 	successChan := make(chan connResult, 1)
+	errChan := make(chan error, 1)
 	concurrentLimiter := make(chan struct{}, maxConcurrentServers)
 
 	log.Debugf("pick server from list: %v", sp.ServerURLs.Load().([]string))
@@ -54,17 +55,17 @@ func (sp *ServerPicker) PickServer(parentCtx context.Context) (*Client, error) {
 		}(url)
 	}
 
-	go sp.processConnResults(connResultChan, successChan)
+	go sp.processConnResults(connResultChan, successChan, errChan)
 
 	select {
 	case cr, ok := <-successChan:
 		if !ok {
-			return nil, errors.New("failed to connect to any relay server: all attempts failed")
+			return nil, <-errChan
 		}
 		log.Infof("chosen home Relay server: %s", cr.Url)
 		return cr.RelayClient, nil
 	case <-ctx.Done():
-		return nil, fmt.Errorf("failed to connect to any relay server: %w", ctx.Err())
+		return nil, fmt.Errorf("connect to relay server: %w", ctx.Err())
 	}
 }
 
@@ -80,12 +81,14 @@ func (sp *ServerPicker) startConnection(ctx context.Context, resultChan chan con
 	}
 }
 
-func (sp *ServerPicker) processConnResults(resultChan chan connResult, successChan chan connResult) {
+func (sp *ServerPicker) processConnResults(resultChan chan connResult, successChan chan connResult, errChan chan error) {
 	var hasSuccess bool
+	var errs []error
 	for numOfResults := 0; numOfResults < cap(resultChan); numOfResults++ {
 		cr := <-resultChan
 		if cr.Err != nil {
 			log.Tracef("failed to connect to Relay server: %s: %v", cr.Url, cr.Err)
+			errs = append(errs, cr.Err)
 			continue
 		}
 		log.Infof("connected to Relay server: %s", cr.Url)
@@ -101,5 +104,16 @@ func (sp *ServerPicker) processConnResults(resultChan chan connResult, successCh
 		hasSuccess = true
 		successChan <- cr
 	}
+	if !hasSuccess {
+		errChan <- pickErr(errs)
+	}
 	close(successChan)
 }
+
+// pickErr combines per-server connection failures into a single error.
+func pickErr(errs []error) error {
+	if len(errs) == 0 {
+		return errors.New("no relay server available")
+	}
+	return errors.Join(errs...)
+}