Add healthcheck for grpc server and client

Signed-off-by: Jian Qiu <jqiu@redhat.com>

Author: qiujian16

pkg/cloudevents/generic/options/grpc/agentoptions.go

@@ -2,6 +2,7 @@ package grpc
 
 import (
 	"context"
+	"time"
 
 	cloudevents "github.com/cloudevents/sdk-go/v2"
 
@@ -47,6 +48,7 @@ func (o *grpcAgentOptions) Protocol(ctx context.Context, dataType types.CloudEve
 			ClusterName: o.clusterName,
 			DataType:    dataType.String(),
 		}),
+		protocol.WithReconnectErrorOption(o.errorChan, 10*time.Second),
 	)
 	if err != nil {
 		return nil, err

pkg/cloudevents/generic/options/grpc/protocol/message.go

@@ -19,8 +19,8 @@ const (
 	prefix      = "ce-"
 	contenttype = "contenttype"
 	// dataSchema  = "dataschema"
-	subject = "subject"
-	time    = "time"
+	subject   = "subject"
+	timestamp = "time"
 )
 
 var specs = spec.WithPrefix(prefix)

pkg/cloudevents/generic/options/grpc/protocol/option.go

@@ -2,6 +2,7 @@ package protocol
 
 import (
 	"fmt"
+	"time"
 )
 
 // Option is the function signature
@@ -24,3 +25,14 @@ func WithSubscribeOption(subscribeOpt *SubscribeOption) Option {
 		return nil
 	}
 }
+
+func WithReconnectErrorOption(reconnectError chan error, interval time.Duration) Option {
+	return func(p *Protocol) error {
+		if reconnectError == nil {
+			return fmt.Errorf("the reconnect error option must not be nil")
+		}
+		p.reconnectErrorChan = reconnectError
+		p.checkInterval = interval
+		return nil
+	}
+}

pkg/cloudevents/generic/options/grpc/protocol/protocol.go

@@ -3,8 +3,10 @@ package protocol
 import (
 	"context"
 	"fmt"
+	healthpb "google.golang.org/grpc/health/grpc_health_v1"
 	"io"
 	"sync"
+	"time"
 
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/codes"
@@ -30,6 +32,10 @@ type Protocol struct {
 	openerMutex sync.Mutex
 
 	closeChan chan struct{}
+
+	// errorChan is to send an error messsage to restart the connection
+	reconnectErrorChan chan error
+	checkInterval      time.Duration
 }
 
 var (
@@ -137,6 +143,10 @@ func (p *Protocol) OpenInbound(ctx context.Context) error {
 		}
 	}()
 
+	if p.reconnectErrorChan != nil {
+		go p.healthCheck(ctx)
+	}
+
 	// Wait until external or internal context done
 	select {
 	case <-ctx.Done():
@@ -164,3 +174,38 @@ func (p *Protocol) Close(ctx context.Context) error {
 	close(p.closeChan)
 	return nil
 }
+
+// healthCheck check status of the server
+func (p *Protocol) healthCheck(ctx context.Context) {
+	logger := cecontext.LoggerFrom(ctx)
+	healthClient := healthpb.NewHealthClient(p.clientConn)
+	stream, err := healthClient.Watch(ctx, &healthpb.HealthCheckRequest{Service: ""})
+	if err != nil {
+		p.reconnectErrorChan <- err
+		return
+	}
+
+	for {
+		timeout := time.NewTimer(p.checkInterval)
+		done := make(chan struct{})
+
+		go func() {
+			resp, err := stream.Recv()
+			if err != nil {
+				p.reconnectErrorChan <- err
+				close(done)
+				return
+			}
+			logger.Infof("Received server health status %s", resp.Status)
+			close(done)
+		}()
+
+		select {
+		case <-timeout.C:
+			p.reconnectErrorChan <- fmt.Errorf("timeout waiting for health check")
+			return
+		case <-done:
+			timeout.Stop()
+		}
+	}
+}

pkg/cloudevents/generic/options/grpc/protocol/write_message.go

@@ -136,13 +136,13 @@ func (b *pbEventWriter) SetAttribute(attribute spec.Attribute, value interface{}
 		}
 	case spec.Time:
 		if value == nil {
-			delete(b.Attributes, prefix+time)
+			delete(b.Attributes, prefix+timestamp)
 		} else {
 			attrVal, err := attributeFor(value)
 			if err != nil {
 				return err
 			}
-			b.Attributes[prefix+time] = attrVal
+			b.Attributes[prefix+timestamp] = attrVal
 		}
 	default:
 		if value == nil {

pkg/cloudevents/generic/options/grpc/sourceoptions.go

@@ -2,6 +2,7 @@ package grpc
 
 import (
 	"context"
+	"time"
 
 	cloudevents "github.com/cloudevents/sdk-go/v2"
 
@@ -42,6 +43,7 @@ func (o *gRPCSourceOptions) Protocol(ctx context.Context, dataType types.CloudEv
 			Source:   o.sourceID,
 			DataType: dataType.String(),
 		}),
+		protocol.WithReconnectErrorOption(o.errorChan, 10*time.Second),
 	)
 	if err != nil {
 		return nil, err

pkg/server/grpc/health/health.go

@@ -0,0 +1,42 @@
+package health
+
+import (
+	"context"
+	"google.golang.org/grpc"
+	"time"
+
+	healthpb "google.golang.org/grpc/health/grpc_health_v1"
+)
+
+// Custom health server with periodic broadcasts
+type heartbeatHealthServer struct {
+	interval time.Duration
+	healthpb.UnimplementedHealthServer
+	status healthpb.HealthCheckResponse_ServingStatus
+}
+
+func RegisterHeartbeatHealthServer(srv *grpc.Server, interval time.Duration) {
+	healthpb.RegisterHealthServer(srv, &heartbeatHealthServer{
+		interval: interval,
+	})
+}
+
+func (s *heartbeatHealthServer) Check(ctx context.Context, req *healthpb.HealthCheckRequest) (*healthpb.HealthCheckResponse, error) {
+	return &healthpb.HealthCheckResponse{Status: s.status}, nil
+}
+
+func (s *heartbeatHealthServer) Watch(req *healthpb.HealthCheckRequest, stream healthpb.Health_WatchServer) error {
+	ticker := time.NewTicker(s.interval) // send every 5s
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ticker.C:
+			if err := stream.Send(&healthpb.HealthCheckResponse{Status: s.status}); err != nil {
+				return err
+			}
+		case <-stream.Context().Done():
+			return stream.Context().Err()
+		}
+	}
+}

pkg/server/grpc/server.go

@@ -6,7 +6,9 @@ import (
 	"crypto/x509"
 	"fmt"
 	"net"
+	"open-cluster-management.io/sdk-go/pkg/server/grpc/health"
 	"os"
+	"time"
 
 	grpcprom "github.com/grpc-ecosystem/go-grpc-middleware/providers/prometheus"
 	"google.golang.org/grpc"
@@ -133,6 +135,8 @@ func (b *GRPCServer) Run(ctx context.Context) error {
 	metrics.RegisterGRPCMetrics(promMiddleware, b.extraMetrics...)
 	// initialize grpc server metrics with appropriate value.
 	promMiddleware.InitializeMetrics(grpcServer)
+	// register health server
+	health.RegisterHeartbeatHealthServer(grpcServer, 5*time.Second)
 
 	for _, r := range b.registerFuncs {
 		r(grpcServer)