cmd/frostfs-node/config.go

@@ -446,6 +446,8 @@ type cfgGRPC struct {
 
 	servers []*grpc.Server
 
+	endpoints []string
+
 	maxChunkSize uint64
 
 	maxAddrAmount uint64

cmd/frostfs-node/grpc.go

@@ -69,11 +69,14 @@ func initGRPC(c *cfg) {
 
 		lis, err := net.Listen("tcp", sc.Endpoint())
 		if err != nil {
+			c.metricsCollector.GrpcServerMetrics().MarkUnhealthy(sc.Endpoint())
 			c.log.Error(logs.FrostFSNodeCantListenGRPCEndpoint, zap.Error(err))
 			return
 		}
+		c.metricsCollector.GrpcServerMetrics().MarkHealthy(sc.Endpoint())
 
 		c.cfgGRPC.listeners = append(c.cfgGRPC.listeners, lis)
+		c.cfgGRPC.endpoints = append(c.cfgGRPC.endpoints, sc.Endpoint())
 
 		srv := grpc.NewServer(serverOpts...)
 
@@ -96,6 +99,7 @@ func serveGRPC(c *cfg) {
 
 		srv := c.cfgGRPC.servers[i]
 		lis := c.cfgGRPC.listeners[i]
+		endpoint := c.cfgGRPC.endpoints[i]
 
 		go func() {
 			defer func() {
@@ -111,6 +115,7 @@ func serveGRPC(c *cfg) {
 			)
 
 			if err := srv.Serve(lis); err != nil {
+				c.metricsCollector.GrpcServerMetrics().MarkUnhealthy(endpoint)
 				fmt.Println("gRPC server error", err)
 			}
 		}()

pkg/metrics/consts.go

@@ -19,6 +19,7 @@ const (
 	stateSubsystem          = "state"
 	treeServiceSubsystem    = "treeservice"
 	writeCacheSubsystem     = "writecache"
+	grpcServerSubsystem     = "grpc_server"
 
 	successLabel          = "success"
 	shardIDLabel          = "shard_id"
@@ -35,6 +36,7 @@ const (
 	containerIDLabelKey   = "cid"
 	storageLabel          = "storage"
 	operationLabel        = "operation"
+	endpointLabel         = "endpoint"
 
 	readWriteMode = "READ_WRITE"
 	readOnlyMode  = "READ_ONLY"

pkg/metrics/grpc.go

@@ -0,0 +1,35 @@
+package metrics
+
+import (
+	"git.frostfs.info/TrueCloudLab/frostfs-observability/metrics"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+type GrpcServerMetrics interface {
+	MarkHealthy(endpoint string)
+	MarkUnhealthy(endpoint string)
+}
+
+type grpcServerMetrics struct {
+	endpointHealth *prometheus.GaugeVec
+}
+
+func newGrpcServerMetrics() *grpcServerMetrics {
+	return &grpcServerMetrics{
+		endpointHealth: metrics.NewGaugeVec(
+			prometheus.GaugeOpts{
+				Namespace: namespace,
+				Subsystem: grpcServerSubsystem,
+				Name:      "health",
+				Help:      "GRPC Server Endpoint health",
+			}, []string{endpointLabel}),
+	}
+}
+
+func (m *grpcServerMetrics) MarkHealthy(endpoint string) {
+	m.endpointHealth.With(prometheus.Labels{endpointLabel: endpoint}).Set(float64(1))
+}
+
+func (m *grpcServerMetrics) MarkUnhealthy(endpoint string) {
+	m.endpointHealth.With(prometheus.Labels{endpointLabel: endpoint}).Set(float64(0))
+}

pkg/metrics/node.go

@@ -17,6 +17,7 @@ type NodeMetrics struct {
 	blobobvnizca  *blobovnizca
 	metabase      *metabaseMetrics
 	pilorama      *piloramaMetrics
+	grpc          *grpcServerMetrics
 }
 
 func NewNodeMetrics() *NodeMetrics {
@@ -37,6 +38,7 @@ func NewNodeMetrics() *NodeMetrics {
 		blobobvnizca: newBlobovnizca(),
 		metabase:     newMetabaseMetrics(),
 		pilorama:     newPiloramaMetrics(),
+		grpc:         newGrpcServerMetrics(),
 	}
 }
 
@@ -84,3 +86,7 @@ func (m *NodeMetrics) MetabaseMetrics() MetabaseMetrics {
 func (m *NodeMetrics) PiloramaMetrics() PiloramaMetrics {
 	return m.pilorama
 }
+
+func (m *NodeMetrics) GrpcServerMetrics() GrpcServerMetrics {
+	return m.grpc
+}