metric[engine]:Replace framework metric

engine/executor/http.go

@@ -20,7 +20,7 @@ import (
 
 	"github.com/pingcap/tiflow/dm/dm/common"
 	"github.com/pingcap/tiflow/dm/pkg/log"
-	"github.com/prometheus/client_golang/prometheus/promhttp"
+	"github.com/pingcap/tiflow/engine/pkg/promutil"
 )
 
 func httpHandler(lis net.Listener) error {
@@ -31,7 +31,7 @@ func httpHandler(lis net.Listener) error {
 	mux.HandleFunc("/debug/pprof/profile", pprof.Profile)
 	mux.HandleFunc("/debug/pprof/symbol", pprof.Symbol)
 	mux.HandleFunc("/debug/pprof/trace", pprof.Trace)
-	mux.Handle("/metrics", promhttp.Handler())
+	mux.Handle("/metrics", promutil.HTTPHandlerForMetric())
 
 	httpS := &http.Server{
 		Handler: mux,

engine/executor/metrics_server.go

@@ -14,18 +14,17 @@
 package executor
 
 import (
+	"github.com/pingcap/tiflow/engine/pkg/promutil"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
-var executorTaskNumGauge = prometheus.NewGaugeVec(
-	prometheus.GaugeOpts{
-		Namespace: "dataflow",
-		Subsystem: "executor",
-		Name:      "task_num",
-		Help:      "number of task in this executor",
-	}, []string{"status"})
-
-// initServerMetrics registers statistics of executor server
-func initServerMetrics(registry *prometheus.Registry) {
-	registry.MustRegister(executorTaskNumGauge)
-}
+var (
+	executorFactory      = promutil.NewFactory4Framework()
+	executorTaskNumGauge = executorFactory.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Namespace: "dataflow",
+			Subsystem: "executor",
+			Name:      "task_num",
+			Help:      "number of task in this executor",
+		}, []string{"status"})
+)

engine/executor/server.go

@@ -325,8 +325,6 @@ func (s *Server) Run(ctx context.Context) error {
 		return s.startForTest(ctx)
 	}
 
-	registerMetrics()
-
 	wg, ctx := errgroup.WithContext(ctx)
 	s.taskRunner = worker.NewTaskRunner(defaultRuntimeIncomingQueueLen, defaultRuntimeInitConcurrency)
 	s.taskCommitter = worker.NewTaskCommitter(s.taskRunner, defaultTaskPreDispatchRequestTTL)

engine/executor/server_test.go

@@ -53,7 +53,6 @@ func TestStartTCPSrv(t *testing.T) {
 	s := NewServer(cfg, nil)
 
 	s.grpcSrv = grpc.NewServer()
-	registerMetrics()
 	wg, ctx := errgroup.WithContext(context.Background())
 	err = s.startTCPService(ctx, wg)
 	require.Nil(t, err)
@@ -114,7 +113,6 @@ func TestCollectMetric(t *testing.T) {
 	s.taskRunner = worker.NewTaskRunner(defaultRuntimeIncomingQueueLen, defaultRuntimeInitConcurrency)
 
 	s.grpcSrv = grpc.NewServer()
-	registerMetrics()
 	err = s.startTCPService(ctx, wg)
 	require.Nil(t, err)
 

engine/lib/base_jobmaster.go

@@ -18,6 +18,7 @@ import (
 
 	"github.com/pingcap/errors"
 	"github.com/pingcap/tiflow/dm/pkg/log"
+	"go.uber.org/zap"
 
 	"github.com/pingcap/tiflow/engine/executor/worker"
 	libModel "github.com/pingcap/tiflow/engine/lib/model"
@@ -180,13 +181,14 @@ func (d *DefaultBaseJobMaster) GetWorkers() map[libModel.WorkerID]WorkerHandle {
 
 // Close implements BaseJobMaster.Close
 func (d *DefaultBaseJobMaster) Close(ctx context.Context) error {
-	if err := d.impl.CloseImpl(ctx); err != nil {
-		return errors.Trace(err)
+	err := d.impl.CloseImpl(ctx)
+	if err != nil {
+		log.L().Error("Failed to close JobMasterImpl", zap.Error(err))
 	}
 
 	d.master.doClose()
 	d.worker.doClose()
-	return nil
+	return errors.Trace(err)
 }
 
 // OnError implements BaseJobMaster.OnError

engine/lib/master.go

@@ -435,16 +435,18 @@ func (m *DefaultBaseMaster) doClose() {
 		log.L().Warn("Failed to clean up message handlers",
 			zap.String("master-id", m.id))
 	}
+	promutil.UnregisterWorkerMetrics(m.id)
 }
 
 // Close implements BaseMaster.Close
 func (m *DefaultBaseMaster) Close(ctx context.Context) error {
-	if err := m.Impl.CloseImpl(ctx); err != nil {
-		return errors.Trace(err)
+	err := m.Impl.CloseImpl(ctx)
+	if err != nil {
+		log.L().Error("Failed to close MasterImpl", zap.Error(err))
 	}
 
 	m.doClose()
-	return nil
+	return errors.Trace(err)
 }
 
 // OnError implements BaseMaster.OnError

engine/lib/worker.go

@@ -350,17 +350,18 @@ func (w *DefaultBaseWorker) doClose() {
 	}
 
 	w.wg.Wait()
+	promutil.UnregisterWorkerMetrics(w.id)
 }
 
 // Close implements BaseWorker.Close
 func (w *DefaultBaseWorker) Close(ctx context.Context) error {
-	if err := w.Impl.CloseImpl(ctx); err != nil {
+	err := w.Impl.CloseImpl(ctx)
+	if err != nil {
 		log.L().Error("Failed to close WorkerImpl", zap.Error(err))
-		return errors.Trace(err)
 	}
 
 	w.doClose()
-	return nil
+	return errors.Trace(err)
 }
 
 // ID implements BaseWorker.ID

engine/pkg/promutil/implement.go

@@ -87,16 +87,25 @@ func (f *wrappingFactory) NewHistogramVec(opts prometheus.HistogramOpts, labelNa
 }
 
 func wrapCounterOpts(prefix string, constLabels prometheus.Labels, opts *prometheus.CounterOpts) *prometheus.CounterOpts {
+	if opts.ConstLabels == nil && constLabels != nil {
+		opts.ConstLabels = make(prometheus.Labels)
+	}
 	wrapOptsCommon(prefix, constLabels, &opts.Namespace, opts.ConstLabels)
 	return opts
 }
 
 func wrapGaugeOpts(prefix string, constLabels prometheus.Labels, opts *prometheus.GaugeOpts) *prometheus.GaugeOpts {
+	if opts.ConstLabels == nil && constLabels != nil {
+		opts.ConstLabels = make(prometheus.Labels)
+	}
 	wrapOptsCommon(prefix, constLabels, &opts.Namespace, opts.ConstLabels)
 	return opts
 }
 
 func wrapHistogramOpts(prefix string, constLabels prometheus.Labels, opts *prometheus.HistogramOpts) *prometheus.HistogramOpts {
+	if opts.ConstLabels == nil && constLabels != nil {
+		opts.ConstLabels = make(prometheus.Labels)
+	}
 	wrapOptsCommon(prefix, constLabels, &opts.Namespace, opts.ConstLabels)
 	return opts
 }

engine/pkg/promutil/implement_test.go

@@ -69,6 +69,17 @@ func TestWrapCounterOpts(t *testing.T) {
 				},
 			},
 		},
+		{
+			constLabels: prometheus.Labels{
+				"k2": "v2",
+			},
+			inputOpts: &prometheus.CounterOpts{},
+			outputOpts: &prometheus.CounterOpts{
+				ConstLabels: prometheus.Labels{
+					"k2": "v2",
+				},
+			},
+		},
 	}
 
 	for _, c := range cases {

engine/pkg/promutil/util.go

@@ -72,3 +72,10 @@ func NewFactory4Worker(info tenant.ProjectInfo, jobType libModel.JobType, jobID
 func NewFactory4Framework() Factory {
 	return NewFactory4FrameworkImpl(globalMetricRegistry)
 }
+
+// UnregisterWorkerMetrics unregisters all metric of workerID
+// IF 'worker' is a job master, use job id as workerID
+// IF 'worker' is a worker, use worke id as workerID
+func UnregisterWorkerMetrics(workerID libModel.WorkerID) {
+	globalMetricRegistry.Unregister(workerID)
+}

engine/servermaster/metrics_server.go

@@ -14,28 +14,24 @@
 package servermaster
 
 import (
+	"github.com/pingcap/tiflow/engine/pkg/promutil"
 	"github.com/prometheus/client_golang/prometheus"
 )
 
 var (
-	serverExecutorNumGauge = prometheus.NewGaugeVec(
+	serverFactory          = promutil.NewFactory4Framework()
+	serverExecutorNumGauge = serverFactory.NewGaugeVec(
 		prometheus.GaugeOpts{
 			Namespace: "dataflow",
 			Subsystem: "server_master",
 			Name:      "executor_num",
 			Help:      "number of executor servers in this cluster",
 		}, []string{"status"})
-	serverJobNumGauge = prometheus.NewGaugeVec(
+	serverJobNumGauge = serverFactory.NewGaugeVec(
 		prometheus.GaugeOpts{
 			Namespace: "dataflow",
 			Subsystem: "server_master",
 			Name:      "job_num",
 			Help:      "number of jobs in this cluster",
 		}, []string{"status"})
 )
-
-// initServerMetrics registers statistics of server
-func initServerMetrics(registry *prometheus.Registry) {
-	registry.MustRegister(serverExecutorNumGauge)
-	registry.MustRegister(serverJobNumGauge)
-}

engine/servermaster/server.go

@@ -27,7 +27,6 @@ import (
 	"github.com/pingcap/tiflow/dm/pkg/log"
 	p2pProtocol "github.com/pingcap/tiflow/proto/p2p"
 	"github.com/prometheus/client_golang/prometheus"
-	"github.com/prometheus/client_golang/prometheus/promhttp"
 	clientv3 "go.etcd.io/etcd/client/v3"
 	"go.etcd.io/etcd/client/v3/concurrency"
 	"go.etcd.io/etcd/server/v3/embed"
@@ -56,6 +55,7 @@ import (
 	"github.com/pingcap/tiflow/engine/pkg/meta/metaclient"
 	pkgOrm "github.com/pingcap/tiflow/engine/pkg/orm"
 	"github.com/pingcap/tiflow/engine/pkg/p2p"
+	"github.com/pingcap/tiflow/engine/pkg/promutil"
 	"github.com/pingcap/tiflow/engine/pkg/rpcutil"
 	"github.com/pingcap/tiflow/engine/pkg/serverutils"
 	"github.com/pingcap/tiflow/engine/pkg/tenant"
@@ -447,8 +447,6 @@ func (s *Server) Run(ctx context.Context) (err error) {
 		return s.startForTest(ctx)
 	}
 
-	registerMetrics()
-
 	err = s.registerMetaStore()
 	if err != nil {
 		return err
@@ -570,7 +568,7 @@ func (s *Server) startGrpcSrv(ctx context.Context) (err error) {
 
 	httpHandlers := map[string]http.Handler{
 		"/debug/":  getDebugHandler(),
-		"/metrics": promhttp.Handler(),
+		"/metrics": promutil.HTTPHandlerForMetric(),
 	}
 
 	// generate grpcServer

engine/servermaster/server_test.go

@@ -86,7 +86,6 @@ func TestStartGrpcSrv(t *testing.T) {
 	defer cleanup()
 
 	s := &Server{cfg: cfg}
-	registerMetrics()
 	ctx := context.Background()
 	err := s.startGrpcSrv(ctx)
 	require.Nil(t, err)
@@ -313,7 +312,6 @@ func TestCollectMetric(t *testing.T) {
 	masterAddr, cfg, cleanup := prepareServerEnv(t, "test-collect-metric")
 	defer cleanup()
 
-	registerMetrics()
 	s := &Server{
 		cfg:     cfg,
 		metrics: newServerMasterMetric(),