feat: add boostd-data metrics (#1784)

* add boostd-data metrics * go mod tidy * separate metrics * cleanup incorrect metrics * reduce metrics name size * add new dashboards * add some traces to boostd-daata * add API metrics to boostd * improve dashboard, remove older * remove info view name * use unique info view name * add storage to docker, fix tests * fix context * remove extra else and metric
filecoin-project · Nov 1, 2023 · 059e592 · 059e592
1 parent b70bf8c
commit 059e592
Show file tree

Hide file tree

Showing 31 changed files with 8,582 additions and 5,223 deletions.
diff --git a/cmd/boostd/run.go b/cmd/boostd/run.go
@@ -2,10 +2,18 @@ package main
 
 import (
 	"fmt"
+	"net/http"
+	_ "net/http/pprof"
+	"time"
 
 	"github.com/filecoin-project/boost/api"
+	"github.com/filecoin-project/boost/build"
+	"github.com/filecoin-project/boost/metrics"
 	"github.com/filecoin-project/boost/node"
 	"github.com/filecoin-project/boost/node/modules/dtypes"
+	"go.opencensus.io/stats"
+	"go.opencensus.io/stats/view"
+	"go.opencensus.io/tag"
 
 	lapi "github.com/filecoin-project/lotus/api"
 	"github.com/filecoin-project/lotus/api/v0api"
@@ -16,9 +24,6 @@ import (
 	lcliutil "github.com/filecoin-project/lotus/cli/util"
 	lotus_repo "github.com/filecoin-project/lotus/node/repo"
 
-	"net/http"
-	_ "net/http/pprof"
-
 	"github.com/urfave/cli/v2"
 )
 
@@ -35,6 +40,10 @@ var runCmd = &cli.Command{
 			Name:  "nosync",
 			Usage: "dont wait for the full node to sync with the chain",
 		},
+		&cli.BoolFlag{
+			Name:  "no-metrics",
+			Usage: "stops emitting information about the node as metrics (param is used by tests)",
+		},
 	},
 	Action: func(cctx *cli.Context) error {
 		if cctx.Bool("pprof") {
@@ -55,6 +64,23 @@ var runCmd = &cli.Command{
 
 		ctx := lcli.ReqContext(cctx)
 
+		if !cctx.Bool("no-metrics") {
+			ctx, _ = tag.New(ctx,
+				tag.Insert(metrics.Version, build.BuildVersion),
+				tag.Insert(metrics.Commit, build.CurrentCommit),
+				tag.Insert(metrics.NodeType, "boostd"),
+				tag.Insert(metrics.StartedAt, time.Now().String()),
+			)
+			// Register all metric views
+			if err = view.Register(
+				metrics.DefaultViews...,
+			); err != nil {
+				log.Fatalf("Cannot register the view: %v", err)
+			}
+			// Set the metric to one so, it is published to the exporter
+			stats.Record(ctx, metrics.BoostInfo.M(1))
+		}
+
 		log.Debug("Checking full node version")
 
 		v, err := fullnodeApi.Version(ctx)

diff --git a/cmd/booster-bitswap/multiminer_retrieval_test.go b/cmd/booster-bitswap/multiminer_retrieval_test.go
@@ -122,6 +122,7 @@ func runBoosterBitswap(ctx context.Context, repo string, minerApiInfo []string,
 		"--api-fullnode=" + fullNodeApiInfo,
 		"--api-lid=" + lidApiInfo,
 		"--api-version-check=false",
+		"--no-metrics",
 	}
 	for _, apiInfo := range minerApiInfo {
 		args = append(args, "--api-storage="+apiInfo)

diff --git a/cmd/booster-bitswap/run.go b/cmd/booster-bitswap/run.go
@@ -4,7 +4,9 @@ import (
 	"fmt"
 	"net/http"
 	_ "net/http/pprof"
+	"time"
 
+	"github.com/filecoin-project/boost/build"
 	"github.com/filecoin-project/boost/cmd/lib"
 	"github.com/filecoin-project/boost/cmd/lib/filters"
 	"github.com/filecoin-project/boost/cmd/lib/remoteblockstore"
@@ -16,6 +18,9 @@ import (
 	"github.com/libp2p/go-libp2p/core/peer"
 	"github.com/mitchellh/go-homedir"
 	"github.com/urfave/cli/v2"
+	"go.opencensus.io/stats"
+	"go.opencensus.io/stats/view"
+	"go.opencensus.io/tag"
 )
 
 var runCmd = &cli.Command{
@@ -120,6 +125,10 @@ var runCmd = &cli.Command{
 			Hidden: true,
 			Value:  true,
 		},
+		&cli.BoolFlag{
+			Name:  "no-metrics",
+			Usage: "stops emitting information about the node as metrics (param is used by tests)",
+		},
 	},
 	Action: func(cctx *cli.Context) error {
 		if cctx.Bool("pprof") {
@@ -133,6 +142,22 @@ var runCmd = &cli.Command{
 		}
 
 		ctx := lcli.ReqContext(cctx)
+		if !cctx.Bool("no-metrics") {
+			ctx, _ = tag.New(ctx,
+				tag.Insert(metrics.Version, build.BuildVersion),
+				tag.Insert(metrics.Commit, build.CurrentCommit),
+				tag.Insert(metrics.NodeType, "booster-bitswap"),
+				tag.Insert(metrics.StartedAt, time.Now().String()),
+			)
+			// Register all metric views
+			if err := view.Register(
+				metrics.DefaultViews...,
+			); err != nil {
+				log.Fatalf("Cannot register the view: %v", err)
+			}
+			// Set the metric to one so, it is published to the exporter
+			stats.Record(ctx, metrics.BoostInfo.M(1))
+		}
 
 		// Instantiate the tracer and exporter
 		if cctx.Bool("tracing") {

diff --git a/cmd/booster-http/multiminer_retrieval_test.go b/cmd/booster-http/multiminer_retrieval_test.go
@@ -108,6 +108,7 @@ func runBoosterHttp(ctx context.Context, t *testing.T, minerApiInfo []string, fu
 		"--api-fullnode=" + fullNodeApiInfo,
 		"--api-lid=" + lidApiInfo,
 		"--api-version-check=false",
+		"--no-metrics",
 	}, args...)
 	for _, apiInfo := range minerApiInfo {
 		args = append(args, "--api-storage="+apiInfo)

diff --git a/cmd/booster-http/run.go b/cmd/booster-http/run.go
@@ -8,7 +8,9 @@ import (
 	"net/http"
 	_ "net/http/pprof"
 	"os"
+	"time"
 
+	"github.com/filecoin-project/boost/build"
 	"github.com/filecoin-project/boost/cmd/lib"
 	"github.com/filecoin-project/boost/cmd/lib/filters"
 	"github.com/filecoin-project/boost/cmd/lib/remoteblockstore"
@@ -24,6 +26,9 @@ import (
 	"github.com/ipfs/go-cid"
 	"github.com/mitchellh/go-homedir"
 	"github.com/urfave/cli/v2"
+	"go.opencensus.io/stats"
+	"go.opencensus.io/stats/view"
+	"go.opencensus.io/tag"
 )
 
 const (
@@ -153,6 +158,10 @@ var runCmd = &cli.Command{
 			Hidden: true,
 			Value:  true,
 		},
+		&cli.BoolFlag{
+			Name:  "no-metrics",
+			Usage: "stops emitting information about the node as metrics (param is used by tests)",
+		},
 	},
 	Action: func(cctx *cli.Context) error {
 		servePieces := cctx.Bool("serve-pieces")
@@ -171,8 +180,26 @@ var runCmd = &cli.Command{
 			}()
 		}
 
-		// Connect to the local index directory service
 		ctx := lcli.ReqContext(cctx)
+
+		if !cctx.Bool("no-metrics") {
+			ctx, _ = tag.New(ctx,
+				tag.Insert(metrics.Version, build.BuildVersion),
+				tag.Insert(metrics.Commit, build.CurrentCommit),
+				tag.Insert(metrics.NodeType, "booster-http"),
+				tag.Insert(metrics.StartedAt, time.Now().String()),
+			)
+			// Register all metric views
+			if err := view.Register(
+				metrics.DefaultViews...,
+			); err != nil {
+				log.Fatalf("Cannot register the view: %v", err)
+			}
+			// Set the metric to one so, it is published to the exporter
+			stats.Record(ctx, metrics.BoostInfo.M(1))
+		}
+
+		// Connect to the local index directory service
 		cl := bdclient.NewStore()
 		defer cl.Close(ctx)
 		err := cl.Dial(ctx, cctx.String("api-lid"))

diff --git a/docker/monitoring/.env b/docker/monitoring/.env
@@ -0,0 +1 @@
+MONITORING_DATA=${HOME}/.boost-monitoring
diff --git a/docker/monitoring/docker-compose.yaml b/docker/monitoring/docker-compose.yaml
@@ -12,7 +12,7 @@ services:
     volumes:
       - ./tempo-local.yaml:/etc/tempo.yaml
       - ./overrides.yaml:/etc/overrides.yaml
-      - ./tempo-data:/tmp/tempo
+      - ${MONITORING_DATA}/tempo-data:/tmp/tempo
     ports:
       - "14268:14268"  # jaeger ingest
     restart: unless-stopped
@@ -44,6 +44,7 @@ services:
     command: [ "--config.file=/etc/prometheus.yaml" ]
     volumes:
       - ./${PROMETHEUS_CONFIG_FILE:-prometheus.yaml}:/etc/prometheus.yaml
+      - ${MONITORING_DATA}/prometheus-data:/prometheus
     restart: unless-stopped
     ports:
       - "9190:9090"
@@ -56,6 +57,7 @@ services:
       #- "booster-http:host-gateway"
       #- "booster-bitswap:host-gateway"
       #- "lotus-miner:host-gateway"
+      #- "boostd-data:host-gateway"
 
     logging:
       driver: loki
@@ -76,7 +78,7 @@ services:
       #- GF_AUTH_DISABLE_LOGIN_FORM=true
       GF_AUTH_SIGV4_AUTH_ENABLED: true
       AWS_SDK_LOAD_CONFIG: true
-      GF_DASHBOARDS_DEFAULT_HOME_DASHBOARD_PATH: "/var/lib/grafana/dashboards/exported_dashboard.json"
+      GF_DASHBOARDS_DEFAULT_HOME_DASHBOARD_PATH: "/var/lib/grafana/dashboards/boostd.json"
     ports:
       - "3333:3000"
     restart: unless-stopped