feat (dot/telemetry): implement telemetry system.interval message

dot/network/config.go

@@ -19,6 +19,7 @@ package network
 import (
 	"errors"
 	"path"
+	"time"
 
 	log "github.com/ChainSafe/log15"
 	"github.com/libp2p/go-libp2p-core/crypto"
@@ -93,6 +94,9 @@ type Config struct {
 
 	// PublishMetrics enables collection of network metrics
 	PublishMetrics bool
+
+	// telemetryInterval how often to send telemetry metrics
+	telemetryInterval time.Duration
 }
 
 // build checks the configuration, sets up the private key for the network service,
@@ -134,6 +138,11 @@ func (c *Config) build() error {
 		c.logger.Warn("Bootstrap is enabled but no bootstrap nodes are defined")
 	}
 
+	// set telemetryInterval to default
+	if c.telemetryInterval.Microseconds() == 0 {
+		c.telemetryInterval = time.Second * 5
+	}
+
 	return nil
 }
 

dot/network/host.go

@@ -27,6 +27,7 @@ import (
 	badger "github.com/ipfs/go-ds-badger2"
 	"github.com/libp2p/go-libp2p"
 	libp2phost "github.com/libp2p/go-libp2p-core/host"
+	"github.com/libp2p/go-libp2p-core/metrics"
 	libp2pnetwork "github.com/libp2p/go-libp2p-core/network"
 	"github.com/libp2p/go-libp2p-core/peer"
 	"github.com/libp2p/go-libp2p-core/peerstore"
@@ -59,6 +60,7 @@ type host struct {
 	cm              *ConnManager
 	ds              *badger.Datastore
 	messageCache    *messageCache
+	bwc             *metrics.BandwidthCounter
 }
 
 // newHost creates a host wrapper with a new libp2p host instance
@@ -167,6 +169,8 @@ func newHost(ctx context.Context, cfg *Config) (*host, error) {
 		return nil, err
 	}
 
+	bwc := metrics.NewBandwidthCounter()
+
 	host := &host{
 		ctx:             ctx,
 		h:               h,
@@ -177,6 +181,7 @@ func newHost(ctx context.Context, cfg *Config) (*host, error) {
 		ds:              ds,
 		persistentPeers: pps,
 		messageCache:    msgCache,
+		bwc:             bwc,
 	}
 
 	cm.host = host
@@ -311,8 +316,14 @@ func (h *host) writeToStream(s libp2pnetwork.Stream, msg Message) error {
 	lenBytes := uint64ToLEB128(msgLen)
 	encMsg = append(lenBytes, encMsg...)
 
-	_, err = s.Write(encMsg)
-	return err
+	sent, err := s.Write(encMsg)
+	if err != nil {
+		return err
+	}
+
+	h.bwc.LogSentMessage(int64(sent))
+
+	return nil
 }
 
 // getOutboundStream returns the outbound message stream for the given peer or returns

dot/network/service.go

@@ -26,11 +26,11 @@ import (
 	"time"
 
 	gssmrmetrics "github.com/ChainSafe/gossamer/dot/metrics"
+	"github.com/ChainSafe/gossamer/dot/telemetry"
 	"github.com/ChainSafe/gossamer/lib/common"
 	"github.com/ChainSafe/gossamer/lib/services"
-	"github.com/ethereum/go-ethereum/metrics"
-
 	log "github.com/ChainSafe/log15"
+	"github.com/ethereum/go-ethereum/metrics"
 	libp2pnetwork "github.com/libp2p/go-libp2p-core/network"
 	"github.com/libp2p/go-libp2p-core/peer"
 	"github.com/libp2p/go-libp2p-core/protocol"
@@ -89,6 +89,10 @@ type Service struct {
 	noDiscover  bool
 	noMDNS      bool
 	noGossip    bool // internal option
+
+	// telemetry
+	telemetryInterval    time.Duration
+	doneNetworkTelemetry chan interface{}
 }
 
 // NewService creates a new network service from the configuration and message channels
@@ -142,6 +146,8 @@ func NewService(cfg *Config) (*Service, error) {
 		syncer:                 cfg.Syncer,
 		notificationsProtocols: make(map[byte]*notificationsProtocol),
 		lightRequest:           make(map[peer.ID]struct{}),
+		telemetryInterval:      cfg.telemetryInterval,
+		doneNetworkTelemetry:   make(chan interface{}),
 	}
 
 	network.syncQueue = newSyncQueue(network)
@@ -246,6 +252,9 @@ func (s *Service) Start() error {
 	}
 
 	go s.logPeerCount()
+	go s.publishNetworkTelemetry(s.doneNetworkTelemetry)
+	go s.sentBlockIntervalTelemetry()
+
 	return nil
 }
 
@@ -279,6 +288,47 @@ func (s *Service) logPeerCount() {
 	}
 }
 
+func (s *Service) publishNetworkTelemetry(done chan interface{}) {
+	ticker := time.NewTicker(s.telemetryInterval)
+	defer ticker.Stop()
+
+main:
+	for {
+		select {
+		case <-done:
+			break main
+
+		case <-ticker.C:
+			o := s.host.bwc.GetBandwidthTotals()
+			telemetry.GetInstance().SendNetworkData(telemetry.NewNetworkData(s.host.peerCount(), o.RateIn, o.RateOut))
+		}
+
+	}
+}
+
+func (s *Service) sentBlockIntervalTelemetry() {
+	for {
+		best, err := s.blockState.BestBlockHeader()
+		if err != nil {
+			continue
+		}
+		finalized, err := s.blockState.GetFinalizedHeader(0, 0) //nolint
+		if err != nil {
+			continue
+		}
+
+		telemetry.GetInstance().SendBlockIntervalData(&telemetry.BlockIntervalData{
+			BestHash:           best.Hash(),
+			BestHeight:         best.Number,
+			FinalizedHash:      finalized.Hash(),
+			FinalizedHeight:    finalized.Number,
+			TXCount:            0, // todo (ed) determine where to get tx count
+			UsedStateCacheSize: 0, // todo (ed) determine where to get used_state_cache_size
+		})
+		time.Sleep(s.telemetryInterval)
+	}
+}
+
 func (s *Service) handleConn(conn libp2pnetwork.Conn) {
 	// give new peers a slight weight
 	// TODO: do this once handshake is received
@@ -344,6 +394,13 @@ func (s *Service) Stop() error {
 		logger.Error("Failed to close host", "error", err)
 	}
 
+	defer func() {
+		if r := recover(); r != nil {
+			logger.Error("failed to close telemetry service")
+		}
+	}()
+	close(s.doneNetworkTelemetry)
+
 	return nil
 }
 
@@ -521,6 +578,8 @@ func (s *Service) readStream(stream libp2pnetwork.Stream, peer peer.ID, decoder
 			_ = stream.Close()
 			return
 		}
+
+		s.host.bwc.LogRecvMessage(int64(tot))
 	}
 }
 

dot/telemetry/telemetry.go

@@ -24,16 +24,17 @@ import (
 	"sync"
 	"time"
 
+	"github.com/ChainSafe/gossamer/lib/common"
 	"github.com/ChainSafe/gossamer/lib/genesis"
 	"github.com/gorilla/websocket"
 	log "github.com/sirupsen/logrus"
 )
 
 // Handler struct for holding telemetry related things
 type Handler struct {
-	buf             bytes.Buffer
-	wsConn          []*websocket.Conn
-	telemetryLogger *log.Entry
+	buf    bytes.Buffer
+	wsConn []*websocket.Conn
+	sync.RWMutex
 }
 
 // MyJSONFormatter struct for defining JSON Formatter
@@ -65,13 +66,16 @@ func GetInstance() *Handler {
 				}
 				log.SetOutput(&handlerInstance.buf)
 				log.SetFormatter(new(MyJSONFormatter))
+				go handlerInstance.sender()
 			})
 	}
 	return handlerInstance
 }
 
 // AddConnections adds connections to telemetry sever
 func (h *Handler) AddConnections(conns []*genesis.TelemetryEndpoint) {
+	h.Lock()
+	defer h.Unlock()
 	for _, v := range conns {
 		c, _, err := websocket.DefaultDialer.Dial(v.Endpoint, nil)
 		if err != nil {
@@ -96,29 +100,87 @@ type ConnectionData struct {
 
 // SendConnection sends connection request message to telemetry connection
 func (h *Handler) SendConnection(data *ConnectionData) {
+	h.Lock()
+	defer h.Unlock()
 	payload := log.Fields{"authority": data.Authority, "chain": data.Chain, "config": "", "genesis_hash": data.GenesisHash,
 		"implementation": data.SystemName, "msg": "system.connected", "name": data.NodeName, "network_id": data.NetworkID, "startup_time": data.StartTime,
 		"version": data.SystemVersion}
-	h.telemetryLogger = log.WithFields(log.Fields{"id": 1, "payload": payload, "ts": time.Now()})
-	h.telemetryLogger.Print()
-	h.sendTelemtry()
+	telemetryLogger := log.WithFields(log.Fields{"id": 1, "payload": payload, "ts": time.Now()})
+	telemetryLogger.Print()
 }
 
 // SendBlockImport sends block imported message to telemetry connection
 func (h *Handler) SendBlockImport(bestHash string, height *big.Int) {
+	h.Lock()
+	defer h.Unlock()
 	payload := log.Fields{"best": bestHash, "height": height.Int64(), "msg": "block.import", "origin": "NetworkInitialSync"}
-	h.telemetryLogger = log.WithFields(log.Fields{"id": 1, "payload": payload, "ts": time.Now()})
-	h.telemetryLogger.Print()
-	h.sendTelemtry()
+	telemetryLogger := log.WithFields(log.Fields{"id": 1, "payload": payload, "ts": time.Now()})
+	telemetryLogger.Print()
 }
 
-func (h *Handler) sendTelemtry() {
-	for _, c := range h.wsConn {
-		err := c.WriteMessage(websocket.TextMessage, h.buf.Bytes())
+// NetworkData struct to hold network data telemetry information
+type NetworkData struct {
+	peers   int
+	rateIn  float64
+	rateOut float64
+}
+
+// NewNetworkData creates networkData struct
+func NewNetworkData(peers int, rateIn, rateOut float64) *NetworkData {
+	return &NetworkData{
+		peers:   peers,
+		rateIn:  rateIn,
+		rateOut: rateOut,
+	}
+}
+
+// SendNetworkData send network data system.interval message to telemetry connection
+func (h *Handler) SendNetworkData(data *NetworkData) {
+	h.Lock()
+	defer h.Unlock()
+	payload := log.Fields{"bandwidth_download": data.rateIn, "bandwidth_upload": data.rateOut, "msg": "system.interval", "peers": data.peers}
+	telemetryLogger := log.WithFields(log.Fields{"id": 1, "payload": payload, "ts": time.Now()})
+	telemetryLogger.Print()
+}
+
+// BlockIntervalData struct to hold data for block system.interval message
+type BlockIntervalData struct {
+	BestHash           common.Hash
+	BestHeight         *big.Int
+	FinalizedHash      common.Hash
+	FinalizedHeight    *big.Int
+	TXCount            int
+	UsedStateCacheSize int
+}
+
+// SendBlockIntervalData send block data system interval information to telemetry connection
+func (h *Handler) SendBlockIntervalData(data *BlockIntervalData) {
+	h.Lock()
+	defer h.Unlock()
+	payload := log.Fields{"best": data.BestHash.String(), "finalized_hash": data.FinalizedHash.String(), // nolint
+		"finalized_height": data.FinalizedHeight, "height": data.BestHeight, "msg": "system.interval", "txcount": data.TXCount, // nolint
+		"used_state_cache_size": data.UsedStateCacheSize}
+	telemetryLogger := log.WithFields(log.Fields{"id": 1, "payload": payload, "ts": time.Now()})
+	telemetryLogger.Print()
+}
+
+func (h *Handler) sender() {
+	for {
+		h.RLock()
+		line, err := h.buf.ReadBytes(byte(10))
+		h.RUnlock()
 		if err != nil {
-			// TODO (ed) determine how to handle this error
-			fmt.Printf("ERROR connecting to telemetry %v\n", err)
+			continue
+		}
+
+		for _, c := range h.wsConn {
+			h.Lock()
+			err := c.WriteMessage(websocket.TextMessage, line)
+			h.Unlock()
+			if err != nil {
+				// TODO (ed) determine how to handle this error
+				fmt.Printf("ERROR connecting to telemetry %v\n", err)
+			}
 		}
 	}
-	h.buf.Reset()
 }