metrics.go 26.7 KB
Newer Older
1
// Package metrics provides a set of metrics for the op-node.
2 3 4 5
package metrics

import (
	"context"
6
	"encoding/binary"
7 8 9 10
	"errors"
	"fmt"
	"net"
	"strconv"
11
	"time"
12

13
	ophttp "github.com/ethereum-optimism/optimism/op-node/http"
14
	"github.com/ethereum-optimism/optimism/op-node/p2p/store"
15 16
	"github.com/ethereum-optimism/optimism/op-service/metrics"

17
	pb "github.com/libp2p/go-libp2p-pubsub/pb"
18
	libp2pmetrics "github.com/libp2p/go-libp2p/core/metrics"
19 20 21
	"github.com/prometheus/client_golang/prometheus"
	"github.com/prometheus/client_golang/prometheus/collectors"
	"github.com/prometheus/client_golang/prometheus/promhttp"
22

23
	"github.com/ethereum/go-ethereum"
24
	"github.com/ethereum/go-ethereum/common"
25
	"github.com/ethereum/go-ethereum/rpc"
26 27

	"github.com/ethereum-optimism/optimism/op-node/eth"
28 29 30 31 32 33 34 35 36 37 38
)

const (
	Namespace = "op_node"

	RPCServerSubsystem = "rpc_server"
	RPCClientSubsystem = "rpc_client"

	BatchMethod = "<batch>"
)

39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56
type Metricer interface {
	RecordInfo(version string)
	RecordUp()
	RecordRPCServerRequest(method string) func()
	RecordRPCClientRequest(method string) func(err error)
	RecordRPCClientResponse(method string, err error)
	SetDerivationIdle(status bool)
	RecordPipelineReset()
	RecordSequencingError()
	RecordPublishingError()
	RecordDerivationError()
	RecordReceivedUnsafePayload(payload *eth.ExecutionPayload)
	recordRef(layer string, name string, num uint64, timestamp uint64, h common.Hash)
	RecordL1Ref(name string, ref eth.L1BlockRef)
	RecordL2Ref(name string, ref eth.L2BlockRef)
	RecordUnsafePayloadsBuffer(length uint64, memSize uint64, next eth.BlockID)
	CountSequencedTxs(count int)
	RecordL1ReorgDepth(d uint64)
57 58
	RecordSequencerInconsistentL1Origin(from eth.BlockID, to eth.BlockID)
	RecordSequencerReset()
59 60 61 62 63 64
	RecordGossipEvent(evType int32)
	IncPeerCount()
	DecPeerCount()
	IncStreamCount()
	DecStreamCount()
	RecordBandwidth(ctx context.Context, bwc *libp2pmetrics.BandwidthCounter)
65 66
	RecordSequencerBuildingDiffTime(duration time.Duration)
	RecordSequencerSealingTime(duration time.Duration)
67
	Document() []metrics.DocumentedMetric
68
	RecordChannelInputBytes(num int)
69
	// P2P Metrics
70
	SetPeerScores(allScores []store.PeerScores)
71 72 73
	ClientPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration)
	ServerPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration)
	PayloadsQuarantineSize(n int)
74 75 76 77
	RecordPeerUnban()
	RecordIPUnban()
	RecordDial(allow bool)
	RecordAccept(allow bool)
78 79
}

80
// Metrics tracks all the metrics for the op-node.
81
type Metrics struct {
82 83 84
	Info *prometheus.GaugeVec
	Up   prometheus.Gauge

85 86 87 88 89 90
	RPCServerRequestsTotal          *prometheus.CounterVec
	RPCServerRequestDurationSeconds *prometheus.HistogramVec
	RPCClientRequestsTotal          *prometheus.CounterVec
	RPCClientRequestDurationSeconds *prometheus.HistogramVec
	RPCClientResponsesTotal         *prometheus.CounterVec

91
	L1SourceCache *CacheMetrics
92
	L2SourceCache *CacheMetrics
93

94 95 96 97 98 99 100 101
	DerivationIdle prometheus.Gauge

	PipelineResets   *EventMetrics
	UnsafePayloads   *EventMetrics
	DerivationErrors *EventMetrics
	SequencingErrors *EventMetrics
	PublishingErrors *EventMetrics

102 103 104 105 106 107
	P2PReqDurationSeconds *prometheus.HistogramVec
	P2PReqTotal           *prometheus.CounterVec
	P2PPayloadByNumber    *prometheus.GaugeVec

	PayloadsQuarantineTotal prometheus.Gauge

108 109 110
	SequencerInconsistentL1Origin *EventMetrics
	SequencerResets               *EventMetrics

111 112
	L1RequestDurationSeconds *prometheus.HistogramVec

113 114 115 116 117 118
	SequencerBuildingDiffDurationSeconds prometheus.Histogram
	SequencerBuildingDiffTotal           prometheus.Counter

	SequencerSealingDurationSeconds prometheus.Histogram
	SequencerSealingTotal           prometheus.Counter

119 120 121
	UnsafePayloadsBufferLen     prometheus.Gauge
	UnsafePayloadsBufferMemSize prometheus.Gauge

122 123 124 125 126 127 128 129 130 131
	RefsNumber  *prometheus.GaugeVec
	RefsTime    *prometheus.GaugeVec
	RefsHash    *prometheus.GaugeVec
	RefsSeqNr   *prometheus.GaugeVec
	RefsLatency *prometheus.GaugeVec
	// hash of the last seen block per name, so we don't reduce/increase latency on updates of the same data,
	// and only count the first occurrence
	LatencySeen map[string]common.Hash

	L1ReorgDepth prometheus.Histogram
132

133 134
	TransactionsSequencedTotal prometheus.Counter

135
	// P2P Metrics
136 137 138 139 140 141 142 143 144
	PeerCount         prometheus.Gauge
	StreamCount       prometheus.Gauge
	GossipEventsTotal *prometheus.CounterVec
	BandwidthTotal    *prometheus.GaugeVec
	PeerUnbans        prometheus.Counter
	IPUnbans          prometheus.Counter
	Dials             *prometheus.CounterVec
	Accepts           *prometheus.CounterVec
	PeerScores        *prometheus.HistogramVec
145

146
	ChannelInputBytes prometheus.Counter
147

148
	registry *prometheus.Registry
149
	factory  metrics.Factory
150 151
}

152 153
var _ Metricer = (*Metrics)(nil)

Andreas Bigger's avatar
Andreas Bigger committed
154
// NewMetrics creates a new [Metrics] instance with the given process name.
155 156 157 158 159 160 161 162 163
func NewMetrics(procName string) *Metrics {
	if procName == "" {
		procName = "default"
	}
	ns := Namespace + "_" + procName

	registry := prometheus.NewRegistry()
	registry.MustRegister(collectors.NewProcessCollector(collectors.ProcessCollectorOpts{}))
	registry.MustRegister(collectors.NewGoCollector())
164
	factory := metrics.With(registry)
165

166
	return &Metrics{
167
		Info: factory.NewGaugeVec(prometheus.GaugeOpts{
168 169 170 171 172 173
			Namespace: ns,
			Name:      "info",
			Help:      "Pseudo-metric tracking version and config info",
		}, []string{
			"version",
		}),
174
		Up: factory.NewGauge(prometheus.GaugeOpts{
175 176 177 178
			Namespace: ns,
			Name:      "up",
			Help:      "1 if the op node has finished starting up",
		}),
179

180
		RPCServerRequestsTotal: factory.NewCounterVec(prometheus.CounterOpts{
181 182 183 184 185 186 187
			Namespace: ns,
			Subsystem: RPCServerSubsystem,
			Name:      "requests_total",
			Help:      "Total requests to the RPC server",
		}, []string{
			"method",
		}),
188
		RPCServerRequestDurationSeconds: factory.NewHistogramVec(prometheus.HistogramOpts{
189 190 191 192 193 194 195 196
			Namespace: ns,
			Subsystem: RPCServerSubsystem,
			Name:      "request_duration_seconds",
			Buckets:   []float64{.005, .01, .025, .05, .1, .25, .5, 1, 2.5, 5, 10},
			Help:      "Histogram of RPC server request durations",
		}, []string{
			"method",
		}),
197
		RPCClientRequestsTotal: factory.NewCounterVec(prometheus.CounterOpts{
198 199 200 201 202 203 204
			Namespace: ns,
			Subsystem: RPCClientSubsystem,
			Name:      "requests_total",
			Help:      "Total RPC requests initiated by the opnode's RPC client",
		}, []string{
			"method",
		}),
205
		RPCClientRequestDurationSeconds: factory.NewHistogramVec(prometheus.HistogramOpts{
206 207 208 209 210 211 212 213
			Namespace: ns,
			Subsystem: RPCClientSubsystem,
			Name:      "request_duration_seconds",
			Buckets:   []float64{.005, .01, .025, .05, .1, .25, .5, 1, 2.5, 5, 10},
			Help:      "Histogram of RPC client request durations",
		}, []string{
			"method",
		}),
214
		RPCClientResponsesTotal: factory.NewCounterVec(prometheus.CounterOpts{
215 216 217 218 219 220 221 222
			Namespace: ns,
			Subsystem: RPCClientSubsystem,
			Name:      "responses_total",
			Help:      "Total RPC request responses received by the opnode's RPC client",
		}, []string{
			"method",
			"error",
		}),
223

224 225
		L1SourceCache: NewCacheMetrics(factory, ns, "l1_source_cache", "L1 Source cache"),
		L2SourceCache: NewCacheMetrics(factory, ns, "l2_source_cache", "L2 Source cache"),
226

227
		DerivationIdle: factory.NewGauge(prometheus.GaugeOpts{
228 229 230 231
			Namespace: ns,
			Name:      "derivation_idle",
			Help:      "1 if the derivation pipeline is idle",
		}),
232

233 234 235 236 237
		PipelineResets:   NewEventMetrics(factory, ns, "pipeline_resets", "derivation pipeline resets"),
		UnsafePayloads:   NewEventMetrics(factory, ns, "unsafe_payloads", "unsafe payloads"),
		DerivationErrors: NewEventMetrics(factory, ns, "derivation_errors", "derivation errors"),
		SequencingErrors: NewEventMetrics(factory, ns, "sequencing_errors", "sequencing errors"),
		PublishingErrors: NewEventMetrics(factory, ns, "publishing_errors", "p2p publishing errors"),
238

239 240 241
		SequencerInconsistentL1Origin: NewEventMetrics(factory, ns, "sequencer_inconsistent_l1_origin", "events when the sequencer selects an inconsistent L1 origin"),
		SequencerResets:               NewEventMetrics(factory, ns, "sequencer_resets", "sequencer resets"),

242
		UnsafePayloadsBufferLen: factory.NewGauge(prometheus.GaugeOpts{
243 244 245 246
			Namespace: ns,
			Name:      "unsafe_payloads_buffer_len",
			Help:      "Number of buffered L2 unsafe payloads",
		}),
247
		UnsafePayloadsBufferMemSize: factory.NewGauge(prometheus.GaugeOpts{
248 249 250 251 252
			Namespace: ns,
			Name:      "unsafe_payloads_buffer_mem_size",
			Help:      "Total estimated memory size of buffered L2 unsafe payloads",
		}),

253
		RefsNumber: factory.NewGaugeVec(prometheus.GaugeOpts{
254
			Namespace: ns,
255 256 257 258 259
			Name:      "refs_number",
			Help:      "Gauge representing the different L1/L2 reference block numbers",
		}, []string{
			"layer",
			"type",
260
		}),
261
		RefsTime: factory.NewGaugeVec(prometheus.GaugeOpts{
262
			Namespace: ns,
263 264 265 266 267
			Name:      "refs_time",
			Help:      "Gauge representing the different L1/L2 reference block timestamps",
		}, []string{
			"layer",
			"type",
268
		}),
269
		RefsHash: factory.NewGaugeVec(prometheus.GaugeOpts{
270
			Namespace: ns,
271 272 273 274 275
			Name:      "refs_hash",
			Help:      "Gauge representing the different L1/L2 reference block hashes truncated to float values",
		}, []string{
			"layer",
			"type",
276
		}),
277
		RefsSeqNr: factory.NewGaugeVec(prometheus.GaugeOpts{
278
			Namespace: ns,
279 280 281 282
			Name:      "refs_seqnr",
			Help:      "Gauge representing the different L2 reference sequence numbers",
		}, []string{
			"type",
283
		}),
284
		RefsLatency: factory.NewGaugeVec(prometheus.GaugeOpts{
285
			Namespace: ns,
286 287
			Name:      "refs_latency",
			Help:      "Gauge representing the different L1/L2 reference block timestamps minus current time, in seconds",
288
		}, []string{
289
			"layer",
290 291
			"type",
		}),
292 293
		LatencySeen: make(map[string]common.Hash),

294
		L1ReorgDepth: factory.NewHistogram(prometheus.HistogramOpts{
295 296 297 298 299
			Namespace: ns,
			Name:      "l1_reorg_depth",
			Buckets:   []float64{0.5, 1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5, 9.5, 10.5, 20.5, 50.5, 100.5},
			Help:      "Histogram of L1 Reorg Depths",
		}),
300

301
		TransactionsSequencedTotal: factory.NewGauge(prometheus.GaugeOpts{
302 303 304 305 306
			Namespace: ns,
			Name:      "transactions_sequenced_total",
			Help:      "Count of total transactions sequenced",
		}),

307
		PeerCount: factory.NewGauge(prometheus.GaugeOpts{
308 309 310 311 312
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "peer_count",
			Help:      "Count of currently connected p2p peers",
		}),
313
		PeerScores: factory.NewHistogramVec(prometheus.HistogramOpts{
asnared's avatar
asnared committed
314 315
			Namespace: ns,
			Name:      "peer_scores",
316 317 318
			Help:      "Histogram of currrently connected peer scores",
			Buckets:   []float64{-100, -40, -20, -10, -5, -2, -1, -0.5, -0.05, 0, 0.05, 0.5, 1, 2, 5, 10, 20, 40},
		}, []string{"type"}),
319 320 321 322 323
		StreamCount: factory.NewGauge(prometheus.GaugeOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "stream_count",
			Help:      "Count of currently connected p2p streams",
asnared's avatar
asnared committed
324
		}),
325
		GossipEventsTotal: factory.NewCounterVec(prometheus.CounterOpts{
326 327 328 329 330 331 332
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "gossip_events_total",
			Help:      "Count of gossip events by type",
		}, []string{
			"type",
		}),
333
		BandwidthTotal: factory.NewGaugeVec(prometheus.GaugeOpts{
334 335 336 337 338 339 340
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "bandwidth_bytes_total",
			Help:      "P2P bandwidth by direction",
		}, []string{
			"direction",
		}),
341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364
		PeerUnbans: factory.NewCounter(prometheus.CounterOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "peer_unbans",
			Help:      "Count of peer unbans",
		}),
		IPUnbans: factory.NewCounter(prometheus.CounterOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "ip_unbans",
			Help:      "Count of IP unbans",
		}),
		Dials: factory.NewCounterVec(prometheus.CounterOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "dials",
			Help:      "Count of outgoing dial attempts, with label to filter to allowed attempts",
		}, []string{"allow"}),
		Accepts: factory.NewCounterVec(prometheus.CounterOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "accepts",
			Help:      "Count of incoming dial attempts to accept, with label to filter to allowed attempts",
		}, []string{"allow"}),
365

366
		ChannelInputBytes: factory.NewCounter(prometheus.CounterOpts{
367 368 369 370 371
			Namespace: ns,
			Name:      "channel_input_bytes",
			Help:      "Number of compressed bytes added to the channel",
		}),

372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409
		P2PReqDurationSeconds: factory.NewHistogramVec(prometheus.HistogramOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "req_duration_seconds",
			Buckets:   []float64{},
			Help:      "Duration of P2P requests",
		}, []string{
			"p2p_role", // "client" or "server"
			"p2p_method",
			"result_code",
		}),

		P2PReqTotal: factory.NewCounterVec(prometheus.CounterOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "req_total",
			Help:      "Number of P2P requests",
		}, []string{
			"p2p_role", // "client" or "server"
			"p2p_method",
			"result_code",
		}),

		P2PPayloadByNumber: factory.NewGaugeVec(prometheus.GaugeOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "payload_by_number",
			Help:      "Payload by number requests",
		}, []string{
			"p2p_role", // "client" or "server"
		}),
		PayloadsQuarantineTotal: factory.NewGauge(prometheus.GaugeOpts{
			Namespace: ns,
			Subsystem: "p2p",
			Name:      "payloads_quarantine_total",
			Help:      "number of unverified execution payloads buffered in quarantine",
		}),

410 411 412 413 414 415 416 417
		L1RequestDurationSeconds: factory.NewHistogramVec(prometheus.HistogramOpts{
			Namespace: ns,
			Name:      "l1_request_seconds",
			Buckets: []float64{
				.005, .01, .025, .05, .1, .25, .5, 1, 2.5, 5, 10},
			Help: "Histogram of L1 request time",
		}, []string{"request"}),

418
		SequencerBuildingDiffDurationSeconds: factory.NewHistogram(prometheus.HistogramOpts{
419 420 421 422 423 424 425
			Namespace: ns,
			Name:      "sequencer_building_diff_seconds",
			Buckets: []float64{
				-10, -5, -2.5, -1, -.5, -.25, -.1, -0.05, -0.025, -0.01, -0.005,
				.005, .01, .025, .05, .1, .25, .5, 1, 2.5, 5, 10},
			Help: "Histogram of Sequencer building time, minus block time",
		}),
426
		SequencerBuildingDiffTotal: factory.NewCounter(prometheus.CounterOpts{
427 428 429 430
			Namespace: ns,
			Name:      "sequencer_building_diff_total",
			Help:      "Number of sequencer block building jobs",
		}),
431
		SequencerSealingDurationSeconds: factory.NewHistogram(prometheus.HistogramOpts{
432 433 434 435 436
			Namespace: ns,
			Name:      "sequencer_sealing_seconds",
			Buckets:   []float64{.005, .01, .025, .05, .1, .25, .5, 1, 2.5, 5, 10},
			Help:      "Histogram of Sequencer block sealing time",
		}),
437
		SequencerSealingTotal: factory.NewCounter(prometheus.CounterOpts{
438 439 440 441 442
			Namespace: ns,
			Name:      "sequencer_sealing_total",
			Help:      "Number of sequencer block sealing jobs",
		}),

443
		registry: registry,
444
		factory:  factory,
445 446 447
	}
}

448 449 450
// SetPeerScores updates the peer score metrics.
// Accepts a slice of peer scores in any order.
func (m *Metrics) SetPeerScores(allScores []store.PeerScores) {
451
	for _, scores := range allScores {
452 453 454 455 456 457 458
		m.PeerScores.WithLabelValues("total").Observe(scores.Gossip.Total)
		m.PeerScores.WithLabelValues("ipColocation").Observe(scores.Gossip.IPColocationFactor)
		m.PeerScores.WithLabelValues("behavioralPenalty").Observe(scores.Gossip.BehavioralPenalty)
		m.PeerScores.WithLabelValues("blocksFirstMessage").Observe(scores.Gossip.Blocks.FirstMessageDeliveries)
		m.PeerScores.WithLabelValues("blocksTimeInMesh").Observe(scores.Gossip.Blocks.TimeInMesh)
		m.PeerScores.WithLabelValues("blocksMessageDeliveries").Observe(scores.Gossip.Blocks.MeshMessageDeliveries)
		m.PeerScores.WithLabelValues("blocksInvalidMessageDeliveries").Observe(scores.Gossip.Blocks.InvalidMessageDeliveries)
459 460 461 462

		m.PeerScores.WithLabelValues("reqRespValidResponses").Observe(scores.ReqResp.ValidResponses)
		m.PeerScores.WithLabelValues("reqRespErrorResponses").Observe(scores.ReqResp.ErrorResponses)
		m.PeerScores.WithLabelValues("reqRespRejectedPayloads").Observe(scores.ReqResp.RejectedPayloads)
463 464 465
	}
}

466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524
// RecordInfo sets a pseudo-metric that contains versioning and
// config info for the opnode.
func (m *Metrics) RecordInfo(version string) {
	m.Info.WithLabelValues(version).Set(1)
}

// RecordUp sets the up metric to 1.
func (m *Metrics) RecordUp() {
	prometheus.MustRegister()
	m.Up.Set(1)
}

// RecordRPCServerRequest is a helper method to record an incoming RPC
// call to the opnode's RPC server. It bumps the requests metric,
// and tracks how long it takes to serve a response.
func (m *Metrics) RecordRPCServerRequest(method string) func() {
	m.RPCServerRequestsTotal.WithLabelValues(method).Inc()
	timer := prometheus.NewTimer(m.RPCServerRequestDurationSeconds.WithLabelValues(method))
	return func() {
		timer.ObserveDuration()
	}
}

// RecordRPCClientRequest is a helper method to record an RPC client
// request. It bumps the requests metric, tracks the response
// duration, and records the response's error code.
func (m *Metrics) RecordRPCClientRequest(method string) func(err error) {
	m.RPCClientRequestsTotal.WithLabelValues(method).Inc()
	timer := prometheus.NewTimer(m.RPCClientRequestDurationSeconds.WithLabelValues(method))
	return func(err error) {
		m.RecordRPCClientResponse(method, err)
		timer.ObserveDuration()
	}
}

// RecordRPCClientResponse records an RPC response. It will
// convert the passed-in error into something metrics friendly.
// Nil errors get converted into <nil>, RPC errors are converted
// into rpc_<error code>, HTTP errors are converted into
// http_<status code>, and everything else is converted into
// <unknown>.
func (m *Metrics) RecordRPCClientResponse(method string, err error) {
	var errStr string
	var rpcErr rpc.Error
	var httpErr rpc.HTTPError
	if err == nil {
		errStr = "<nil>"
	} else if errors.As(err, &rpcErr) {
		errStr = fmt.Sprintf("rpc_%d", rpcErr.ErrorCode())
	} else if errors.As(err, &httpErr) {
		errStr = fmt.Sprintf("http_%d", httpErr.StatusCode)
	} else if errors.Is(err, ethereum.NotFound) {
		errStr = "<not found>"
	} else {
		errStr = "<unknown>"
	}
	m.RPCClientResponsesTotal.WithLabelValues(method, errStr).Inc()
}

525 526 527 528 529 530 531 532
func (m *Metrics) SetDerivationIdle(status bool) {
	var val float64
	if status {
		val = 1
	}
	m.DerivationIdle.Set(val)
}

533 534
func (m *Metrics) RecordPipelineReset() {
	m.PipelineResets.RecordEvent()
535 536
}

537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578
func (m *Metrics) RecordSequencingError() {
	m.SequencingErrors.RecordEvent()
}

func (m *Metrics) RecordPublishingError() {
	m.PublishingErrors.RecordEvent()
}

func (m *Metrics) RecordDerivationError() {
	m.DerivationErrors.RecordEvent()
}

func (m *Metrics) RecordReceivedUnsafePayload(payload *eth.ExecutionPayload) {
	m.UnsafePayloads.RecordEvent()
	m.recordRef("l2", "received_payload", uint64(payload.BlockNumber), uint64(payload.Timestamp), payload.BlockHash)
}

func (m *Metrics) recordRef(layer string, name string, num uint64, timestamp uint64, h common.Hash) {
	m.RefsNumber.WithLabelValues(layer, name).Set(float64(num))
	if timestamp != 0 {
		m.RefsTime.WithLabelValues(layer, name).Set(float64(timestamp))
		// only meter the latency when we first see this hash for the given label name
		if m.LatencySeen[name] != h {
			m.LatencySeen[name] = h
			m.RefsLatency.WithLabelValues(layer, name).Set(float64(timestamp) - (float64(time.Now().UnixNano()) / 1e9))
		}
	}
	// we map the first 8 bytes to a float64, so we can graph changes of the hash to find divergences visually.
	// We don't do math.Float64frombits, just a regular conversion, to keep the value within a manageable range.
	m.RefsHash.WithLabelValues(layer, name).Set(float64(binary.LittleEndian.Uint64(h[:])))
}

func (m *Metrics) RecordL1Ref(name string, ref eth.L1BlockRef) {
	m.recordRef("l1", name, ref.Number, ref.Time, ref.Hash)
}

func (m *Metrics) RecordL2Ref(name string, ref eth.L2BlockRef) {
	m.recordRef("l2", name, ref.Number, ref.Time, ref.Hash)
	m.recordRef("l1_origin", name, ref.L1Origin.Number, 0, ref.L1Origin.Hash)
	m.RefsSeqNr.WithLabelValues(name).Set(float64(ref.SequenceNumber))
}

579 580 581 582 583 584
func (m *Metrics) RecordUnsafePayloadsBuffer(length uint64, memSize uint64, next eth.BlockID) {
	m.recordRef("l2", "l2_buffer_unsafe", next.Number, 0, next.Hash)
	m.UnsafePayloadsBufferLen.Set(float64(length))
	m.UnsafePayloadsBufferMemSize.Set(float64(memSize))
}

585 586
func (m *Metrics) CountSequencedTxs(count int) {
	m.TransactionsSequencedTotal.Add(float64(count))
587 588
}

589 590 591 592
func (m *Metrics) RecordL1ReorgDepth(d uint64) {
	m.L1ReorgDepth.Observe(float64(d))
}

593 594 595 596 597 598 599 600 601 602
func (m *Metrics) RecordSequencerInconsistentL1Origin(from eth.BlockID, to eth.BlockID) {
	m.SequencerInconsistentL1Origin.RecordEvent()
	m.recordRef("l1_origin", "inconsistent_from", from.Number, 0, from.Hash)
	m.recordRef("l1_origin", "inconsistent_to", to.Number, 0, to.Hash)
}

func (m *Metrics) RecordSequencerReset() {
	m.SequencerResets.RecordEvent()
}

603 604 605 606
func (m *Metrics) RecordGossipEvent(evType int32) {
	m.GossipEventsTotal.WithLabelValues(pb.TraceEvent_Type_name[evType]).Inc()
}

Matthew Slipper's avatar
Matthew Slipper committed
607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623
func (m *Metrics) IncPeerCount() {
	m.PeerCount.Inc()
}

func (m *Metrics) DecPeerCount() {
	m.PeerCount.Dec()
}

func (m *Metrics) IncStreamCount() {
	m.StreamCount.Inc()
}

func (m *Metrics) DecStreamCount() {
	m.StreamCount.Dec()
}

func (m *Metrics) RecordBandwidth(ctx context.Context, bwc *libp2pmetrics.BandwidthCounter) {
624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
	tick := time.NewTicker(10 * time.Second)
	defer tick.Stop()

	for {
		select {
		case <-tick.C:
			bwTotals := bwc.GetBandwidthTotals()
			m.BandwidthTotal.WithLabelValues("in").Set(float64(bwTotals.TotalIn))
			m.BandwidthTotal.WithLabelValues("out").Set(float64(bwTotals.TotalOut))
		case <-ctx.Done():
			return
		}
	}
}

639 640 641 642 643
// RecordL1RequestTime tracks the amount of time the derivation pipeline spent waiting for L1 data requests.
func (m *Metrics) RecordL1RequestTime(method string, duration time.Duration) {
	m.L1RequestDurationSeconds.WithLabelValues(method).Observe(float64(duration) / float64(time.Second))
}

644 645 646 647 648 649 650 651 652 653 654 655 656 657 658
// RecordSequencerBuildingDiffTime tracks the amount of time the sequencer was allowed between
// start to finish, incl. sealing, minus the block time.
// Ideally this is 0, realistically the sequencer scheduler may be busy with other jobs like syncing sometimes.
func (m *Metrics) RecordSequencerBuildingDiffTime(duration time.Duration) {
	m.SequencerBuildingDiffTotal.Inc()
	m.SequencerBuildingDiffDurationSeconds.Observe(float64(duration) / float64(time.Second))
}

// RecordSequencerSealingTime tracks the amount of time the sequencer took to finish sealing the block.
// Ideally this is 0, realistically it may take some time.
func (m *Metrics) RecordSequencerSealingTime(duration time.Duration) {
	m.SequencerSealingTotal.Inc()
	m.SequencerSealingDurationSeconds.Observe(float64(duration) / float64(time.Second))
}

659 660 661 662
// Serve starts the metrics server on the given hostname and port.
// The server will be closed when the passed-in context is cancelled.
func (m *Metrics) Serve(ctx context.Context, hostname string, port int) error {
	addr := net.JoinHostPort(hostname, strconv.Itoa(port))
663
	server := ophttp.NewHttpServer(promhttp.InstrumentMetricHandler(
664 665 666
		m.registry, promhttp.HandlerFor(m.registry, promhttp.HandlerOpts{}),
	))
	server.Addr = addr
667 668 669 670 671 672
	go func() {
		<-ctx.Done()
		server.Close()
	}()
	return server.ListenAndServe()
}
673

674 675 676 677
func (m *Metrics) Document() []metrics.DocumentedMetric {
	return m.factory.Document()
}

678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698
func (m *Metrics) ClientPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration) {
	if resultCode > 4 { // summarize all high codes to reduce metrics overhead
		resultCode = 5
	}
	code := strconv.FormatUint(uint64(resultCode), 10)
	m.P2PReqTotal.WithLabelValues("client", "payload_by_number", code).Inc()
	m.P2PReqDurationSeconds.WithLabelValues("client", "payload_by_number", code).Observe(float64(duration) / float64(time.Second))
	m.P2PPayloadByNumber.WithLabelValues("client").Set(float64(num))
}

func (m *Metrics) ServerPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration) {
	code := strconv.FormatUint(uint64(resultCode), 10)
	m.P2PReqTotal.WithLabelValues("server", "payload_by_number", code).Inc()
	m.P2PReqDurationSeconds.WithLabelValues("server", "payload_by_number", code).Observe(float64(duration) / float64(time.Second))
	m.P2PPayloadByNumber.WithLabelValues("server").Set(float64(num))
}

func (m *Metrics) PayloadsQuarantineSize(n int) {
	m.PayloadsQuarantineTotal.Set(float64(n))
}

699
func (m *Metrics) RecordChannelInputBytes(inputCompressedBytes int) {
700
	m.ChannelInputBytes.Add(float64(inputCompressedBytes))
701 702
}

703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726
func (m *Metrics) RecordPeerUnban() {
	m.PeerUnbans.Inc()
}

func (m *Metrics) RecordIPUnban() {
	m.IPUnbans.Inc()
}

func (m *Metrics) RecordDial(allow bool) {
	if allow {
		m.Dials.WithLabelValues("true").Inc()
	} else {
		m.Dials.WithLabelValues("false").Inc()
	}
}

func (m *Metrics) RecordAccept(allow bool) {
	if allow {
		m.Accepts.WithLabelValues("true").Inc()
	} else {
		m.Accepts.WithLabelValues("false").Inc()
	}
}

727 728
type noopMetricer struct{}

729
var NoopMetrics Metricer = new(noopMetricer)
730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783

func (n *noopMetricer) RecordInfo(version string) {
}

func (n *noopMetricer) RecordUp() {
}

func (n *noopMetricer) RecordRPCServerRequest(method string) func() {
	return func() {}
}

func (n *noopMetricer) RecordRPCClientRequest(method string) func(err error) {
	return func(err error) {}
}

func (n *noopMetricer) RecordRPCClientResponse(method string, err error) {
}

func (n *noopMetricer) SetDerivationIdle(status bool) {
}

func (n *noopMetricer) RecordPipelineReset() {
}

func (n *noopMetricer) RecordSequencingError() {
}

func (n *noopMetricer) RecordPublishingError() {
}

func (n *noopMetricer) RecordDerivationError() {
}

func (n *noopMetricer) RecordReceivedUnsafePayload(payload *eth.ExecutionPayload) {
}

func (n *noopMetricer) recordRef(layer string, name string, num uint64, timestamp uint64, h common.Hash) {
}

func (n *noopMetricer) RecordL1Ref(name string, ref eth.L1BlockRef) {
}

func (n *noopMetricer) RecordL2Ref(name string, ref eth.L2BlockRef) {
}

func (n *noopMetricer) RecordUnsafePayloadsBuffer(length uint64, memSize uint64, next eth.BlockID) {
}

func (n *noopMetricer) CountSequencedTxs(count int) {
}

func (n *noopMetricer) RecordL1ReorgDepth(d uint64) {
}

784 785 786 787 788 789
func (n *noopMetricer) RecordSequencerInconsistentL1Origin(from eth.BlockID, to eth.BlockID) {
}

func (n *noopMetricer) RecordSequencerReset() {
}

790 791 792
func (n *noopMetricer) RecordGossipEvent(evType int32) {
}

793
func (n *noopMetricer) SetPeerScores(allScores []store.PeerScores) {
794 795
}

796 797 798 799 800 801 802 803 804 805 806 807 808 809
func (n *noopMetricer) IncPeerCount() {
}

func (n *noopMetricer) DecPeerCount() {
}

func (n *noopMetricer) IncStreamCount() {
}

func (n *noopMetricer) DecStreamCount() {
}

func (n *noopMetricer) RecordBandwidth(ctx context.Context, bwc *libp2pmetrics.BandwidthCounter) {
}
810 811 812 813 814 815

func (n *noopMetricer) RecordSequencerBuildingDiffTime(duration time.Duration) {
}

func (n *noopMetricer) RecordSequencerSealingTime(duration time.Duration) {
}
816 817 818 819

func (n *noopMetricer) Document() []metrics.DocumentedMetric {
	return nil
}
820 821 822 823 824 825 826 827 828

func (n *noopMetricer) ClientPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration) {
}

func (n *noopMetricer) ServerPayloadByNumberEvent(num uint64, resultCode byte, duration time.Duration) {
}

func (n *noopMetricer) PayloadsQuarantineSize(int) {
}
829 830 831

func (n *noopMetricer) RecordChannelInputBytes(int) {
}
832 833 834 835 836 837 838 839 840 841 842 843

func (n *noopMetricer) RecordPeerUnban() {
}

func (n *noopMetricer) RecordIPUnban() {
}

func (n *noopMetricer) RecordDial(allow bool) {
}

func (n *noopMetricer) RecordAccept(allow bool) {
}