@aztec/telemetry-client 0.0.0-test.1 → 0.0.1-commit.b655e406

package/src/metrics.ts

@@ -6,9 +6,15 @@
  * @see {@link https://opentelemetry.io/docs/specs/semconv/general/metrics/ | OpenTelemetry Metrics} for naming conventions.
  */
 
+export const BLOB_SINK_STORE_REQUESTS = 'aztec.blob_sink.store_request_count';
+export const BLOB_SINK_RETRIEVE_REQUESTS = 'aztec.blob_sink.retrieve_request_count';
 export const BLOB_SINK_OBJECTS_IN_BLOB_STORE = 'aztec.blob_sink.objects_in_blob_store';
 export const BLOB_SINK_BLOB_SIZE = 'aztec.blob_sink.blob_size';
 
+export const BLOB_SINK_ARCHIVE_BLOB_REQUEST_COUNT = 'aztec.blob_sink.archive.block_request_count';
+export const BLOB_SINK_ARCHIVE_BLOCK_REQUEST_COUNT = 'aztec.blob_sink.archive.blob_request_count';
+export const BLOB_SINK_ARCHIVE_BLOB_COUNT = 'aztec.blob_sink.archive.blob_count';
+
 /** How long it takes to simulate a circuit */
 export const CIRCUIT_SIMULATION_DURATION = 'aztec.circuit.simulation.duration';
 export const CIRCUIT_SIMULATION_INPUT_SIZE = 'aztec.circuit.simulation.input_size';
@@ -30,32 +36,48 @@ export const MEMPOOL_TX_COUNT = 'aztec.mempool.tx_count';
 export const MEMPOOL_TX_SIZE = 'aztec.mempool.tx_size';
 export const DB_NUM_ITEMS = 'aztec.db.num_items';
 export const DB_MAP_SIZE = 'aztec.db.map_size';
+export const DB_PHYSICAL_FILE_SIZE = 'aztec.db.physical_file_size';
 export const DB_USED_SIZE = 'aztec.db.used_size';
 
 export const MEMPOOL_ATTESTATIONS_COUNT = 'aztec.mempool.attestations_count';
 export const MEMPOOL_ATTESTATIONS_SIZE = 'aztec.mempool.attestations_size';
 
-export const ARCHIVER_SYNC_DURATION = 'aztec.archiver.sync_duration';
-export const ARCHIVER_L1_BLOCKS_SYNCED = 'aztec.archiver.l1_blocks_synced';
 export const ARCHIVER_L1_BLOCK_HEIGHT = 'aztec.archiver.l1_block_height';
 export const ARCHIVER_BLOCK_HEIGHT = 'aztec.archiver.block_height';
-export const ARCHIVER_TX_COUNT = 'aztec.archiver.tx_count';
 export const ARCHIVER_ROLLUP_PROOF_DELAY = 'aztec.archiver.rollup_proof_delay';
 export const ARCHIVER_ROLLUP_PROOF_COUNT = 'aztec.archiver.rollup_proof_count';
+
+export const ARCHIVER_MANA_PER_BLOCK = 'aztec.archiver.block.mana_count';
+export const ARCHIVER_TXS_PER_BLOCK = 'aztec.archiver.block.tx_count';
+export const ARCHIVER_SYNC_PER_BLOCK = 'aztec.archiver.block.sync_per_item_duration';
+export const ARCHIVER_SYNC_BLOCK_COUNT = 'aztec.archiver.block.sync_count';
+
+export const ARCHIVER_SYNC_PER_MESSAGE = 'aztec.archiver.message.sync_per_item_duration';
+export const ARCHIVER_SYNC_MESSAGE_COUNT = 'aztec.archiver.message.sync_count';
+
+export const ARCHIVER_PRUNE_DURATION = 'aztec.archiver.prune_duration';
 export const ARCHIVER_PRUNE_COUNT = 'aztec.archiver.prune_count';
 
+export const ARCHIVER_TOTAL_TXS = 'aztec.archiver.tx_count';
+
 export const NODE_RECEIVE_TX_DURATION = 'aztec.node.receive_tx.duration';
 export const NODE_RECEIVE_TX_COUNT = 'aztec.node.receive_tx.count';
 
+export const NODE_SNAPSHOT_DURATION = 'aztec.node.snapshot_duration';
+export const NODE_SNAPSHOT_ERROR_COUNT = 'aztec.node.snapshot_error_count';
+
 export const SEQUENCER_STATE_TRANSITION_BUFFER_DURATION = 'aztec.sequencer.state_transition_buffer.duration';
 export const SEQUENCER_BLOCK_BUILD_DURATION = 'aztec.sequencer.block.build_duration';
 export const SEQUENCER_BLOCK_BUILD_MANA_PER_SECOND = 'aztec.sequencer.block.build_mana_per_second';
 export const SEQUENCER_BLOCK_COUNT = 'aztec.sequencer.block.count';
-export const SEQUENCER_CURRENT_STATE = 'aztec.sequencer.current.state';
-export const SEQUENCER_CURRENT_BLOCK_NUMBER = 'aztec.sequencer.current.block_number';
-export const SEQUENCER_CURRENT_BLOCK_SIZE = 'aztec.sequencer.current.block_size';
-export const SEQUENCER_TIME_TO_COLLECT_ATTESTATIONS = 'aztec.sequencer.time_to_collect_attestations';
-export const SEQUENCER_BLOCK_BUILD_INSERTION_TIME = 'aztec.sequencer.block_builder_tree_insertion_duration';
+export const SEQUENCER_CURRENT_BLOCK_REWARDS = 'aztec.sequencer.current_block_rewards';
+export const SEQUENCER_SLOT_COUNT = 'aztec.sequencer.slot.total_count';
+export const SEQUENCER_FILLED_SLOT_COUNT = 'aztec.sequencer.slot.filled_count';
+
+export const SEQUENCER_COLLECTED_ATTESTATIONS_COUNT = 'aztec.sequencer.attestations.collected_count';
+export const SEQUENCER_REQUIRED_ATTESTATIONS_COUNT = 'aztec.sequencer.attestations.required_count';
+export const SEQUENCER_COLLECT_ATTESTATIONS_DURATION = 'aztec.sequencer.attestations.collect_duration';
+export const SEQUENCER_COLLECT_ATTESTATIONS_TIME_ALLOWANCE = 'aztec.sequencer.attestations.collect_allowance';
 
 export const L1_PUBLISHER_GAS_PRICE = 'aztec.l1_publisher.gas_price';
 export const L1_PUBLISHER_TX_COUNT = 'aztec.l1_publisher.tx_count';
@@ -70,15 +92,48 @@ export const L1_PUBLISHER_BLOB_INCLUSION_BLOCKS = 'aztec.l1_publisher.blob_inclu
 export const L1_PUBLISHER_BLOB_TX_SUCCESS = 'aztec.l1_publisher.blob_tx_success';
 export const L1_PUBLISHER_BLOB_TX_FAILURE = 'aztec.l1_publisher.blob_tx_failure';
 export const L1_PUBLISHER_BALANCE = 'aztec.l1_publisher.balance';
+export const L1_PUBLISHER_TX_TOTAL_FEE = 'aztec.l1_publisher.tx_total_fee';
+
+export const L1_BLOCK_HEIGHT = 'aztec.l1.block_height';
+export const L1_BALANCE_ETH = 'aztec.l1.balance';
+export const L1_GAS_PRICE_WEI = 'aztec.l1.gas_price';
+export const L1_BLOB_BASE_FEE_WEI = 'aztec.l1.blob_base_fee';
+
+export const L1_TX_MINED_DURATION = 'aztec.l1_tx.mined_duration';
+export const L1_TX_MINED_COUNT = 'aztec.l1_tx.mined_count';
+export const L1_TX_REVERTED_COUNT = 'aztec.l1_tx.reverted_count';
+export const L1_TX_CANCELLED_COUNT = 'aztec.l1_tx.cancelled_count';
+export const L1_TX_NOT_MINED_COUNT = 'aztec.l1_tx.not_mined_count';
+export const L1_TX_ATTEMPTS_UNTIL_MINED = 'aztec.l1_tx.attempts_until_mined';
+export const L1_TX_MAX_PRIORITY_FEE = 'aztec.l1_tx.max_priority_fee';
+export const L1_TX_MAX_FEE = 'aztec.l1_tx.max_fee';
+export const L1_TX_BLOB_FEE = 'aztec.l1_tx.blob_fee';
 
 export const PEER_MANAGER_GOODBYES_SENT = 'aztec.peer_manager.goodbyes_sent';
 export const PEER_MANAGER_GOODBYES_RECEIVED = 'aztec.peer_manager.goodbyes_received';
+export const PEER_MANAGER_PEER_COUNT = 'aztec.peer_manager.peer_count';
 
 export const P2P_REQ_RESP_SENT_REQUESTS = 'aztec.p2p.req_resp.sent_requests';
 export const P2P_REQ_RESP_RECEIVED_REQUESTS = 'aztec.p2p.req_resp.received_requests';
 export const P2P_REQ_RESP_FAILED_OUTBOUND_REQUESTS = 'aztec.p2p.req_resp.failed_outbound_requests';
 export const P2P_REQ_RESP_FAILED_INBOUND_REQUESTS = 'aztec.p2p.req_resp.failed_inbound_requests';
 
+export const P2P_GOSSIP_MESSAGE_VALIDATION_DURATION = 'aztec.p2p.gossip.message_validation_duration';
+export const P2P_GOSSIP_MESSAGE_PREVALIDATION_COUNT = 'aztec.p2p.gossip.message_validation_count';
+export const P2P_GOSSIP_MESSAGE_LATENCY = 'aztec.p2p.gossip.message_latency';
+
+export const P2P_GOSSIP_AGG_MESSAGE_LATENCY_MIN = 'aztec.p2p.gossip.agg_message_latency_min';
+export const P2P_GOSSIP_AGG_MESSAGE_LATENCY_MAX = 'aztec.p2p.gossip.agg_message_latency_max';
+export const P2P_GOSSIP_AGG_MESSAGE_LATENCY_P50 = 'aztec.p2p.gossip.agg_message_latency_p50';
+export const P2P_GOSSIP_AGG_MESSAGE_LATENCY_P90 = 'aztec.p2p.gossip.agg_message_latency_p90';
+export const P2P_GOSSIP_AGG_MESSAGE_LATENCY_AVG = 'aztec.p2p.gossip.agg_message_latency_avg';
+
+export const P2P_GOSSIP_AGG_MESSAGE_VALIDATION_DURATION_MIN = 'aztec.p2p.gossip.agg_message_validation_duration_min';
+export const P2P_GOSSIP_AGG_MESSAGE_VALIDATION_DURATION_MAX = 'aztec.p2p.gossip.agg_message_validation_duration_max';
+export const P2P_GOSSIP_AGG_MESSAGE_VALIDATION_DURATION_P50 = 'aztec.p2p.gossip.agg_message_validation_duration_p50';
+export const P2P_GOSSIP_AGG_MESSAGE_VALIDATION_DURATION_P90 = 'aztec.p2p.gossip.agg_message_validation_duration_p90';
+export const P2P_GOSSIP_AGG_MESSAGE_VALIDATION_DURATION_AVG = 'aztec.p2p.gossip.agg_message_validation_duration_avg';
+
 export const PUBLIC_PROCESSOR_TX_DURATION = 'aztec.public_processor.tx_duration';
 export const PUBLIC_PROCESSOR_TX_COUNT = 'aztec.public_processor.tx_count';
 export const PUBLIC_PROCESSOR_TX_PHASE_COUNT = 'aztec.public_processor.tx_phase_count';
@@ -91,15 +146,21 @@ export const PUBLIC_PROCESSOR_TOTAL_GAS_HISTOGRAM = 'aztec.public_processor.tota
 export const PUBLIC_PROCESSOR_GAS_RATE = 'aztec.public_processor.gas_rate';
 export const PUBLIC_PROCESSOR_TREE_INSERTION = 'aztec.public_processor.tree_insertion';
 
+export const PUBLIC_EXECUTOR_PREFIX = 'aztec.public_executor.';
 export const PUBLIC_EXECUTOR_SIMULATION_COUNT = 'aztec.public_executor.simulation_count';
 export const PUBLIC_EXECUTOR_SIMULATION_DURATION = 'aztec.public_executor.simulation_duration';
 export const PUBLIC_EXECUTOR_SIMULATION_MANA_PER_SECOND = 'aztec.public_executor.simulation_mana_per_second';
-export const PUBLIC_EXECUTION_SIMULATION_BYTECODE_SIZE = 'aztec.public_executor.simulation_bytecode_size';
-export const PUBLIC_EXECUTION_PRIVATE_EFFECTS_INSERTION = 'aztec.public_executor.private_effects_insertion';
+export const PUBLIC_EXECUTOR_SIMULATION_MANA_USED = 'aztec.public_executor.simulation_mana_used';
+export const PUBLIC_EXECUTOR_SIMULATION_TOTAL_INSTRUCTIONS = 'aztec.public_executor.simulation_total_instructions';
+export const PUBLIC_EXECUTOR_TX_HASHING = 'aztec.public_executor.tx_hashing';
+export const PUBLIC_EXECUTOR_PRIVATE_EFFECTS_INSERTION = 'aztec.public_executor.private_effects_insertion';
+export const PUBLIC_EXECUTOR_SIMULATION_BYTECODE_SIZE = 'aztec.public_executor.simulation_bytecode_size';
 
 export const PROVING_ORCHESTRATOR_BASE_ROLLUP_INPUTS_DURATION =
   'aztec.proving_orchestrator.base_rollup.inputs_duration';
 
+export const PROVING_ORCHESTRATOR_AVM_FALLBACK_COUNT = 'aztec.proving_orchestrator.avm.fallback_count';
+
 export const PROVING_QUEUE_JOB_SIZE = 'aztec.proving_queue.job_size';
 export const PROVING_QUEUE_SIZE = 'aztec.proving_queue.size';
 export const PROVING_QUEUE_TOTAL_JOBS = 'aztec.proving_queue.enqueued_jobs_count';
@@ -121,15 +182,18 @@ export const PROVER_NODE_EXECUTION_DURATION = 'aztec.prover_node.execution.durat
 export const PROVER_NODE_JOB_DURATION = 'aztec.prover_node.job_duration';
 export const PROVER_NODE_JOB_BLOCKS = 'aztec.prover_node.job_blocks';
 export const PROVER_NODE_JOB_TRANSACTIONS = 'aztec.prover_node.job_transactions';
+export const PROVER_NODE_REWARDS_TOTAL = 'aztec.prover_node.rewards_total';
+export const PROVER_NODE_REWARDS_PER_EPOCH = 'aztec.prover_node.rewards_per_epoch';
 
 export const WORLD_STATE_FORK_DURATION = 'aztec.world_state.fork.duration';
 export const WORLD_STATE_SYNC_DURATION = 'aztec.world_state.sync.duration';
 export const WORLD_STATE_MERKLE_TREE_SIZE = 'aztec.world_state.merkle_tree_size';
 export const WORLD_STATE_DB_SIZE = 'aztec.world_state.db_size';
 export const WORLD_STATE_DB_MAP_SIZE = 'aztec.world_state.db_map_size';
+export const WORLD_STATE_DB_PHYSICAL_SIZE = 'aztec.world_state.db_physical_size';
 export const WORLD_STATE_TREE_SIZE = 'aztec.world_state.tree_size';
-export const WORLD_STATE_UNFINALISED_HEIGHT = 'aztec.world_state.unfinalised_height';
-export const WORLD_STATE_FINALISED_HEIGHT = 'aztec.world_state.finalised_height';
+export const WORLD_STATE_UNFINALIZED_HEIGHT = 'aztec.world_state.unfinalized_height';
+export const WORLD_STATE_FINALIZED_HEIGHT = 'aztec.world_state.finalized_height';
 export const WORLD_STATE_OLDEST_BLOCK = 'aztec.world_state.oldest_block';
 export const WORLD_STATE_DB_USED_SIZE = 'aztec.world_state.db_used_size';
 export const WORLD_STATE_DB_NUM_ITEMS = 'aztec.world_state.db_num_items';
@@ -139,7 +203,13 @@ export const WORLD_STATE_CRITICAL_ERROR_COUNT = 'aztec.world_state.critical_erro
 export const PROOF_VERIFIER_COUNT = 'aztec.proof_verifier.count';
 
 export const VALIDATOR_RE_EXECUTION_TIME = 'aztec.validator.re_execution_time';
+export const VALIDATOR_RE_EXECUTION_MANA = 'aztec.validator.re_execution_mana';
+export const VALIDATOR_RE_EXECUTION_TX_COUNT = 'aztec.validator.re_execution_tx_count';
+
 export const VALIDATOR_FAILED_REEXECUTION_COUNT = 'aztec.validator.failed_reexecution_count';
+export const VALIDATOR_ATTESTATION_SUCCESS_COUNT = 'aztec.validator.attestation_success_count';
+export const VALIDATOR_ATTESTATION_FAILED_BAD_PROPOSAL_COUNT = 'aztec.validator.attestation_failed_bad_proposal_count';
+export const VALIDATOR_ATTESTATION_FAILED_NODE_ISSUE_COUNT = 'aztec.validator.attestation_failed_node_issue_count';
 
 export const NODEJS_EVENT_LOOP_DELAY_MIN = 'nodejs.eventloop.delay.min';
 export const NODEJS_EVENT_LOOP_DELAY_MEAN = 'nodejs.eventloop.delay.mean';
@@ -151,3 +221,27 @@ export const NODEJS_EVENT_LOOP_DELAY_P99 = 'nodejs.eventloop.delay.p99';
 
 export const NODEJS_EVENT_LOOP_UTILIZATION = 'nodejs.eventloop.utilization';
 export const NODEJS_EVENT_LOOP_TIME = 'nodejs.eventloop.time';
+
+export const NODEJS_MEMORY_HEAP_USAGE = 'nodejs.memory.v8_heap.usage';
+export const NODEJS_MEMORY_HEAP_TOTAL = 'nodejs.memory.v8_heap.total';
+export const NODEJS_MEMORY_NATIVE_USAGE = 'nodejs.memory.native.usage';
+export const NODEJS_MEMORY_BUFFER_USAGE = 'nodejs.memory.array_buffer.usage';
+
+export const TX_PROVIDER_TXS_FROM_PROPOSALS_COUNT = 'aztec.tx_collector.txs_from_proposal_count';
+export const TX_PROVIDER_TXS_FROM_MEMPOOL_COUNT = 'aztec.tx_collector.txs_from_mempool_count';
+export const TX_PROVIDER_TXS_FROM_P2P_COUNT = 'aztec.tx_collector.txs_from_p2p_count';
+export const TX_PROVIDER_MISSING_TXS_COUNT = 'aztec.tx_collector.missing_txs_count';
+
+export const TX_COLLECTOR_COUNT = 'aztec.tx_collector.tx_count';
+export const TX_COLLECTOR_DURATION_PER_REQUEST = 'aztec.tx_collector.duration_per_request';
+export const TX_COLLECTOR_DURATION_PER_TX = 'aztec.tx_collector.duration_per_tx';
+
+export const IVC_VERIFIER_TIME = 'aztec.ivc_verifier.time';
+export const IVC_VERIFIER_TOTAL_TIME = 'aztec.ivc_verifier.total_time';
+export const IVC_VERIFIER_FAILURE_COUNT = 'aztec.ivc_verifier.failure_count';
+
+export const IVC_VERIFIER_AGG_DURATION_MIN = 'aztec.ivc_verifier.agg_duration_min';
+export const IVC_VERIFIER_AGG_DURATION_MAX = 'aztec.ivc_verifier.agg_duration_max';
+export const IVC_VERIFIER_AGG_DURATION_P50 = 'aztec.ivc_verifier.agg_duration_p50';
+export const IVC_VERIFIER_AGG_DURATION_P90 = 'aztec.ivc_verifier.agg_duration_p90';
+export const IVC_VERIFIER_AGG_DURATION_AVG = 'aztec.ivc_verifier.agg_duration_avg';

package/src/{event_loop_monitor.ts → nodejs_metrics_monitor.ts}

@@ -1,3 +1,4 @@
+import type { Observable } from '@opentelemetry/api';
 import { type EventLoopUtilization, type IntervalHistogram, monitorEventLoopDelay, performance } from 'node:perf_hooks';
 
 import * as Attributes from './attributes.js';
@@ -13,7 +14,7 @@ import {
 /**
  * Detector for custom Aztec attributes
  */
-export class EventLoopMonitor {
+export class NodejsMetricsMonitor {
   private eventLoopDelayGauges: {
     min: ObservableGauge;
     max: ObservableGauge;
@@ -24,6 +25,10 @@ export class EventLoopMonitor {
     p99: ObservableGauge;
   };
 
+  // skip `rss` because that's already tracked by @opentelemetry/host-metrics
+  // description of each field here https://nodejs.org/api/process.html#processmemoryusage
+  private memoryGauges: Record<Exclude<keyof NodeJS.MemoryUsage, 'rss'>, ObservableGauge>;
+
   private eventLoopUilization: ObservableGauge;
   private eventLoopTime: UpDownCounter;
 
@@ -62,6 +67,25 @@ export class EventLoopMonitor {
     });
 
     this.eventLoopDelay = monitorEventLoopDelay();
+
+    this.memoryGauges = {
+      heapUsed: meter.createObservableGauge(Metrics.NODEJS_MEMORY_HEAP_USAGE, {
+        unit: 'By',
+        description: 'Memory used by the V8 heap',
+      }),
+      heapTotal: meter.createObservableGauge(Metrics.NODEJS_MEMORY_HEAP_TOTAL, {
+        unit: 'By',
+        description: 'The max size the V8 heap can grow to',
+      }),
+      arrayBuffers: meter.createObservableGauge(Metrics.NODEJS_MEMORY_BUFFER_USAGE, {
+        unit: 'By',
+        description: 'Memory allocated for buffers (includes native memory used)',
+      }),
+      external: meter.createObservableGauge(Metrics.NODEJS_MEMORY_NATIVE_USAGE, {
+        unit: 'By',
+        description: 'Memory allocated for native C++ objects',
+      }),
+    };
   }
 
   start(): void {
@@ -74,6 +98,7 @@ export class EventLoopMonitor {
     this.meter.addBatchObservableCallback(this.measure, [
       this.eventLoopUilization,
       ...Object.values(this.eventLoopDelayGauges),
+      ...Object.values(this.memoryGauges),
     ]);
   }
 
@@ -84,6 +109,7 @@ export class EventLoopMonitor {
     this.meter.removeBatchObservableCallback(this.measure, [
       this.eventLoopUilization,
       ...Object.values(this.eventLoopDelayGauges),
+      ...Object.values(this.memoryGauges),
     ]);
     this.eventLoopDelay.disable();
     this.eventLoopDelay.reset();
@@ -91,6 +117,20 @@ export class EventLoopMonitor {
   }
 
   private measure = (obs: BatchObservableResult): void => {
+    this.measureMemoryUsage(obs);
+    this.measureEventLoopDelay(obs);
+  };
+
+  private measureMemoryUsage = (observer: BatchObservableResult) => {
+    const mem = process.memoryUsage();
+
+    observer.observe(this.memoryGauges.heapUsed, mem.heapUsed);
+    observer.observe(this.memoryGauges.heapTotal, mem.heapTotal);
+    observer.observe(this.memoryGauges.arrayBuffers, mem.arrayBuffers);
+    observer.observe(this.memoryGauges.external, mem.external);
+  };
+
+  private measureEventLoopDelay = (obs: BatchObservableResult): void => {
     const newELU = performance.eventLoopUtilization();
     const delta = performance.eventLoopUtilization(newELU, this.lastELU);
     this.lastELU = newELU;
@@ -103,17 +143,26 @@ export class EventLoopMonitor {
     // - https://youtu.be/WetXnEPraYM
     obs.observe(this.eventLoopUilization, delta.utilization);
 
-    this.eventLoopTime.add(Math.floor(delta.idle), { [Attributes.NODEJS_EVENT_LOOP_STATE]: 'idle' });
-    this.eventLoopTime.add(Math.floor(delta.active), { [Attributes.NODEJS_EVENT_LOOP_STATE]: 'active' });
+    this.eventLoopTime.add(Math.trunc(delta.idle), { [Attributes.NODEJS_EVENT_LOOP_STATE]: 'idle' });
+    this.eventLoopTime.add(Math.trunc(delta.active), { [Attributes.NODEJS_EVENT_LOOP_STATE]: 'active' });
 
-    obs.observe(this.eventLoopDelayGauges.min, Math.floor(this.eventLoopDelay.min));
-    obs.observe(this.eventLoopDelayGauges.mean, Math.floor(this.eventLoopDelay.mean));
-    obs.observe(this.eventLoopDelayGauges.max, Math.floor(this.eventLoopDelay.max));
-    obs.observe(this.eventLoopDelayGauges.stddev, Math.floor(this.eventLoopDelay.stddev));
-    obs.observe(this.eventLoopDelayGauges.p50, Math.floor(this.eventLoopDelay.percentile(50)));
-    obs.observe(this.eventLoopDelayGauges.p90, Math.floor(this.eventLoopDelay.percentile(90)));
-    obs.observe(this.eventLoopDelayGauges.p99, Math.floor(this.eventLoopDelay.percentile(99)));
+    safeObserveInt(obs, this.eventLoopDelayGauges.min, this.eventLoopDelay.min);
+    safeObserveInt(obs, this.eventLoopDelayGauges.mean, this.eventLoopDelay.mean);
+    safeObserveInt(obs, this.eventLoopDelayGauges.max, this.eventLoopDelay.max);
+    safeObserveInt(obs, this.eventLoopDelayGauges.stddev, this.eventLoopDelay.stddev);
+    safeObserveInt(obs, this.eventLoopDelayGauges.p50, this.eventLoopDelay.percentile(50));
+    safeObserveInt(obs, this.eventLoopDelayGauges.p90, this.eventLoopDelay.percentile(90));
+    safeObserveInt(obs, this.eventLoopDelayGauges.p99, this.eventLoopDelay.percentile(99));
 
     this.eventLoopDelay.reset();
   };
 }
+
+function safeObserveInt(observer: BatchObservableResult, metric: Observable, value: number, attrs?: object) {
+  // discard NaN, Infinity, -Infinity
+  if (!Number.isFinite(value)) {
+    return;
+  }
+
+  observer.observe(metric, Math.trunc(value), attrs);
+}

package/src/noop.ts

@@ -3,6 +3,9 @@ import { type Meter, type Span, type SpanContext, type Tracer, createNoopMeter }
 import type { TelemetryClient } from './telemetry.js';
 
 export class NoopTelemetryClient implements TelemetryClient {
+  setExportedPublicTelemetry(_prefixes: string[]): void {}
+  setPublicTelemetryCollectFrom(_roles: string[]): void {}
+
   getMeter(): Meter {
     return createNoopMeter();
   }
@@ -32,7 +35,7 @@ export class NoopTracer implements Tracer {
     return new NoopSpan();
   }
 
-  startActiveSpan<F extends (...args: any[]) => any>(_name: string, ...args: (unknown | F)[]): ReturnType<F> {
+  startActiveSpan<F extends (...args: any[]) => any>(_name: string, ...args: unknown[]): ReturnType<F> {
     // there are three different signatures for startActiveSpan, grab the function, we don't care about the rest
     const fn = args.find(arg => typeof arg === 'function') as F;
     return fn(new NoopSpan());

package/src/otel.ts

@@ -28,8 +28,8 @@ import { BatchSpanProcessor, NodeTracerProvider } from '@opentelemetry/sdk-trace
 import { ATTR_SERVICE_NAME, ATTR_SERVICE_VERSION } from '@opentelemetry/semantic-conventions';
 
 import type { TelemetryClientConfig } from './config.js';
-import { EventLoopMonitor } from './event_loop_monitor.js';
-import { OtelFilterMetricExporter } from './otel_filter_metric_exporter.js';
+import { NodejsMetricsMonitor } from './nodejs_metrics_monitor.js';
+import { OtelFilterMetricExporter, PublicOtelFilterMetricExporter } from './otel_filter_metric_exporter.js';
 import { registerOtelLoggerProvider } from './otel_logger_provider.js';
 import { getOtelResource } from './otel_resource.js';
 import type { TelemetryClient } from './telemetry.js';
@@ -38,7 +38,7 @@ export type OpenTelemetryClientFactory = (resource: IResource, log: Logger) => O
 
 export class OpenTelemetryClient implements TelemetryClient {
   hostMetrics: HostMetrics | undefined;
-  eventLoopMonitor: EventLoopMonitor | undefined;
+  nodejsMetricsMonitor: NodejsMetricsMonitor | undefined;
   private meters: Map<string, Meter> = new Map<string, Meter>();
   private tracers: Map<string, Tracer> = new Map<string, Tracer>();
 
@@ -47,9 +47,18 @@ export class OpenTelemetryClient implements TelemetryClient {
     private meterProvider: MeterProvider,
     private traceProvider: TracerProvider,
     private loggerProvider: LoggerProvider | undefined,
+    private publicMetricExporter: PublicOtelFilterMetricExporter | undefined,
     private log: Logger,
   ) {}
 
+  setExportedPublicTelemetry(metrics: string[]): void {
+    this.publicMetricExporter?.setMetricPrefixes(metrics);
+  }
+
+  setPublicTelemetryCollectFrom(roles: string[]): void {
+    this.publicMetricExporter?.setAllowedRoles(roles);
+  }
+
   getMeter(name: string): Meter {
     let meter = this.meters.get(name);
     if (!meter) {
@@ -91,12 +100,12 @@ export class OpenTelemetryClient implements TelemetryClient {
       meterProvider: this.meterProvider,
     });
 
-    this.eventLoopMonitor = new EventLoopMonitor(
+    this.nodejsMetricsMonitor = new NodejsMetricsMonitor(
       this.meterProvider.getMeter(this.resource.attributes[ATTR_SERVICE_NAME] as string),
     );
 
     this.hostMetrics.start();
-    this.eventLoopMonitor.start();
+    this.nodejsMetricsMonitor.start();
   }
 
   public isEnabled() {
@@ -112,7 +121,7 @@ export class OpenTelemetryClient implements TelemetryClient {
   }
 
   public async stop() {
-    this.eventLoopMonitor?.stop();
+    this.nodejsMetricsMonitor?.stop();
 
     const flushAndShutdown = async (provider?: { forceFlush: () => Promise<void>; shutdown: () => Promise<void> }) => {
       if (!provider) {
@@ -131,16 +140,31 @@ export class OpenTelemetryClient implements TelemetryClient {
 
   public static createMeterProvider(
     resource: IResource,
-    options: Partial<PeriodicExportingMetricReaderOptions>,
+    exporters: Array<PeriodicExportingMetricReaderOptions>,
   ): MeterProvider {
     return new MeterProvider({
       resource,
-      readers: options.exporter
-        ? [new PeriodicExportingMetricReader(options as PeriodicExportingMetricReaderOptions)]
-        : [],
+      readers: exporters.map(options => new PeriodicExportingMetricReader(options)),
 
       views: [
         // Every histogram matching the selector (type + unit) gets these custom buckets assigned
+        new View({
+          instrumentType: InstrumentType.HISTOGRAM,
+          instrumentUnit: 'Mmana',
+          aggregation: new ExplicitBucketHistogramAggregation(
+            [0.1, 0.5, 1, 2, 4, 8, 10, 25, 50, 100, 500, 1000, 5000, 10000],
+            true,
+          ),
+        }),
+        new View({
+          instrumentType: InstrumentType.HISTOGRAM,
+          instrumentUnit: 'tx',
+          aggregation: new ExplicitBucketHistogramAggregation(
+            // TPS
+            [0.1 * 36, 0.2 * 36, 0.5 * 36, 1 * 36, 2 * 36, 5 * 36, 10 * 36, 15 * 36].map(Math.ceil),
+            true,
+          ),
+        }),
         new View({
           instrumentType: InstrumentType.HISTOGRAM,
           instrumentUnit: 's',
@@ -257,20 +281,41 @@ export class OpenTelemetryClient implements TelemetryClient {
 
       tracerProvider.register();
 
-      const meterProvider = OpenTelemetryClient.createMeterProvider(resource, {
-        exporter: config.metricsCollectorUrl
-          ? new OtelFilterMetricExporter(
-              new OTLPMetricExporter({ url: config.metricsCollectorUrl.href }),
-              config.otelExcludeMetrics ?? [],
-            )
-          : undefined,
-        exportTimeoutMillis: config.otelExportTimeoutMs,
-        exportIntervalMillis: config.otelCollectIntervalMs,
-      });
+      const exporters: PeriodicExportingMetricReaderOptions[] = [];
+      if (config.metricsCollectorUrl) {
+        exporters.push({
+          exporter: new OtelFilterMetricExporter(
+            new OTLPMetricExporter({ url: config.metricsCollectorUrl.href }),
+            config.otelExcludeMetrics,
+            'deny',
+          ),
+          exportTimeoutMillis: config.otelExportTimeoutMs,
+          exportIntervalMillis: config.otelCollectIntervalMs,
+        });
+      }
+
+      let publicExporter: PublicOtelFilterMetricExporter | undefined;
+      if (config.publicMetricsCollectorUrl && !config.publicMetricsOptOut) {
+        log.info(`Exporting public metrics: ${config.publicIncludeMetrics}`, {
+          publicMetrics: config.publicIncludeMetrics,
+          collectorUrl: config.publicMetricsCollectorUrl,
+        });
+        publicExporter = new PublicOtelFilterMetricExporter(
+          config.publicMetricsCollectFrom,
+          new OTLPMetricExporter({ url: config.publicMetricsCollectorUrl.href }),
+          config.publicIncludeMetrics,
+        );
+        exporters.push({
+          exporter: publicExporter,
+          exportTimeoutMillis: config.otelExportTimeoutMs,
+          exportIntervalMillis: config.otelCollectIntervalMs,
+        });
+      }
 
+      const meterProvider = OpenTelemetryClient.createMeterProvider(resource, exporters);
       const loggerProvider = registerOtelLoggerProvider(resource, config.logsCollectorUrl);
 
-      return new OpenTelemetryClient(resource, meterProvider, tracerProvider, loggerProvider, log);
+      return new OpenTelemetryClient(resource, meterProvider, tracerProvider, loggerProvider, publicExporter, log);
     };
   }
 

package/src/otel_filter_metric_exporter.ts

@@ -1,8 +1,14 @@
-import type { ExportResult } from '@opentelemetry/core';
+import { type ExportResult, ExportResultCode } from '@opentelemetry/core';
 import type { MetricData, PushMetricExporter, ResourceMetrics } from '@opentelemetry/sdk-metrics';
 
+import { AZTEC_NODE_ROLE } from './attributes.js';
+
 export class OtelFilterMetricExporter implements PushMetricExporter {
-  constructor(private readonly exporter: PushMetricExporter, private readonly excludeMetricPrefixes: string[]) {
+  constructor(
+    private readonly exporter: PushMetricExporter,
+    private metricPrefix: string[],
+    private readonly filter: 'allow' | 'deny' = 'deny',
+  ) {
     if (exporter.selectAggregation) {
       (this as PushMetricExporter).selectAggregation = exporter.selectAggregation.bind(exporter);
     }
@@ -23,9 +29,17 @@ export class OtelFilterMetricExporter implements PushMetricExporter {
   }
 
   private filterMetrics(metrics: MetricData[]): MetricData[] {
-    return metrics.filter(
-      metric => !this.excludeMetricPrefixes.some(prefix => metric.descriptor.name.startsWith(prefix)),
-    );
+    return metrics.filter(metric => {
+      const matched = this.metricPrefix.some(prefix => metric.descriptor.name.startsWith(prefix));
+
+      if (this.filter === 'deny') {
+        return !matched;
+      }
+
+      if (this.filter === 'allow') {
+        return matched;
+      }
+    });
   }
 
   public forceFlush(): Promise<void> {
@@ -35,4 +49,32 @@ export class OtelFilterMetricExporter implements PushMetricExporter {
   public shutdown(): Promise<void> {
     return this.exporter.shutdown();
   }
+
+  public setMetricPrefixes(metrics: string[]) {
+    this.metricPrefix = metrics;
+  }
+}
+
+export class PublicOtelFilterMetricExporter extends OtelFilterMetricExporter {
+  constructor(
+    private allowedRoles: string[],
+    exporter: PushMetricExporter,
+    metricPrefix: string[],
+  ) {
+    super(exporter, metricPrefix, 'allow');
+  }
+
+  public override export(metrics: ResourceMetrics, resultCallback: (result: ExportResult) => void): void {
+    const role = String(metrics.resource.attributes[AZTEC_NODE_ROLE] ?? '');
+    if (!role || !this.allowedRoles.includes(role)) {
+      // noop
+      return resultCallback({ code: ExportResultCode.SUCCESS });
+    }
+
+    super.export(metrics, resultCallback);
+  }
+
+  public setAllowedRoles(roles: string[]) {
+    this.allowedRoles = roles;
+  }
 }

package/src/otel_resource.ts

@@ -1,16 +1,54 @@
 import {
+  type DetectorSync,
   type IResource,
+  Resource,
   detectResourcesSync,
   envDetectorSync,
   osDetectorSync,
-  processDetectorSync,
   serviceInstanceIdDetectorSync,
 } from '@opentelemetry/resources';
+import { SEMRESATTRS_SERVICE_NAME } from '@opentelemetry/semantic-conventions';
+
+import { AZTEC_NODE_ROLE, AZTEC_REGISTRY_ADDRESS, AZTEC_ROLLUP_ADDRESS, AZTEC_ROLLUP_VERSION } from './attributes.js';
 
 export function getOtelResource(): IResource {
   const resource = detectResourcesSync({
-    detectors: [osDetectorSync, envDetectorSync, processDetectorSync, serviceInstanceIdDetectorSync],
+    detectors: [
+      aztecNetworkDetectorSync,
+      osDetectorSync,
+      envDetectorSync,
+      // this detector is disabled because:
+      // 1. our software runs in a docker container, a lot of the attributes detected would be identical across different machines (e.g. all run node v22, executing the same script, running PID 1, etc)
+      // 2. it catures process.argv which could contain sensitive values in plain text (e.g. validator private keys)
+      // processDetectorSync,
+      serviceInstanceIdDetectorSync,
+    ],
   });
 
   return resource;
 }
+
+const aztecNetworkDetectorSync: DetectorSync = {
+  detect(): IResource {
+    let role: string | undefined;
+    if (process.argv.includes('--sequencer')) {
+      role = 'sequencer';
+    } else if (process.argv.includes('--prover-node')) {
+      role = 'prover-node';
+    } else if (process.argv.includes('--node')) {
+      role = 'node';
+    } else if (process.argv.includes('--p2p-bootstrap')) {
+      role = 'bootnode';
+    }
+    const aztecAttributes = {
+      // this gets overwritten by OTEL_RESOURCE_ATTRIBUTES (if set)
+      [SEMRESATTRS_SERVICE_NAME]: role ? `aztec-${role}` : undefined,
+      [AZTEC_NODE_ROLE]: role,
+      [AZTEC_ROLLUP_VERSION]: process.env.ROLLUP_VERSION ?? 'canonical',
+      [AZTEC_ROLLUP_ADDRESS]: process.env.ROLLUP_CONTRACT_ADDRESS,
+      [AZTEC_REGISTRY_ADDRESS]: process.env.REGISTRY_CONTRACT_ADDRESS,
+    };
+
+    return new Resource(aztecAttributes);
+  },
+};