@aztec/telemetry-client 0.0.1-commit.e6bd8901 → 0.0.1-commit.ec7ac5448

package/src/metrics.ts

@@ -167,6 +167,55 @@ export const MEMPOOL_TX_MINED_DELAY: MetricDefinition = {
   valueType: ValueType.INT,
 };
 
+export const MEMPOOL_TX_POOL_V2_EVICTED_COUNT: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.evicted_count',
+  description: 'The number of transactions evicted from the tx pool',
+  valueType: ValueType.INT,
+};
+export const MEMPOOL_TX_POOL_V2_IGNORED_COUNT: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.ignored_count',
+  description: 'The number of transactions ignored in addPendingTxs',
+  valueType: ValueType.INT,
+};
+export const MEMPOOL_TX_POOL_V2_REJECTED_COUNT: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.rejected_count',
+  description: 'The number of transactions rejected in addPendingTxs',
+  valueType: ValueType.INT,
+};
+export const MEMPOOL_TX_POOL_V2_SOFT_DELETED_HITS: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.soft_deleted_hits',
+  description: 'The number of transactions found in the soft-deleted pool',
+  valueType: ValueType.INT,
+};
+export const MEMPOOL_TX_POOL_V2_MISSING_ON_PROTECT: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.missing_on_protect',
+  description: 'The number of truly missing transactions in protectTxs',
+  valueType: ValueType.INT,
+};
+export const MEMPOOL_TX_POOL_V2_MISSING_PREVIOUSLY_EVICTED: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.missing_previously_evicted',
+  description: 'The number of truly missing transactions in protectTxs that were previously evicted',
+  valueType: ValueType.INT,
+};
+export const MEMPOOL_TX_POOL_V2_METADATA_MEMORY: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.metadata_memory',
+  description: 'Estimated total memory consumed by in-memory transaction metadata',
+  unit: 'By',
+  valueType: ValueType.INT,
+};
+
+export const MEMPOOL_TX_POOL_V2_DUPLICATE_ADD: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.duplicate_add',
+  description: 'Transactions received via addPendingTxs that were already in the pool',
+  valueType: ValueType.INT,
+};
+
+export const MEMPOOL_TX_POOL_V2_ALREADY_PROTECTED_ADD: MetricDefinition = {
+  name: 'aztec.mempool.tx_pool_v2.already_protected_add',
+  description: 'Transactions received via addPendingTxs that were already pre-protected',
+  valueType: ValueType.INT,
+};
+
 export const DB_NUM_ITEMS: MetricDefinition = {
   name: 'aztec.db.num_items',
   description: 'LMDB Num Items',
@@ -224,6 +273,11 @@ export const ARCHIVER_BLOCK_HEIGHT: MetricDefinition = {
   description: 'The height of the latest block processed by the archiver',
   valueType: ValueType.INT,
 };
+export const ARCHIVER_CHECKPOINT_HEIGHT: MetricDefinition = {
+  name: 'aztec.archiver.checkpoint_height',
+  description: 'The height of the latest checkpoint processed by the archiver',
+  valueType: ValueType.INT,
+};
 export const ARCHIVER_ROLLUP_PROOF_DELAY: MetricDefinition = {
   name: 'aztec.archiver.rollup_proof_delay',
   description: 'Time after a block is submitted until its proof is published',
@@ -254,6 +308,12 @@ export const ARCHIVER_SYNC_PER_BLOCK: MetricDefinition = {
   unit: 'ms',
   valueType: ValueType.INT,
 };
+export const ARCHIVER_SYNC_PER_CHECKPOINT: MetricDefinition = {
+  name: 'aztec.archiver.checkpoint.sync_per_item_duration',
+  description: 'Duration to sync a checkpoint',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
 export const ARCHIVER_SYNC_BLOCK_COUNT: MetricDefinition = {
   name: 'aztec.archiver.block.sync_count',
   description: 'Number of blocks synced from L1',
@@ -296,6 +356,13 @@ export const ARCHIVER_BLOCK_PROPOSAL_TX_TARGET_COUNT: MetricDefinition = {
   valueType: ValueType.INT,
 };
 
+export const ARCHIVER_CHECKPOINT_L1_INCLUSION_DELAY: MetricDefinition = {
+  name: 'aztec.archiver.checkpoint_l1_inclusion_delay',
+  description: 'Seconds into the L2 slot when the checkpoint L1 tx was included',
+  unit: 's',
+  valueType: ValueType.INT,
+};
+
 export const NODE_RECEIVE_TX_DURATION: MetricDefinition = {
   name: 'aztec.node.receive_tx.duration',
   description: 'The duration of the receiveTx method',
@@ -344,9 +411,15 @@ export const SEQUENCER_BLOCK_COUNT: MetricDefinition = {
   description: 'Number of blocks built by this sequencer',
   valueType: ValueType.INT,
 };
-export const SEQUENCER_CURRENT_BLOCK_REWARDS: MetricDefinition = {
-  name: 'aztec.sequencer.current_block_rewards',
-  description: 'The rewards earned',
+export const SEQUENCER_BLOCK_INTER_BLOCK_TIME: MetricDefinition = {
+  name: 'aztec.sequencer.block.inter_block_time',
+  description: 'Wall-clock time elapsed between consecutive blocks being built by this sequencer',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_CURRENT_SLOT_REWARDS: MetricDefinition = {
+  name: 'aztec.sequencer.current_slot_rewards',
+  description: 'The rewards earned per filled slot',
   valueType: ValueType.DOUBLE,
 };
 export const SEQUENCER_SLOT_COUNT: MetricDefinition = {
@@ -369,12 +442,12 @@ export const SEQUENCER_CHECKPOINT_ATTESTATION_DELAY: MetricDefinition = {
 
 export const SEQUENCER_COLLECTED_ATTESTATIONS_COUNT: MetricDefinition = {
   name: 'aztec.sequencer.attestations.collected_count',
-  description: 'The number of attestations collected for a block proposal',
+  description: 'The number of attestations collected for a checkpoint proposal',
   valueType: ValueType.INT,
 };
 export const SEQUENCER_REQUIRED_ATTESTATIONS_COUNT: MetricDefinition = {
   name: 'aztec.sequencer.attestations.required_count',
-  description: 'The minimum number of attestations required to publish a block',
+  description: 'The minimum number of attestations required to publish a checkpoint',
   valueType: ValueType.INT,
 };
 export const SEQUENCER_COLLECT_ATTESTATIONS_DURATION: MetricDefinition = {
@@ -395,14 +468,42 @@ export const SEQUENCER_BLOCK_PROPOSAL_FAILED_COUNT: MetricDefinition = {
   description: 'The number of times block proposal failed (including validation builds)',
   valueType: ValueType.INT,
 };
-export const SEQUENCER_BLOCK_PROPOSAL_SUCCESS_COUNT: MetricDefinition = {
-  name: 'aztec.sequencer.block.proposal_success_count',
-  description: 'The number of times block proposal succeeded (including validation builds)',
+export const SEQUENCER_CHECKPOINT_PROPOSAL_SUCCESS_COUNT: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.proposal_success_count',
+  description: 'The number of times checkpoint proposal succeeded',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_CHECKPOINT_PRECHECK_FAILED_COUNT: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.precheck_failed_count',
+  description: 'The number of times checkpoint pre-build checks failed',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_CHECKPOINT_PROPOSAL_FAILED_COUNT: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.proposal_failed_count',
+  description: 'The number of times checkpoint proposal failed',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_CHECKPOINT_BUILD_DURATION: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.build_duration',
+  description: 'Total duration to build all blocks in a checkpoint',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_CHECKPOINT_BLOCK_COUNT: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.block_count',
+  description: 'Number of blocks built in a checkpoint',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_CHECKPOINT_TX_COUNT: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.tx_count',
+  description: 'Total number of transactions across all blocks in a checkpoint',
+  unit: 'tx',
   valueType: ValueType.INT,
 };
-export const SEQUENCER_BLOCK_PROPOSAL_PRECHECK_FAILED_COUNT: MetricDefinition = {
-  name: 'aztec.sequencer.block.proposal_precheck_failed_count',
-  description: 'The number of times block proposal pre-build checks failed',
+export const SEQUENCER_CHECKPOINT_TOTAL_MANA: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.total_mana',
+  description: 'Total L2 mana used across all blocks in a checkpoint',
+  unit: 'mana',
   valueType: ValueType.INT,
 };
 export const SEQUENCER_SLASHING_ATTEMPTS_COUNT: MetricDefinition = {
@@ -415,6 +516,16 @@ export const SEQUENCER_CHECKPOINT_SUCCESS_COUNT: MetricDefinition = {
   description: 'The number of times checkpoint publishing succeeded',
   valueType: ValueType.INT,
 };
+export const SEQUENCER_PIPELINE_DEPTH: MetricDefinition = {
+  name: 'aztec.sequencer.pipeline.depth',
+  description: 'Current pipeline depth when builder pipelining is enabled',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_PIPELINE_DISCARDS_COUNT: MetricDefinition = {
+  name: 'aztec.sequencer.pipeline.discards_count',
+  description: 'The number of times a pipeline was discarded',
+  valueType: ValueType.INT,
+};
 
 // Fisherman fee analysis metrics
 export const FISHERMAN_FEE_ANALYSIS_WOULD_BE_INCLUDED: MetricDefinition = {
@@ -474,6 +585,29 @@ export const FISHERMAN_FEE_ANALYSIS_MINED_BLOB_TX_TOTAL_COST: MetricDefinition =
   unit: 'eth',
   valueType: ValueType.DOUBLE,
 };
+export const FISHERMAN_FEE_ANALYSIS_PENDING_BLOB_COUNT: MetricDefinition = {
+  name: 'aztec.fisherman.fee_analysis.pending_blob_count',
+  description: 'Total number of blobs in pending blob transactions',
+  unit: 'blobs',
+  valueType: ValueType.INT,
+};
+export const FISHERMAN_FEE_ANALYSIS_INCLUDED_BLOB_COUNT: MetricDefinition = {
+  name: 'aztec.fisherman.fee_analysis.included_blob_count',
+  description: 'Total number of blobs included in the mined block',
+  unit: 'blobs',
+  valueType: ValueType.INT,
+};
+export const FISHERMAN_FEE_ANALYSIS_BLOCK_BLOBS_FULL: MetricDefinition = {
+  name: 'aztec.fisherman.fee_analysis.block_blobs_full',
+  description: 'Whether the mined block reached 100% blob capacity',
+  valueType: ValueType.INT,
+};
+export const FISHERMAN_FEE_ANALYSIS_MAX_BLOB_CAPACITY: MetricDefinition = {
+  name: 'aztec.fisherman.fee_analysis.max_blob_capacity',
+  description: 'Maximum blob capacity for the analyzed block based on L1 upgrade schedule',
+  unit: 'blobs',
+  valueType: ValueType.INT,
+};
 
 export const VALIDATOR_INVALID_ATTESTATION_RECEIVED_COUNT: MetricDefinition = {
   name: 'aztec.validator.invalid_attestation_received_count',
@@ -652,6 +786,12 @@ export const PEER_MANAGER_PEER_COUNT: MetricDefinition = {
   unit: 'peers',
   valueType: ValueType.INT,
 };
+export const PEER_MANAGER_HEALTHY_PEER_COUNT: MetricDefinition = {
+  name: 'aztec.peer_manager.healthy_peer_count',
+  description: 'Number of healthy (non-protected, non-banned) peers',
+  unit: 'peers',
+  valueType: ValueType.INT,
+};
 export const PEER_MANAGER_LOW_SCORE_DISCONNECTS: MetricDefinition = {
   name: 'aztec.peer_manager.low_score_disconnects',
   description: 'Number of peers disconnected due to low score',
@@ -936,6 +1076,11 @@ export const PROVING_QUEUE_REJECTED_JOBS: MetricDefinition = {
   description: 'Number of rejected proving jobs',
   valueType: ValueType.INT,
 };
+export const PROVING_QUEUE_ABORTED_JOBS: MetricDefinition = {
+  name: 'aztec.proving_queue.aborted_jobs_count',
+  description: 'Number of aborted proving jobs',
+  valueType: ValueType.INT,
+};
 export const PROVING_QUEUE_RETRIED_JOBS: MetricDefinition = {
   name: 'aztec.proving_queue.retried_jobs_count',
   description: 'Number of retried proving jobs',
@@ -1010,6 +1155,36 @@ export const PROVER_NODE_JOB_TRANSACTIONS: MetricDefinition = {
   description: 'Number of transactions in a proven epoch',
   valueType: ValueType.INT,
 };
+export const PROVER_NODE_BLOB_PROCESSING_LAST_DURATION: MetricDefinition = {
+  name: 'aztec.prover_node.blob_processing.last_duration',
+  description: 'Duration of blob processing step in epoch proving job',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const PROVER_NODE_CHONK_VERIFIER_LAST_DURATION: MetricDefinition = {
+  name: 'aztec.prover_node.chonk_verifier.last_duration',
+  description: 'Duration of chonk verifier enqueuing in epoch proving job',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const PROVER_NODE_BLOCK_PROCESSING_DURATION: MetricDefinition = {
+  name: 'aztec.prover_node.block_processing.duration',
+  description: 'Duration of processing a single block in epoch proving job',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const PROVER_NODE_CHECKPOINT_PROCESSING_DURATION: MetricDefinition = {
+  name: 'aztec.prover_node.checkpoint_processing.duration',
+  description: 'Duration of processing a single checkpoint in epoch proving job',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const PROVER_NODE_ALL_CHECKPOINTS_PROCESSING_LAST_DURATION: MetricDefinition = {
+  name: 'aztec.prover_node.all_checkpoints_processing.last_duration',
+  description: 'Duration of processing all checkpoints in epoch proving job',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
 export const PROVER_NODE_REWARDS_TOTAL: MetricDefinition = {
   name: 'aztec.prover_node.rewards_total',
   description: 'The rewards earned (total)',
@@ -1144,6 +1319,16 @@ export const VALIDATOR_ATTESTATION_FAILED_NODE_ISSUE_COUNT: MetricDefinition = {
   description: 'The number of failed attestations due to node issues (timeout, missing data, etc.)',
   valueType: ValueType.INT,
 };
+export const VALIDATOR_CURRENT_EPOCH: MetricDefinition = {
+  name: 'aztec.validator.current_epoch',
+  description: 'The current epoch number, reflecting total epochs elapsed since genesis',
+  valueType: ValueType.INT,
+};
+export const VALIDATOR_ATTESTED_EPOCH_COUNT: MetricDefinition = {
+  name: 'aztec.validator.attested_epoch_count',
+  description: 'The number of epochs in which this node successfully submitted at least one attestation',
+  valueType: ValueType.INT,
+};
 
 export const NODEJS_EVENT_LOOP_DELAY_MIN: MetricDefinition = {
   name: 'nodejs.eventloop.delay.min',
@@ -1271,6 +1456,55 @@ export const TX_COLLECTOR_DURATION_PER_TX: MetricDefinition = {
   valueType: ValueType.INT,
 };
 
+export const TX_FILE_STORE_UPLOADS_SUCCESS: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.uploads_success',
+  description: 'Number of successful tx uploads to file storage',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_UPLOADS_FAILED: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.uploads_failed',
+  description: 'Number of failed tx uploads to file storage',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_UPLOADS_SKIPPED: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.uploads_skipped',
+  description: 'Number of tx uploads skipped (duplicates)',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_UPLOAD_DURATION: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.upload_duration',
+  description: 'Duration to upload a tx to file storage',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_QUEUE_SIZE: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.queue_size',
+  description: 'Number of txs pending upload',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_DOWNLOADS_SUCCESS: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.downloads_success',
+  description: 'Number of successful tx downloads from file storage',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_DOWNLOADS_FAILED: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.downloads_failed',
+  description: 'Number of failed tx downloads from file storage',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_DOWNLOAD_DURATION: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.download_duration',
+  description: 'Duration to download a tx from file storage',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const TX_FILE_STORE_DOWNLOAD_SIZE: MetricDefinition = {
+  name: 'aztec.p2p.tx_file_store.download_size',
+  description: 'Size of a downloaded tx from file storage',
+  unit: 'By',
+  valueType: ValueType.INT,
+};
+
 export const IVC_VERIFIER_TIME: MetricDefinition = {
   name: 'aztec.ivc_verifier.time',
   description: 'Duration to verify chonk proofs',
@@ -1319,3 +1553,51 @@ export const IVC_VERIFIER_AGG_DURATION_AVG: MetricDefinition = {
   unit: 'ms',
   valueType: ValueType.DOUBLE,
 };
+
+// HA Signer metrics
+export const HA_SIGNER_SIGNING_DURATION: MetricDefinition = {
+  name: 'aztec.ha_signer.signing_duration',
+  description: 'End-to-end duration for signing with HA protection (lock + sign + record)',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_SIGNING_SUCCESS_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.signing_success_count',
+  description: 'Count of successful signing operations',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_DUTY_ALREADY_SIGNED_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.duty_already_signed_count',
+  description: 'Count of DutyAlreadySignedError (expected in HA; another node signed first)',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_SLASHING_PROTECTION_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.slashing_protection_count',
+  description: 'Count of SlashingProtectionError (attempted to sign different data)',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_SIGNING_ERROR_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.signing_error_count',
+  description: 'Count of signing function failures (lock deleted for retry)',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_LOCK_ACQUIRED_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.lock_acquired_count',
+  description: 'Count of lock acquisitions (new lock acquired vs existing record found)',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_CLEANUP_STUCK_DUTIES_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.cleanup_stuck_duties_count',
+  description: 'Number of stuck duties cleaned up per cleanup run',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_CLEANUP_OLD_DUTIES_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.cleanup_old_duties_count',
+  description: 'Number of old duties cleaned up',
+  valueType: ValueType.INT,
+};
+export const HA_SIGNER_CLEANUP_OUTDATED_ROLLUP_DUTIES_COUNT: MetricDefinition = {
+  name: 'aztec.ha_signer.cleanup_outdated_rollup_duties_count',
+  description: 'Number of duties cleaned due to rollup upgrade',
+  valueType: ValueType.INT,
+};

package/src/nodejs_metrics_monitor.ts

@@ -2,6 +2,7 @@ import type { Observable } from '@opentelemetry/api';
 import { type EventLoopUtilization, type IntervalHistogram, monitorEventLoopDelay, performance } from 'node:perf_hooks';
 
 import * as Attributes from './attributes.js';
+import { createUpDownCounterWithDefault } from './metric-utils.js';
 import * as Metrics from './metrics.js';
 import type { BatchObservableResult, Meter, ObservableGauge, UpDownCounter } from './telemetry.js';
 
@@ -41,7 +42,9 @@ export class NodejsMetricsMonitor {
     };
 
     this.eventLoopUilization = meter.createObservableGauge(Metrics.NODEJS_EVENT_LOOP_UTILIZATION);
-    this.eventLoopTime = meter.createUpDownCounter(Metrics.NODEJS_EVENT_LOOP_TIME);
+    this.eventLoopTime = createUpDownCounterWithDefault(meter, Metrics.NODEJS_EVENT_LOOP_TIME, {
+      [Attributes.NODEJS_EVENT_LOOP_STATE]: ['idle', 'active'],
+    });
     this.eventLoopDelay = monitorEventLoopDelay();
 
     this.memoryGauges = {

package/src/otel_resource.ts

@@ -7,10 +7,26 @@ import {
   osDetectorSync,
   serviceInstanceIdDetectorSync,
 } from '@opentelemetry/resources';
-import { SEMRESATTRS_SERVICE_NAME } from '@opentelemetry/semantic-conventions';
+import { ATTR_SERVICE_NAME, ATTR_SERVICE_VERSION } from '@opentelemetry/semantic-conventions';
+import { readFileSync } from 'fs';
+import { dirname, resolve } from 'path';
+import { fileURLToPath } from 'url';
 
 import { AZTEC_NODE_ROLE, AZTEC_REGISTRY_ADDRESS, AZTEC_ROLLUP_ADDRESS, AZTEC_ROLLUP_VERSION } from './attributes.js';
 
+/** Reads the Aztec client version from the release manifest. */
+function getAztecVersion(): string | undefined {
+  try {
+    const releasePleasePath = resolve(
+      dirname(fileURLToPath(import.meta.url)),
+      '../../../.release-please-manifest.json',
+    );
+    return JSON.parse(readFileSync(releasePleasePath, 'utf-8'))['.'];
+  } catch {
+    return undefined;
+  }
+}
+
 export function getOtelResource(): IResource {
   const resource = detectResourcesSync({
     detectors: [
@@ -42,7 +58,8 @@ const aztecNetworkDetectorSync: DetectorSync = {
     }
     const aztecAttributes = {
       // this gets overwritten by OTEL_RESOURCE_ATTRIBUTES (if set)
-      [SEMRESATTRS_SERVICE_NAME]: role ? `aztec-${role}` : undefined,
+      [ATTR_SERVICE_NAME]: role ? `aztec-${role}` : undefined,
+      [ATTR_SERVICE_VERSION]: getAztecVersion(),
       [AZTEC_NODE_ROLE]: role,
       [AZTEC_ROLLUP_VERSION]: process.env.ROLLUP_VERSION ?? 'canonical',
       [AZTEC_ROLLUP_ADDRESS]: process.env.ROLLUP_CONTRACT_ADDRESS,

package/src/prom_otel_adapter.ts

@@ -1,4 +1,4 @@
-import { type Logger, createLogger } from '@aztec/foundation/log';
+import { type Logger, type LoggerBindings, createLogger } from '@aztec/foundation/log';
 import { Timer } from '@aztec/foundation/timer';
 
 import { Registry } from 'prom-client';
@@ -410,12 +410,11 @@ function parseLabelsSafely<Labels extends LabelsGeneric>(labelStr: string, logge
  */
 export class OtelMetricsAdapter extends Registry implements MetricsRegister {
   private readonly meter: Meter;
+  private logger: Logger;
 
-  constructor(
-    telemetryClient: TelemetryClient,
-    private logger: Logger = createLogger('telemetry:otel-metrics-adapter'),
-  ) {
+  constructor(telemetryClient: TelemetryClient, bindings?: LoggerBindings) {
     super();
+    this.logger = createLogger('telemetry:otel-metrics-adapter', bindings);
     this.meter = telemetryClient.getMeter('metrics-adapter');
   }
 

package/src/start.ts

@@ -1,4 +1,4 @@
-import { createLogger } from '@aztec/foundation/log';
+import { type LoggerBindings, createLogger } from '@aztec/foundation/log';
 
 import type { TelemetryClientConfig } from './config.js';
 import { NoopTelemetryClient } from './noop.js';
@@ -9,8 +9,11 @@ export * from './config.js';
 let initialized = false;
 let telemetry: TelemetryClient = new NoopTelemetryClient();
 
-export async function initTelemetryClient(config: TelemetryClientConfig): Promise<TelemetryClient> {
-  const log = createLogger('telemetry:client');
+export async function initTelemetryClient(
+  config: TelemetryClientConfig,
+  bindings?: LoggerBindings,
+): Promise<TelemetryClient> {
+  const log = createLogger('telemetry:client', bindings);
   if (initialized) {
     log.warn('Telemetry client has already been initialized once');
     return telemetry;

package/src/telemetry.ts

@@ -20,7 +20,7 @@ import type * as Attributes from './attributes.js';
 import type { MetricDefinition } from './metrics.js';
 import { getTelemetryClient } from './start.js';
 
-export { toMetricOptions } from './metric-utils.js';
+export { toMetricOptions, createUpDownCounterWithDefault } from './metric-utils.js';
 
 export { type Span, SpanStatusCode, ValueType, type Context } from '@opentelemetry/api';
 
@@ -56,8 +56,10 @@ type BannedMetricAttributeNames = (typeof Attributes)[
 /** Global registry of attributes */
 export type AttributesType = Partial<Record<AttributeNames, AttributeValue>>;
 
+export type AllowedAttributeNames = Exclude<AttributeNames, BannedMetricAttributeNames>;
+
 /** Subset of attributes allowed to be added to metrics */
-export type MetricAttributesType = Partial<Record<Exclude<AttributeNames, BannedMetricAttributeNames>, AttributeValue>>;
+export type MetricAttributesType = Partial<Record<AllowedAttributeNames, AttributeValue>>;
 
 /** Re-export MetricDefinition for convenience */
 export type { MetricDefinition } from './metrics.js';

package/src/wrappers/fetch.ts

@@ -9,12 +9,17 @@ import { ATTR_JSONRPC_METHOD, ATTR_JSONRPC_REQUEST_ID } from '../vendor/attribut
 
 /**
  * Makes a fetch function that retries based on the given attempts and propagates trace information.
- * @param retries - Sequence of intervals (in seconds) to retry.
+ * @param retries - Sequence of intervals (in seconds) to retry, or a factory function returning an iterator for custom/indefinite backoff.
  * @param noRetry - Whether to stop retries on server errors.
  * @param log - Optional logger for logging attempts.
  * @returns A fetch function.
  */
-export function makeTracedFetch(retries: number[], defaultNoRetry: boolean, fetch = defaultFetch, log?: Logger) {
+export function makeTracedFetch(
+  retries: number[] | (() => Generator<number>),
+  defaultNoRetry: boolean,
+  fetch = defaultFetch,
+  log?: Logger,
+) {
   return (host: string, body: unknown, extraHeaders: Record<string, string> = {}, noRetry?: boolean) => {
     const telemetry = getTelemetryClient();
     return telemetry.getTracer('fetch').startActiveSpan(`JsonRpcClient`, { kind: SpanKind.CLIENT }, async span => {
@@ -27,10 +32,11 @@ export function makeTracedFetch(retries: number[], defaultNoRetry: boolean, fetc
         }
         const headers = { ...extraHeaders };
         propagation.inject(context.active(), headers);
+        const backoff = typeof retries === 'function' ? retries() : makeBackoff(retries);
         return await retry(
           () => fetch(host, body, headers, noRetry ?? defaultNoRetry),
           `JsonRpcClient request to ${host}`,
-          makeBackoff(retries),
+          backoff,
           log,
           false,
         );