@aztec/telemetry-client 0.0.1-commit.2448fdb → 0.0.1-commit.2606882

package/dest/wrappers/l2_block_stream.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"l2_block_stream.d.ts","sourceRoot":"","sources":["../../src/wrappers/l2_block_stream.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,iCAAiC,CAAC;AAE9D,OAAO,EACL,KAAK,aAAa,EAClB,aAAa,EACb,KAAK,yBAAyB,EAC9B,KAAK,8BAA8B,EACpC,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,KAAK,SAAS,EAAE,KAAK,MAAM,EAAa,MAAM,yBAAyB,CAAC;AAEjF,kFAAkF;AAClF,qBAAa,sBAAuB,SAAQ,aAAc,YAAW,SAAS;aAQ1D,MAAM,EAAE,MAAM;IAC9B,OAAO,CAAC,QAAQ,CAAC,IAAI;IARvB,YACE,aAAa,EAAE,IAAI,CACjB,aAAa,EACb,WAAW,GAAG,gBAAgB,GAAG,WAAW,GAAG,gBAAgB,GAAG,uBAAuB,CAC1F,EACD,SAAS,EAAE,8BAA8B,EACzC,OAAO,EAAE,yBAAyB,EAClB,MAAM,EAAE,MAAM,EACb,IAAI,GAAE,MAAwB,EAC/C,GAAG,yCAAqC,EACxC,IAAI,GAAE;QACJ,MAAM,CAAC,EAAE,OAAO,CAAC;QACjB,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,aAAa,CAAC,EAAE,WAAW,CAAC;KACxB,EAGP;IAOQ,IAAI,kBAEZ;CACF"}
+{"version":3,"file":"l2_block_stream.d.ts","sourceRoot":"","sources":["../../src/wrappers/l2_block_stream.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,WAAW,EAAE,MAAM,iCAAiC,CAAC;AAE9D,OAAO,EACL,KAAK,aAAa,EAClB,aAAa,EACb,KAAK,yBAAyB,EAC9B,KAAK,8BAA8B,EACpC,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EAAE,KAAK,SAAS,EAAE,KAAK,MAAM,EAAa,MAAM,yBAAyB,CAAC;AAEjF,kFAAkF;AAClF,qBAAa,sBAAuB,SAAQ,aAAc,YAAW,SAAS;aAK1D,MAAM,EAAE,MAAM;IAC9B,OAAO,CAAC,QAAQ,CAAC,IAAI;IALvB,YACE,aAAa,EAAE,IAAI,CAAC,aAAa,EAAE,WAAW,GAAG,cAAc,GAAG,WAAW,GAAG,gBAAgB,CAAC,EACjG,SAAS,EAAE,8BAA8B,EACzC,OAAO,EAAE,yBAAyB,EAClB,MAAM,EAAE,MAAM,EACb,IAAI,GAAE,MAAwB,EAC/C,GAAG,yCAAqC,EACxC,IAAI,GAAE;QACJ,MAAM,CAAC,EAAE,OAAO,CAAC;QACjB,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,aAAa,CAAC,EAAE,WAAW,CAAC;KACxB,EAGP;IAOQ,IAAI,kBAEZ;CACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aztec/telemetry-client",
-  "version": "0.0.1-commit.2448fdb",
+  "version": "0.0.1-commit.2606882",
   "inherits": [
     "../package.common.json"
   ],
@@ -27,8 +27,8 @@
     "!*.test.*"
   ],
   "dependencies": {
-    "@aztec/foundation": "0.0.1-commit.2448fdb",
-    "@aztec/stdlib": "0.0.1-commit.2448fdb",
+    "@aztec/foundation": "0.0.1-commit.2606882",
+    "@aztec/stdlib": "0.0.1-commit.2606882",
     "@opentelemetry/api": "^1.9.0",
     "@opentelemetry/api-logs": "^0.55.0",
     "@opentelemetry/core": "^1.28.0",

package/src/attributes.ts

@@ -45,6 +45,8 @@ export const BLOCK_NUMBER = 'aztec.block.number';
 export const BLOCK_HASH = 'aztec.block.hash';
 /** The slot number */
 export const SLOT_NUMBER = 'aztec.slot.number';
+/** Whether an event happened before or after a slot boundary. */
+export const SLOT_BOUNDARY_SIDE = 'aztec.slot_boundary_side';
 /** The checkpoint number */
 export const CHECKPOINT_NUMBER = 'aztec.checkpoint.number';
 /** The parent's block number */
@@ -94,7 +96,6 @@ export const VALIDATOR_STATUS = 'aztec.validator_status';
 
 export const P2P_ID = 'aztec.p2p.id';
 export const P2P_REQ_RESP_PROTOCOL = 'aztec.p2p.req_resp.protocol';
-export const P2P_REQ_RESP_BATCH_REQUESTS_COUNT = 'aztec.p2p.req_resp.batch_requests_count';
 /** The state of a peer (Healthy, Disconnect, Banned) */
 export const P2P_PEER_SCORE_STATE = 'aztec.p2p.peer_score_state';
 export const POOL_NAME = 'aztec.pool.name';
@@ -151,5 +152,11 @@ export const L1_BLOCK_PROPOSAL_TX_TARGET = 'aztec.l1.block_proposal_tx_target';
 /** Whether tracing methods were used to extract block proposal data */
 export const L1_BLOCK_PROPOSAL_USED_TRACE = 'aztec.l1.block_proposal_used_trace';
 
+/** HA signer duty type (e.g., BLOCK_PROPOSAL, CHECKPOINT_ATTESTATION) */
+export const HA_DUTY_TYPE = 'aztec.ha_signer.duty_type';
+
+/** HA signer node identifier */
+export const HA_NODE_ID = 'aztec.ha_signer.node_id';
+
 /** The address of an attester (validator) participating in consensus */
 export const ATTESTER_ADDRESS = 'aztec.attester.address';

package/src/config.ts

@@ -1,4 +1,9 @@
-import { type ConfigMappingsType, booleanConfigHelper, getConfigFromMappings } from '@aztec/foundation/config';
+import {
+  type ConfigMappingsType,
+  booleanConfigHelper,
+  getConfigFromMappings,
+  numberConfigHelper,
+} from '@aztec/foundation/config';
 
 export interface TelemetryClientConfig {
   metricsCollectorUrl?: URL;
@@ -12,35 +17,35 @@ export interface TelemetryClientConfig {
   otelExportTimeoutMs: number;
   otelExcludeMetrics: string[];
   otelIncludeMetrics: string[];
+  otelMinTraceDurationMs: number;
+  otelBspMaxQueueSize: number;
 }
 
 export const telemetryClientConfigMappings: ConfigMappingsType<TelemetryClientConfig> = {
   metricsCollectorUrl: {
     env: 'OTEL_EXPORTER_OTLP_METRICS_ENDPOINT',
     description: 'The URL of the telemetry collector for metrics',
-    parseEnv: (val: string) => val && new URL(val),
+    parseEnv: (val: string) => new URL(val),
   },
   tracesCollectorUrl: {
     env: 'OTEL_EXPORTER_OTLP_TRACES_ENDPOINT',
     description: 'The URL of the telemetry collector for traces',
-    parseEnv: (val: string) => val && new URL(val),
+    parseEnv: (val: string) => new URL(val),
   },
   logsCollectorUrl: {
     env: 'OTEL_EXPORTER_OTLP_LOGS_ENDPOINT',
     description: 'The URL of the telemetry collector for logs',
-    parseEnv: (val: string) => val && new URL(val),
+    parseEnv: (val: string) => new URL(val),
   },
   otelCollectIntervalMs: {
     env: 'OTEL_COLLECT_INTERVAL_MS',
     description: 'The interval at which to collect metrics',
-    defaultValue: 60000, // Default extracted from otel client
-    parseEnv: (val: string) => parseInt(val),
+    ...numberConfigHelper(60000),
   },
   otelExportTimeoutMs: {
     env: 'OTEL_EXPORT_TIMEOUT_MS',
     description: 'The timeout for exporting metrics',
-    defaultValue: 30000, // Default extracted from otel client
-    parseEnv: (val: string) => parseInt(val),
+    ...numberConfigHelper(30000),
   },
   otelExcludeMetrics: {
     env: 'OTEL_EXCLUDE_METRICS',
@@ -54,6 +59,16 @@ export const telemetryClientConfigMappings: ConfigMappingsType<TelemetryClientCo
         : [],
     defaultValue: [],
   },
+  otelMinTraceDurationMs: {
+    env: 'OTEL_MIN_TRACE_DURATION_MS',
+    description: 'The minimum successful trace duration to export in milliseconds. Set to 0 to export all traces.',
+    ...numberConfigHelper(10),
+  },
+  otelBspMaxQueueSize: {
+    env: 'OTEL_BSP_MAX_QUEUE_SIZE',
+    description: 'The maximum number of completed spans to queue before export.',
+    ...numberConfigHelper(2048),
+  },
   otelIncludeMetrics: {
     env: 'OTEL_INCLUDE_METRICS',
     description: 'A list of metric prefixes to include in export (ignored if OTEL_EXCLUDE_METRICS is set)',
@@ -70,7 +85,7 @@ export const telemetryClientConfigMappings: ConfigMappingsType<TelemetryClientCo
   publicMetricsCollectorUrl: {
     env: 'PUBLIC_OTEL_EXPORTER_OTLP_METRICS_ENDPOINT',
     description: 'A URL to publish a subset of metrics for public consumption',
-    parseEnv: (val: string) => val && new URL(val),
+    parseEnv: (val: string) => new URL(val),
   },
   publicMetricsCollectFrom: {
     env: 'PUBLIC_OTEL_COLLECT_FROM',

package/src/metrics.ts

@@ -308,6 +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',
@@ -350,6 +356,19 @@ 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 ARCHIVER_CHECKPOINT_PROMOTED_COUNT: MetricDefinition = {
+  name: 'aztec.archiver.checkpoint_promoted_count',
+  description: 'Number of checkpoints promoted from proposed (blob fetch skipped)',
+  valueType: ValueType.INT,
+};
+
 export const NODE_RECEIVE_TX_DURATION: MetricDefinition = {
   name: 'aztec.node.receive_tx.duration',
   description: 'The duration of the receiveTx method',
@@ -381,6 +400,12 @@ export const SEQUENCER_STATE_TRANSITION_BUFFER_DURATION: MetricDefinition = {
   unit: 'ms',
   valueType: ValueType.INT,
 };
+export const SEQUENCER_STATE_DURATION: MetricDefinition = {
+  name: 'aztec.sequencer.state_duration',
+  description: 'Wall-clock time spent in each sequencer state, labelled by the state being left',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
 export const SEQUENCER_BLOCK_BUILD_DURATION: MetricDefinition = {
   name: 'aztec.sequencer.block.build_duration',
   description: 'Duration to build a block',
@@ -398,6 +423,12 @@ export const SEQUENCER_BLOCK_COUNT: MetricDefinition = {
   description: 'Number of blocks built by this sequencer',
   valueType: ValueType.INT,
 };
+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',
@@ -470,6 +501,25 @@ export const SEQUENCER_CHECKPOINT_BUILD_DURATION: MetricDefinition = {
   unit: 'ms',
   valueType: ValueType.INT,
 };
+export const SEQUENCER_CHECKPOINT_START_TO_FIRST_BLOCK_DURATION: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.start_to_first_block_duration',
+  description: 'Time from starting checkpoint work to the first block finishing build',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_CHECKPOINT_LAST_BLOCK_TO_BROADCAST_DURATION: MetricDefinition = {
+  name: 'aztec.sequencer.checkpoint.last_block_to_broadcast_duration',
+  description: 'Time from the final block finishing build to the checkpoint proposal being broadcast',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const SEQUENCER_PIPELINED_CHECKPOINT_BUILD_START_OFFSET_FROM_SLOT_BOUNDARY: MetricDefinition = {
+  name: 'aztec.sequencer.pipelined_checkpoint.build_start_offset_from_slot_boundary',
+  description:
+    'Absolute offset from the wall-clock slot boundary when a pipelined checkpoint build starts. Use aztec.slot_boundary_side to distinguish before vs after the boundary.',
+  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',
@@ -497,6 +547,21 @@ 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,
+};
+export const SEQUENCER_PIPELINE_PARENT_CHECKPOINT_MISMATCH_COUNT: MetricDefinition = {
+  name: 'aztec.sequencer.pipeline.parent_checkpoint_mismatch_count',
+  description: 'The number of times a pipelined checkpoint was discarded because the parent did not match expectations',
+  valueType: ValueType.INT,
+};
 
 // Fisherman fee analysis metrics
 export const FISHERMAN_FEE_ANALYSIS_WOULD_BE_INCLUDED: MetricDefinition = {
@@ -664,6 +729,24 @@ export const L1_PUBLISHER_TX_TOTAL_FEE: MetricDefinition = {
   unit: 'eth',
   valueType: ValueType.DOUBLE,
 };
+export const PROVER_NODE_ESTIMATED_SUBMISSION_GAS: MetricDefinition = {
+  name: 'aztec.prover_node.estimated_submission.gas',
+  description: 'Estimated gas for a proof submission tx (proof publishing disabled, not actually sent)',
+  unit: 'gas',
+  valueType: ValueType.INT,
+};
+export const PROVER_NODE_ESTIMATED_SUBMISSION_GAS_PRICE: MetricDefinition = {
+  name: 'aztec.prover_node.estimated_submission.gas_price',
+  description: 'Estimated effective gas price for a proof submission tx (proof publishing disabled, not actually sent)',
+  unit: 'gwei',
+  valueType: ValueType.DOUBLE,
+};
+export const PROVER_NODE_ESTIMATED_SUBMISSION_TOTAL_FEE: MetricDefinition = {
+  name: 'aztec.prover_node.estimated_submission.total_fee',
+  description: 'Estimated total L1 fee for a proof submission tx (proof publishing disabled, not actually sent)',
+  unit: 'eth',
+  valueType: ValueType.DOUBLE,
+};
 
 export const L1_BLOCK_HEIGHT: MetricDefinition = {
   name: 'aztec.l1.block_height',
@@ -757,6 +840,12 @@ 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',
@@ -886,6 +975,12 @@ export const P2P_GOSSIP_AGG_MESSAGE_VALIDATION_DURATION_AVG: MetricDefinition =
   valueType: ValueType.INT,
 };
 
+export const P2P_GOSSIP_SLOW_VALIDATION_COUNT: MetricDefinition = {
+  name: 'aztec.p2p.gossip.slow_validation_count',
+  description: 'Number of gossip validations that exceeded 75% of the mcache eviction window',
+  valueType: ValueType.INT,
+};
+
 export const PUBLIC_PROCESSOR_TX_DURATION: MetricDefinition = {
   name: 'aztec.public_processor.tx_duration',
   description: 'Duration to process a public transaction',
@@ -949,6 +1044,17 @@ export const PUBLIC_PROCESSOR_TREE_INSERTION: MetricDefinition = {
   unit: 'ms',
   valueType: ValueType.INT,
 };
+export const PUBLIC_PROCESSOR_SILENTLY_SKIPPED_COUNT: MetricDefinition = {
+  name: 'aztec.public_processor.silently_skipped_count',
+  description: 'Public txs fully processed then skipped (e.g. blob-field limit); not counted as processed or failed',
+  valueType: ValueType.INT,
+};
+export const PUBLIC_PROCESSOR_SILENTLY_SKIPPED_DURATION: MetricDefinition = {
+  name: 'aztec.public_processor.silently_skipped_duration',
+  description: 'Wall-clock time spent processing txs that were then silently skipped',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
 
 export const PUBLIC_EXECUTOR_PREFIX = 'aztec.public_executor.';
 export const PUBLIC_EXECUTOR_SIMULATION_COUNT: MetricDefinition = {
@@ -1263,6 +1369,19 @@ export const VALIDATOR_RE_EXECUTION_TX_COUNT: MetricDefinition = {
   unit: 'tx',
   valueType: ValueType.INT,
 };
+export const VALIDATOR_CHECKPOINT_PROPOSAL_TO_PIPELINED_STATE_DURATION: MetricDefinition = {
+  name: 'aztec.validator.checkpoint_proposal_to_pipelined_state_duration',
+  description: 'Time from receiving a checkpoint proposal to setting proposed checkpoint state for pipelining',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
+export const VALIDATOR_CHECKPOINT_PROPOSAL_RECEIVE_OFFSET_FROM_NEXT_SLOT_BOUNDARY: MetricDefinition = {
+  name: 'aztec.validator.checkpoint.proposal_receive_offset_from_next_slot_boundary',
+  description:
+    'Absolute offset from the next slot boundary when a foreign checkpoint proposal is received. Use aztec.slot_boundary_side to distinguish before vs after the boundary.',
+  unit: 'ms',
+  valueType: ValueType.INT,
+};
 
 export const VALIDATOR_FAILED_REEXECUTION_COUNT: MetricDefinition = {
   name: 'aztec.validator.failed_reexecution_count',
@@ -1518,3 +1637,51 @@ 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/monitored_batch_span_processor.ts

@@ -0,0 +1,93 @@
+import type { Logger } from '@aztec/foundation/log';
+
+import { type Context, SpanStatusCode } from '@opentelemetry/api';
+import { hrTimeToMilliseconds } from '@opentelemetry/core';
+import type { SpanExporter } from '@opentelemetry/sdk-trace-base';
+import { BatchSpanProcessor, type BufferConfig, type ReadableSpan, type Span } from '@opentelemetry/sdk-trace-node';
+
+/** Minimum interval between drop warnings to avoid log spam. */
+const DROP_WARNING_INTERVAL_MS = 30_000;
+
+const DEFAULT_MIN_TRACE_DURATION_MS = 10;
+
+export type MonitoredBatchSpanProcessorConfig = BufferConfig & {
+  minTraceDurationMs?: number;
+};
+
+/**
+ * Wraps BatchSpanProcessor to emit warnings when spans are dropped due to a full queue.
+ * The standard BatchSpanProcessor silently discards spans when its internal queue reaches
+ * maxQueueSize, making telemetry data loss invisible to operators.
+ */
+export class MonitoredBatchSpanProcessor extends BatchSpanProcessor {
+  private readonly maxQueueSize: number;
+  private readonly minTraceDurationMs: number;
+  private readonly log: Logger;
+
+  private approxQueueSize = 0;
+  private droppedSinceLastWarning = 0;
+  private totalDropped = 0;
+  private lastWarningTime = 0;
+
+  constructor(exporter: SpanExporter, log: Logger, config?: MonitoredBatchSpanProcessorConfig) {
+    const maxQueueSize = config?.maxQueueSize ?? 2048;
+    super(exporter, { ...config, maxQueueSize });
+    this.maxQueueSize = maxQueueSize;
+    this.minTraceDurationMs = Math.max(0, config?.minTraceDurationMs ?? DEFAULT_MIN_TRACE_DURATION_MS);
+    this.log = log;
+  }
+
+  override onStart(span: Span, parentContext: Context): void {
+    super.onStart(span, parentContext);
+  }
+
+  override onEnd(span: ReadableSpan): void {
+    if (this.shouldDropShortSpan(span)) {
+      return;
+    }
+
+    if (this.approxQueueSize >= this.maxQueueSize) {
+      this.droppedSinceLastWarning++;
+      this.totalDropped++;
+      this.maybeLogDropWarning();
+    } else {
+      this.approxQueueSize++;
+    }
+    super.onEnd(span);
+  }
+
+  override async forceFlush(): Promise<void> {
+    await super.forceFlush();
+    this.approxQueueSize = 0;
+  }
+
+  override async shutdown(): Promise<void> {
+    if (this.totalDropped > 0) {
+      this.log.warn(`BatchSpanProcessor shutting down with ${this.totalDropped} total spans dropped`, {
+        totalDropped: this.totalDropped,
+      });
+    }
+    await super.shutdown();
+  }
+
+  private shouldDropShortSpan(span: ReadableSpan): boolean {
+    return (
+      this.minTraceDurationMs > 0 &&
+      span.status.code !== SpanStatusCode.ERROR &&
+      hrTimeToMilliseconds(span.duration) < this.minTraceDurationMs
+    );
+  }
+
+  private maybeLogDropWarning(): void {
+    const now = Date.now();
+    if (now - this.lastWarningTime >= DROP_WARNING_INTERVAL_MS) {
+      this.log.warn(
+        `BatchSpanProcessor dropping spans: queue full (maxQueueSize=${this.maxQueueSize}). ` +
+          `${this.droppedSinceLastWarning} dropped since last warning, ${this.totalDropped} total.`,
+        { droppedSinceLastWarning: this.droppedSinceLastWarning, totalDropped: this.totalDropped },
+      );
+      this.droppedSinceLastWarning = 0;
+      this.lastWarningTime = now;
+    }
+  }
+}

package/src/otel.ts

@@ -28,12 +28,13 @@ import {
   type PeriodicExportingMetricReaderOptions,
   View,
 } from '@opentelemetry/sdk-metrics';
-import { BatchSpanProcessor, NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
+import { NodeTracerProvider } from '@opentelemetry/sdk-trace-node';
 import { ATTR_SERVICE_NAME, ATTR_SERVICE_VERSION } from '@opentelemetry/semantic-conventions';
 
 import type { TelemetryClientConfig } from './config.js';
 import { toMetricOptions } from './metric-utils.js';
 import type { MetricDefinition } from './metrics.js';
+import { MonitoredBatchSpanProcessor } from './monitored_batch_span_processor.js';
 import { NodejsMetricsMonitor } from './nodejs_metrics_monitor.js';
 import { OtelFilterMetricExporter, PublicOtelFilterMetricExporter } from './otel_filter_metric_exporter.js';
 import { registerOtelLoggerProvider } from './otel_logger_provider.js';
@@ -334,6 +335,36 @@ export class OpenTelemetryClient implements TelemetryClient {
             true,
           ),
         }),
+        // L1 gas prices in gwei: priority fees ~0.01-10, base fees ~1-500, spikes to 1000+
+        new View({
+          instrumentType: InstrumentType.HISTOGRAM,
+          instrumentUnit: 'gwei',
+          aggregation: new ExplicitBucketHistogramAggregation(
+            [0.1, 0.5, 1, 2, 5, 10, 20, 50, 100, 200, 500, 1_000],
+            true,
+          ),
+        }),
+        // L1 gas consumption: tx gas 100k-30M, calldata/blob gas varies
+        new View({
+          instrumentType: InstrumentType.HISTOGRAM,
+          instrumentUnit: 'gas',
+          aggregation: new ExplicitBucketHistogramAggregation(
+            [
+              10_000, 50_000, 100_000, 250_000, 500_000, 1_000_000, 2_000_000, 5_000_000, 10_000_000, 15_000_000,
+              30_000_000,
+            ],
+            true,
+          ),
+        }),
+        // L1 tx total fee in ETH: typically 0.001 - 1 ETH
+        new View({
+          instrumentType: InstrumentType.HISTOGRAM,
+          instrumentUnit: 'eth',
+          aggregation: new ExplicitBucketHistogramAggregation(
+            [0.0001, 0.0005, 0.001, 0.005, 0.01, 0.05, 0.1, 0.5, 1, 5, 10],
+            true,
+          ),
+        }),
       ],
     });
   }
@@ -343,7 +374,12 @@ export class OpenTelemetryClient implements TelemetryClient {
       const tracerProvider = new NodeTracerProvider({
         resource,
         spanProcessors: config.tracesCollectorUrl
-          ? [new BatchSpanProcessor(new OTLPTraceExporter({ url: config.tracesCollectorUrl.href }))]
+          ? [
+              new MonitoredBatchSpanProcessor(new OTLPTraceExporter({ url: config.tracesCollectorUrl.href }), log, {
+                maxQueueSize: config.otelBspMaxQueueSize,
+                minTraceDurationMs: config.otelMinTraceDurationMs,
+              }),
+            ]
           : [],
       });
 

package/src/otel_propagation.ts

@@ -1,3 +1,5 @@
+import type { DiagnosticsMiddleware } from '@aztec/foundation/json-rpc/server';
+
 import { ROOT_CONTEXT, type Span, SpanKind, SpanStatusCode, propagation } from '@opentelemetry/api';
 import type Koa from 'koa';
 
@@ -17,7 +19,7 @@ export function getOtelJsonRpcPropagationMiddleware(
     const context = propagation.extract(ROOT_CONTEXT, ctx.request.headers);
     const method = (ctx.request.body as any)?.method;
     return tracer.startActiveSpan(
-      `JsonRpcServer.${method ?? 'unknown'}`,
+      `JsonRpcServer.${method ?? 'batch'}`,
       { kind: SpanKind.SERVER },
       context,
       async (span: Span): Promise<void> => {
@@ -48,3 +50,42 @@ export function getOtelJsonRpcPropagationMiddleware(
     );
   };
 }
+
+export function getOtelJsonRpcDiagnosticsMiddleware(): DiagnosticsMiddleware {
+  return function otelJsonRpcDiagnostics(ctx, next) {
+    const [namespace, method] = splitNamespace(ctx.method);
+    const scope = namespace ?? 'UnknownHandler';
+    const tracer = getTelemetryClient().getTracer(scope);
+    return tracer.startActiveSpan(
+      `${scope}.${method}`,
+      { kind: SpanKind.INTERNAL, attributes: { [ATTR_JSONRPC_METHOD]: ctx.method } },
+      async span => {
+        if (ctx.id !== null) {
+          span.setAttribute(ATTR_JSONRPC_REQUEST_ID, ctx.id);
+        }
+
+        try {
+          await next();
+          span.setStatus({ code: SpanStatusCode.OK });
+        } catch (err) {
+          span.setStatus({ code: SpanStatusCode.ERROR, message: err instanceof Error ? err.message : String(err) });
+          if (typeof err === 'string' || err instanceof Error) {
+            span.recordException(err);
+          }
+          throw err;
+        } finally {
+          span.end();
+        }
+      },
+    );
+  };
+}
+
+function splitNamespace(fullMethod: string): [namespace: string | undefined, method: string] {
+  const idx = fullMethod.indexOf('_');
+  if (idx > -1) {
+    return [fullMethod.slice(0, idx), fullMethod.slice(idx + 1)];
+  } else {
+    return [undefined, fullMethod];
+  }
+}

package/src/start.ts

@@ -19,7 +19,12 @@ export async function initTelemetryClient(
     return telemetry;
   }
 
-  if (config.metricsCollectorUrl || config.publicMetricsCollectorUrl) {
+  if (
+    config.metricsCollectorUrl ||
+    config.publicMetricsCollectorUrl ||
+    config.tracesCollectorUrl ||
+    config.logsCollectorUrl
+  ) {
     log.info(`Using OpenTelemetry client with custom collector`);
     // Lazy load OpenTelemetry to avoid loading heavy deps at startup
     const { OpenTelemetryClient } = await import('./otel.js');

package/src/telemetry.ts

@@ -48,7 +48,6 @@ type BannedMetricAttributeNames = (typeof Attributes)[
   | 'TX_HASH'
   | 'PROVING_JOB_ID'
   | 'P2P_ID'
-  | 'P2P_REQ_RESP_BATCH_REQUESTS_COUNT'
   | 'TARGET_ADDRESS'
   | 'MANA_USED'
   | 'TOTAL_INSTRUCTIONS'];

package/src/wrappers/l2_block_stream.ts

@@ -11,10 +11,7 @@ import { type Traceable, type Tracer, trackSpan } from '@aztec/telemetry-client'
 /** Extends an L2BlockStream with a tracer to create a new trace per iteration. */
 export class TraceableL2BlockStream extends L2BlockStream implements Traceable {
   constructor(
-    l2BlockSource: Pick<
-      L2BlockSource,
-      'getBlocks' | 'getBlockHeader' | 'getL2Tips' | 'getCheckpoints' | 'getCheckpointedBlocks'
-    >,
+    l2BlockSource: Pick<L2BlockSource, 'getBlocks' | 'getBlockData' | 'getL2Tips' | 'getCheckpoints'>,
     localData: L2BlockStreamLocalDataProvider,
     handler: L2BlockStreamEventHandler,
     public readonly tracer: Tracer,