@aztec/telemetry-client 0.0.1-commit.e558bd1c → 0.0.1-commit.e57c76e

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 +356,19 @@ 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',
@@ -327,6 +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',
@@ -344,9 +423,15 @@ 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 +454,12 @@ 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 +480,61 @@ 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_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',
   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_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_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 +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 = {
@@ -474,6 +621,29 @@ 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',
@@ -559,6 +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',
@@ -652,6 +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',
@@ -781,6 +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',
@@ -844,6 +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 = {
@@ -936,6 +1147,11 @@ 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 +1226,36 @@ 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)',
@@ -1123,6 +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',
@@ -1144,6 +1403,16 @@ 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',
@@ -1297,6 +1566,28 @@ export const TX_FILE_STORE_QUEUE_SIZE: MetricDefinition = {
   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',
@@ -1346,3 +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,
+              }),
+            ]
           : [],
       });