@aztec/validator-ha-signer 0.0.1-commit.ee80a48 → 0.0.1-commit.f103f88

package/src/db/schema.ts

@@ -9,7 +9,7 @@
 /**
  * Current schema version
  */
-export const SCHEMA_VERSION = 1;
+export const SCHEMA_VERSION = 2;
 
 /**
  * SQL to create the validator_duties table
@@ -20,9 +20,10 @@ CREATE TABLE IF NOT EXISTS validator_duties (
   validator_address VARCHAR(42) NOT NULL,
   slot BIGINT NOT NULL,
   block_number BIGINT NOT NULL,
+  checkpoint_number BIGINT NOT NULL DEFAULT 0,
   block_index_within_checkpoint INTEGER NOT NULL DEFAULT 0,
   duty_type VARCHAR(30) NOT NULL CHECK (duty_type IN ('BLOCK_PROPOSAL', 'CHECKPOINT_PROPOSAL', 'ATTESTATION', 'ATTESTATIONS_AND_SIGNERS', 'GOVERNANCE_VOTE', 'SLASHING_VOTE')),
-  status VARCHAR(20) NOT NULL CHECK (status IN ('signing', 'signed', 'failed')),
+  status VARCHAR(20) NOT NULL CHECK (status IN ('signing', 'signed')),
   message_hash VARCHAR(66) NOT NULL,
   signature VARCHAR(132),
   node_id VARCHAR(255) NOT NULL,
@@ -106,6 +107,7 @@ WITH inserted AS (
     validator_address,
     slot,
     block_number,
+    checkpoint_number,
     block_index_within_checkpoint,
     duty_type,
     status,
@@ -113,13 +115,14 @@ WITH inserted AS (
     node_id,
     lock_token,
     started_at
-  ) VALUES ($1, $2, $3, $4, $5, $6, 'signing', $7, $8, $9, CURRENT_TIMESTAMP)
+  ) VALUES ($1, $2, $3, $4, $5, $6, $7, 'signing', $8, $9, $10, CURRENT_TIMESTAMP)
   ON CONFLICT (rollup_address, validator_address, slot, duty_type, block_index_within_checkpoint) DO NOTHING
   RETURNING
     rollup_address,
     validator_address,
     slot,
     block_number,
+    checkpoint_number,
     block_index_within_checkpoint,
     duty_type,
     status,
@@ -139,6 +142,7 @@ SELECT
   validator_address,
   slot,
   block_number,
+  checkpoint_number,
   block_index_within_checkpoint,
   duty_type,
   status,
@@ -154,8 +158,8 @@ FROM validator_duties
 WHERE rollup_address = $1
   AND validator_address = $2
   AND slot = $3
-  AND duty_type = $6
-  AND block_index_within_checkpoint = $5
+  AND duty_type = $7
+  AND block_index_within_checkpoint = $6
   AND NOT EXISTS (SELECT 1 FROM inserted);
 `;
 
@@ -203,23 +207,34 @@ WHERE status = 'signed'
 
 /**
  * Query to clean up old duties (for maintenance)
- * Removes duties older than a specified timestamp
+ * Removes SIGNED duties older than a specified age (in milliseconds)
  */
 export const CLEANUP_OLD_DUTIES = `
 DELETE FROM validator_duties
-WHERE status IN ('signing', 'signed', 'failed')
-  AND started_at < $1;
+WHERE status = 'signed'
+  AND started_at < CURRENT_TIMESTAMP - ($1 || ' milliseconds')::INTERVAL;
 `;
 
 /**
  * Query to cleanup own stuck duties
  * Removes duties in 'signing' status for a specific node that are older than maxAgeMs
+ * Uses DB's CURRENT_TIMESTAMP to avoid clock skew issues between nodes
  */
 export const CLEANUP_OWN_STUCK_DUTIES = `
 DELETE FROM validator_duties
 WHERE node_id = $1
   AND status = 'signing'
-  AND started_at < $2;
+  AND started_at < CURRENT_TIMESTAMP - ($2 || ' milliseconds')::INTERVAL;
+`;
+
+/**
+ * Query to cleanup duties with outdated rollup address
+ * Removes all duties where the rollup address doesn't match the current one
+ * Used after a rollup upgrade to clean up duties for the old rollup
+ */
+export const CLEANUP_OUTDATED_ROLLUP_DUTIES = `
+DELETE FROM validator_duties
+WHERE rollup_address != $1;
 `;
 
 /**
@@ -242,6 +257,7 @@ SELECT
   validator_address,
   slot,
   block_number,
+  checkpoint_number,
   block_index_within_checkpoint,
   duty_type,
   status,

package/src/db/types.ts

@@ -1,6 +1,7 @@
-import type { BlockNumber, CheckpointNumber, IndexWithinCheckpoint, SlotNumber } from '@aztec/foundation/branded-types';
-import type { EthAddress } from '@aztec/foundation/eth-address';
+import { BlockNumber, CheckpointNumber, type IndexWithinCheckpoint, SlotNumber } from '@aztec/foundation/branded-types';
+import { EthAddress } from '@aztec/foundation/eth-address';
 import type { Signature } from '@aztec/foundation/eth-signature';
+import { DutyType } from '@aztec/stdlib/ha-signing';
 
 /**
  * Row type from PostgreSQL query
@@ -10,6 +11,7 @@ export interface DutyRow {
   validator_address: string;
   slot: string;
   block_number: string;
+  checkpoint_number: string;
   block_index_within_checkpoint: number;
   duty_type: DutyType;
   status: DutyStatus;
@@ -23,24 +25,35 @@ export interface DutyRow {
 }
 
 /**
- * Row type from INSERT_OR_GET_DUTY query (includes is_new flag)
+ * Plain-primitive representation of a duty record suitable for serialization
+ * (e.g. msgpackr for LMDB). All domain types are stored as their string/number
+ * equivalents. Timestamps are Unix milliseconds.
  */
-export interface InsertOrGetRow extends DutyRow {
-  is_new: boolean;
+export interface StoredDutyRecord {
+  rollupAddress: string;
+  validatorAddress: string;
+  slot: string;
+  blockNumber: string;
+  checkpointNumber: string;
+  blockIndexWithinCheckpoint: number;
+  dutyType: DutyType;
+  status: DutyStatus;
+  messageHash: string;
+  signature?: string;
+  nodeId: string;
+  lockToken: string;
+  /** Unix timestamp in milliseconds when signing started */
+  startedAtMs: number;
+  /** Unix timestamp in milliseconds when signing completed */
+  completedAtMs?: number;
+  errorMessage?: string;
 }
 
 /**
- * Type of validator duty being performed
+ * Row type from INSERT_OR_GET_DUTY query (includes is_new flag)
  */
-export enum DutyType {
-  BLOCK_PROPOSAL = 'BLOCK_PROPOSAL',
-  CHECKPOINT_PROPOSAL = 'CHECKPOINT_PROPOSAL',
-  ATTESTATION = 'ATTESTATION',
-  ATTESTATIONS_AND_SIGNERS = 'ATTESTATIONS_AND_SIGNERS',
-  GOVERNANCE_VOTE = 'GOVERNANCE_VOTE',
-  SLASHING_VOTE = 'SLASHING_VOTE',
-  AUTH_REQUEST = 'AUTH_REQUEST',
-  TXS = 'TXS',
+export interface InsertOrGetRow extends DutyRow {
+  is_new: boolean;
 }
 
 /**
@@ -51,8 +64,12 @@ export enum DutyStatus {
   SIGNED = 'signed',
 }
 
+// Re-export DutyType from stdlib
+export { DutyType };
+
 /**
- * Record of a validator duty in the database
+ * Rich representation of a validator duty, with branded types and Date objects.
+ * This is the common output type returned by all SlashingProtectionDatabase implementations.
  */
 export interface ValidatorDutyRecord {
   /** Ethereum address of the rollup contract */
@@ -61,8 +78,10 @@ export interface ValidatorDutyRecord {
   validatorAddress: EthAddress;
   /** Slot number for this duty */
   slot: SlotNumber;
-  /** Block number for this duty */
+  /** Block number for this duty (0 for non-block-proposal duties) */
   blockNumber: BlockNumber;
+  /** Checkpoint number for this duty (0 for attestation and vote duties) */
+  checkpointNumber: CheckpointNumber;
   /** Block index within checkpoint (0, 1, 2... for block proposals, -1 for other duty types) */
   blockIndexWithinCheckpoint: number;
   /** Type of duty being performed */
@@ -81,10 +100,36 @@ export interface ValidatorDutyRecord {
   startedAt: Date;
   /** When the duty signing was completed (success or failure) */
   completedAt?: Date;
-  /** Error message if status is 'failed' */
+  /** Error message (currently unused) */
   errorMessage?: string;
 }
 
+/**
+ * Convert a {@link StoredDutyRecord} (plain-primitive wire format) to a
+ * {@link ValidatorDutyRecord} (rich domain type).
+ *
+ * Shared by LMDB and any future non-Postgres backend implementations.
+ */
+export function recordFromFields(stored: StoredDutyRecord): ValidatorDutyRecord {
+  return {
+    rollupAddress: EthAddress.fromString(stored.rollupAddress),
+    validatorAddress: EthAddress.fromString(stored.validatorAddress),
+    slot: SlotNumber.fromString(stored.slot),
+    blockNumber: BlockNumber.fromString(stored.blockNumber),
+    checkpointNumber: CheckpointNumber.fromString(stored.checkpointNumber),
+    blockIndexWithinCheckpoint: stored.blockIndexWithinCheckpoint,
+    dutyType: stored.dutyType,
+    status: stored.status,
+    messageHash: stored.messageHash,
+    signature: stored.signature,
+    nodeId: stored.nodeId,
+    lockToken: stored.lockToken,
+    startedAt: new Date(stored.startedAtMs),
+    completedAt: stored.completedAtMs !== undefined ? new Date(stored.completedAtMs) : undefined,
+    errorMessage: stored.errorMessage,
+  };
+}
+
 /**
  * Duty identifier for block proposals.
  * blockIndexWithinCheckpoint is REQUIRED and must be >= 0.
@@ -163,8 +208,10 @@ export function getBlockIndexFromDutyIdentifier(duty: DutyIdentifier): number {
  * Additional parameters for checking and recording a new duty
  */
 interface CheckAndRecordExtra {
-  /** Block number for this duty */
-  blockNumber: BlockNumber | CheckpointNumber;
+  /** Block number for this duty (0 for non-block-proposal duties) */
+  blockNumber: BlockNumber;
+  /** Checkpoint number for this duty (0 for attestation and vote duties) */
+  checkpointNumber: CheckpointNumber;
   /** The signing root (hash) for this duty */
   messageHash: string;
   /** Identifier for the node that acquired the lock */

package/src/factory.ts

@@ -1,11 +1,17 @@
 /**
  * Factory functions for creating validator HA signers
  */
+import { DateProvider } from '@aztec/foundation/timer';
+import { createStore } from '@aztec/kv-store/lmdb-v2';
+import type { LocalSignerConfig, ValidatorHASignerConfig } from '@aztec/stdlib/ha-signing';
+import { getTelemetryClient } from '@aztec/telemetry-client';
+
 import { Pool } from 'pg';
 
-import type { ValidatorHASignerConfig } from './config.js';
+import { LmdbSlashingProtectionDatabase } from './db/lmdb.js';
 import { PostgresSlashingProtectionDatabase } from './db/postgres.js';
-import type { CreateHASignerDeps, SlashingProtectionDatabase } from './types.js';
+import { HASignerMetrics } from './metrics.js';
+import type { CreateHASignerDeps, CreateLocalSignerWithProtectionDeps, SlashingProtectionDatabase } from './types.js';
 import { ValidatorHASigner } from './validator_ha_signer.js';
 
 /**
@@ -23,7 +29,6 @@ import { ValidatorHASigner } from './validator_ha_signer.js';
  * ```typescript
  * const { signer, db } = await createHASigner({
  *   databaseUrl: process.env.DATABASE_URL,
- *   haSigningEnabled: true,
  *   nodeId: 'validator-node-1',
  *   pollingIntervalMs: 100,
  *   signingTimeoutMs: 3000,
@@ -55,6 +60,10 @@ export async function createHASigner(
   if (!databaseUrl) {
     throw new Error('databaseUrl is required for createHASigner');
   }
+
+  const telemetryClient = deps?.telemetryClient ?? getTelemetryClient();
+  const dateProvider = deps?.dateProvider ?? new DateProvider();
+
   // Create connection pool (or use provided pool)
   let pool: Pool;
   if (!deps?.pool) {
@@ -75,8 +84,88 @@ export async function createHASigner(
   // Verify database schema is initialized and version matches
   await db.initialize();
 
+  // Create metrics
+  const metrics = new HASignerMetrics(telemetryClient, signerConfig.nodeId);
+
   // Create signer
-  const signer = new ValidatorHASigner(db, { ...signerConfig, databaseUrl });
+  const signer = new ValidatorHASigner(db, signerConfig, { metrics, dateProvider });
 
   return { signer, db };
 }
+
+/**
+ * Create a local (single-node) signing protection signer backed by LMDB.
+ *
+ * This provides double-signing protection for nodes that are NOT running in a
+ * high-availability (multi-node) setup. It prevents a proposer from sending two
+ * proposals for the same slot if the node crashes and restarts mid-proposal.
+ *
+ * When `config.dataDirectory` is set, the protection database is persisted to disk
+ * and survives crashes/restarts. When unset, an ephemeral in-memory store is
+ * used which protects within a single run but not across restarts.
+ *
+ * @param config - Local signer config
+ * @param deps - Optional dependencies (telemetry, date provider).
+ * @returns An object containing the signer and database instances.
+ */
+export async function createLocalSignerWithProtection(
+  config: LocalSignerConfig,
+  deps?: CreateLocalSignerWithProtectionDeps,
+): Promise<{
+  signer: ValidatorHASigner;
+  db: SlashingProtectionDatabase;
+}> {
+  const telemetryClient = deps?.telemetryClient ?? getTelemetryClient();
+  const dateProvider = deps?.dateProvider ?? new DateProvider();
+
+  const kvStore = await createStore('signing-protection', LmdbSlashingProtectionDatabase.SCHEMA_VERSION, {
+    dataDirectory: config.dataDirectory,
+    dataStoreMapSizeKb: config.signingProtectionMapSizeKb ?? config.dataStoreMapSizeKb,
+    l1Contracts: config.l1Contracts,
+  });
+
+  const db = new LmdbSlashingProtectionDatabase(kvStore, dateProvider);
+
+  const signerConfig = {
+    ...config,
+    nodeId: config.nodeId || 'local',
+  };
+
+  const metrics = new HASignerMetrics(telemetryClient, signerConfig.nodeId, 'LocalSigningProtectionMetrics');
+
+  const signer = new ValidatorHASigner(db, signerConfig, { metrics, dateProvider });
+
+  return { signer, db };
+}
+
+/**
+ * Create an in-memory LMDB-backed SlashingProtectionDatabase that can be shared across
+ * multiple validator nodes in the same process. Used for testing HA setups.
+ */
+export async function createSharedSlashingProtectionDb(
+  dateProvider: DateProvider = new DateProvider(),
+): Promise<SlashingProtectionDatabase> {
+  const kvStore = await createStore('shared-signing-protection', LmdbSlashingProtectionDatabase.SCHEMA_VERSION, {
+    dataStoreMapSizeKb: 1024 * 1024,
+  });
+  return new LmdbSlashingProtectionDatabase(kvStore, dateProvider);
+}
+
+/**
+ * Create a ValidatorHASigner backed by a pre-existing SlashingProtectionDatabase.
+ * Used for testing HA setups where multiple nodes share the same protection database.
+ */
+export function createSignerFromSharedDb(
+  db: SlashingProtectionDatabase,
+  config: Pick<
+    ValidatorHASignerConfig,
+    'nodeId' | 'pollingIntervalMs' | 'signingTimeoutMs' | 'maxStuckDutiesAgeMs' | 'l1Contracts'
+  >,
+  deps?: CreateLocalSignerWithProtectionDeps,
+): { signer: ValidatorHASigner; db: SlashingProtectionDatabase } {
+  const telemetryClient = deps?.telemetryClient ?? getTelemetryClient();
+  const dateProvider = deps?.dateProvider ?? new DateProvider();
+  const metrics = new HASignerMetrics(telemetryClient, config.nodeId, 'SharedSigningProtectionMetrics');
+  const signer = new ValidatorHASigner(db, config, { metrics, dateProvider });
+  return { signer, db };
+}

package/src/metrics.ts

@@ -0,0 +1,138 @@
+import {
+  Attributes,
+  type Histogram,
+  Metrics,
+  type TelemetryClient,
+  type UpDownCounter,
+  createUpDownCounterWithDefault,
+} from '@aztec/telemetry-client';
+
+export type HACleanupType = 'stuck' | 'old' | 'outdated_rollup';
+
+/**
+ * Metrics for HA signer tracking signing operations, lock acquisition, and cleanup.
+ */
+export class HASignerMetrics {
+  // Signing lifecycle metrics
+  private signingDuration: Histogram;
+  private signingSuccessCount: UpDownCounter;
+  private dutyAlreadySignedCount: UpDownCounter;
+  private slashingProtectionCount: UpDownCounter;
+  private signingErrorCount: UpDownCounter;
+
+  // Lock acquisition metrics
+  private lockAcquiredCount: UpDownCounter;
+
+  // Cleanup metrics
+  private cleanupStuckDutiesCount: UpDownCounter;
+  private cleanupOldDutiesCount: UpDownCounter;
+  private cleanupOutdatedRollupDutiesCount: UpDownCounter;
+
+  constructor(
+    client: TelemetryClient,
+    private nodeId: string,
+    name = 'HASignerMetrics',
+  ) {
+    const meter = client.getMeter(name);
+
+    // Signing lifecycle
+    this.signingDuration = meter.createHistogram(Metrics.HA_SIGNER_SIGNING_DURATION);
+    this.signingSuccessCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_SIGNING_SUCCESS_COUNT);
+    this.dutyAlreadySignedCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_DUTY_ALREADY_SIGNED_COUNT);
+    this.slashingProtectionCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_SLASHING_PROTECTION_COUNT);
+    this.signingErrorCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_SIGNING_ERROR_COUNT);
+
+    // Lock acquisition
+    this.lockAcquiredCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_LOCK_ACQUIRED_COUNT);
+
+    // Cleanup
+    this.cleanupStuckDutiesCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_CLEANUP_STUCK_DUTIES_COUNT);
+    this.cleanupOldDutiesCount = createUpDownCounterWithDefault(meter, Metrics.HA_SIGNER_CLEANUP_OLD_DUTIES_COUNT);
+    this.cleanupOutdatedRollupDutiesCount = createUpDownCounterWithDefault(
+      meter,
+      Metrics.HA_SIGNER_CLEANUP_OUTDATED_ROLLUP_DUTIES_COUNT,
+    );
+  }
+
+  /**
+   * Record a successful signing operation.
+   * @param dutyType - The type of duty signed
+   * @param durationMs - Duration from start of signWithProtection to completion
+   */
+  public recordSigningSuccess(dutyType: string, durationMs: number): void {
+    const attributes = {
+      [Attributes.HA_DUTY_TYPE]: dutyType,
+      [Attributes.HA_NODE_ID]: this.nodeId,
+    };
+    this.signingSuccessCount.add(1, attributes);
+    this.signingDuration.record(durationMs, attributes);
+  }
+
+  /**
+   * Record a DutyAlreadySignedError (expected in HA; another node signed first).
+   * @param dutyType - The type of duty
+   */
+  public recordDutyAlreadySigned(dutyType: string): void {
+    const attributes = {
+      [Attributes.HA_DUTY_TYPE]: dutyType,
+      [Attributes.HA_NODE_ID]: this.nodeId,
+    };
+    this.dutyAlreadySignedCount.add(1, attributes);
+  }
+
+  /**
+   * Record a SlashingProtectionError (attempted to sign different data for same duty).
+   * @param dutyType - The type of duty
+   */
+  public recordSlashingProtection(dutyType: string): void {
+    const attributes = {
+      [Attributes.HA_DUTY_TYPE]: dutyType,
+      [Attributes.HA_NODE_ID]: this.nodeId,
+    };
+    this.slashingProtectionCount.add(1, attributes);
+  }
+
+  /**
+   * Record a signing function failure (lock will be deleted for retry).
+   * @param dutyType - The type of duty
+   */
+  public recordSigningError(dutyType: string): void {
+    const attributes = {
+      [Attributes.HA_DUTY_TYPE]: dutyType,
+      [Attributes.HA_NODE_ID]: this.nodeId,
+    };
+    this.signingErrorCount.add(1, attributes);
+  }
+
+  /**
+   * Record lock acquisition.
+   * @param acquired - Whether a new lock was acquired (true) or existing record found (false)
+   */
+  public recordLockAcquire(acquired: boolean): void {
+    if (acquired) {
+      const attributes = {
+        [Attributes.HA_NODE_ID]: this.nodeId,
+      };
+      this.lockAcquiredCount.add(1, attributes);
+    }
+  }
+
+  /**
+   * Record cleanup metrics.
+   * @param type - Type of cleanup
+   * @param count - Number of duties cleaned up
+   */
+  public recordCleanup(type: HACleanupType, count: number): void {
+    const attributes = {
+      [Attributes.HA_NODE_ID]: this.nodeId,
+    };
+
+    if (type === 'stuck') {
+      this.cleanupStuckDutiesCount.add(count, attributes);
+    } else if (type === 'old') {
+      this.cleanupOldDutiesCount.add(count, attributes);
+    } else if (type === 'outdated_rollup') {
+      this.cleanupOutdatedRollupDutiesCount.add(count, attributes);
+    }
+  }
+}