@aztec/validator-ha-signer 0.0.1-commit.fffb133c → 0.0.1-private.5d121bfd

package/src/db/lmdb.ts

@@ -0,0 +1,308 @@
+/**
+ * LMDB implementation of SlashingProtectionDatabase
+ *
+ * Provides local (single-node) double-signing protection using LMDB as the backend.
+ * Suitable for nodes that do NOT run in a high-availability multi-node setup.
+ *
+ * The LMDB store is single-writer, making setIfNotExists inherently atomic.
+ * This means we get crash-restart protection without needing an external database.
+ */
+import { SlotNumber } from '@aztec/foundation/branded-types';
+import { randomBytes } from '@aztec/foundation/crypto/random';
+import { EthAddress } from '@aztec/foundation/eth-address';
+import { type Logger, createLogger } from '@aztec/foundation/log';
+import type { DateProvider } from '@aztec/foundation/timer';
+import type { AztecAsyncKVStore, AztecAsyncMap } from '@aztec/kv-store';
+import { openStoreAt } from '@aztec/kv-store/lmdb-v2';
+
+import type { SlashingProtectionDatabase, TryInsertOrGetResult } from '../types.js';
+import {
+  type CheckAndRecordParams,
+  DutyStatus,
+  DutyType,
+  type StoredDutyRecord,
+  getBlockIndexFromDutyIdentifier,
+  recordFromFields,
+} from './types.js';
+
+const DUTIES_MAP_NAME = 'signing-protection-duties';
+const LEGACY_CHECKPOINT_NUMBER = '0';
+
+type StoredDutyRecordV1 = Omit<StoredDutyRecord, 'checkpointNumber'> & { checkpointNumber?: undefined };
+type MigratableStoredDutyRecord = StoredDutyRecord | StoredDutyRecordV1;
+
+function needsCheckpointNumberMigration(record: MigratableStoredDutyRecord): record is StoredDutyRecordV1 {
+  return record.checkpointNumber === undefined;
+}
+
+/**
+ * Migrates local slashing-protection duties from schema 1 to schema 2.
+ */
+export async function migrateLmdbSlashingProtectionDatabase(
+  dataDirectory: string,
+  currentVersion: number,
+  latestVersion: number,
+  dbMapSizeKb?: number,
+): Promise<void> {
+  if (currentVersion !== 1 || latestVersion !== LmdbSlashingProtectionDatabase.SCHEMA_VERSION) {
+    throw new Error(`Unsupported LMDB slashing-protection migration ${currentVersion} -> ${latestVersion}`);
+  }
+
+  const store = await openStoreAt(dataDirectory, dbMapSizeKb);
+  try {
+    const duties = store.openMap<string, MigratableStoredDutyRecord>(DUTIES_MAP_NAME);
+    const migratedRecords: { key: string; value: StoredDutyRecord }[] = [];
+
+    for await (const [key, record] of duties.entriesAsync()) {
+      if (needsCheckpointNumberMigration(record)) {
+        migratedRecords.push({ key, value: { ...record, checkpointNumber: LEGACY_CHECKPOINT_NUMBER } });
+      }
+    }
+
+    if (migratedRecords.length > 0) {
+      await duties.setMany(migratedRecords);
+    }
+  } finally {
+    await store.close();
+  }
+}
+
+function dutyKey(
+  rollupAddress: string,
+  validatorAddress: string,
+  slot: string,
+  dutyType: string,
+  blockIndexWithinCheckpoint: number,
+): string {
+  return `${rollupAddress}:${validatorAddress}:${slot}:${dutyType}:${blockIndexWithinCheckpoint}`;
+}
+
+/**
+ * LMDB-backed implementation of SlashingProtectionDatabase.
+ *
+ * Provides single-node double-signing protection that survives crashes and restarts.
+ * Does not provide cross-node coordination (that requires the PostgreSQL implementation).
+ */
+export class LmdbSlashingProtectionDatabase implements SlashingProtectionDatabase {
+  public static readonly SCHEMA_VERSION = 2;
+
+  private readonly duties: AztecAsyncMap<string, StoredDutyRecord>;
+  private readonly log: Logger;
+
+  constructor(
+    private readonly store: AztecAsyncKVStore,
+    private readonly dateProvider: DateProvider,
+  ) {
+    this.log = createLogger('slashing-protection:lmdb');
+    this.duties = store.openMap<string, StoredDutyRecord>(DUTIES_MAP_NAME);
+  }
+
+  /**
+   * Atomically try to insert a new duty record, or get the existing one if present.
+   *
+   * LMDB is single-writer so the read-then-write inside transactionAsync is naturally atomic.
+   */
+  public async tryInsertOrGetExisting(params: CheckAndRecordParams): Promise<TryInsertOrGetResult> {
+    const blockIndexWithinCheckpoint = getBlockIndexFromDutyIdentifier(params);
+    const key = dutyKey(
+      params.rollupAddress.toString(),
+      params.validatorAddress.toString(),
+      params.slot.toString(),
+      params.dutyType,
+      blockIndexWithinCheckpoint,
+    );
+
+    const lockToken = randomBytes(16).toString('hex');
+    const now = this.dateProvider.now();
+
+    const result = await this.store.transactionAsync(async () => {
+      const existing = await this.duties.getAsync(key);
+      if (existing) {
+        return { isNew: false as const, record: { ...existing, lockToken: '' } };
+      }
+
+      const newRecord: StoredDutyRecord = {
+        rollupAddress: params.rollupAddress.toString(),
+        validatorAddress: params.validatorAddress.toString(),
+        slot: params.slot.toString(),
+        blockNumber: params.blockNumber.toString(),
+        checkpointNumber: params.checkpointNumber.toString(),
+        blockIndexWithinCheckpoint,
+        dutyType: params.dutyType,
+        status: DutyStatus.SIGNING,
+        messageHash: params.messageHash,
+        nodeId: params.nodeId,
+        lockToken,
+        startedAtMs: now,
+      };
+      await this.duties.set(key, newRecord);
+      return { isNew: true as const, record: newRecord };
+    });
+
+    if (result.isNew) {
+      this.log.debug(`Acquired lock for duty ${params.dutyType} at slot ${params.slot}`, {
+        validatorAddress: params.validatorAddress.toString(),
+        nodeId: params.nodeId,
+      });
+    }
+
+    return { isNew: result.isNew, record: recordFromFields(result.record) };
+  }
+
+  /**
+   * Update a duty to 'signed' status with the signature.
+   * Only succeeds if the lockToken matches.
+   */
+  public updateDutySigned(
+    rollupAddress: EthAddress,
+    validatorAddress: EthAddress,
+    slot: SlotNumber,
+    dutyType: DutyType,
+    signature: string,
+    lockToken: string,
+    blockIndexWithinCheckpoint: number,
+  ): Promise<boolean> {
+    const key = dutyKey(
+      rollupAddress.toString(),
+      validatorAddress.toString(),
+      slot.toString(),
+      dutyType,
+      blockIndexWithinCheckpoint,
+    );
+
+    return this.store.transactionAsync(async () => {
+      const existing = await this.duties.getAsync(key);
+      if (!existing) {
+        this.log.warn('Failed to update duty to signed: duty not found', {
+          rollupAddress: rollupAddress.toString(),
+          validatorAddress: validatorAddress.toString(),
+          slot: slot.toString(),
+          dutyType,
+          blockIndexWithinCheckpoint,
+        });
+        return false;
+      }
+
+      if (existing.lockToken !== lockToken) {
+        this.log.warn('Failed to update duty to signed: invalid token', {
+          rollupAddress: rollupAddress.toString(),
+          validatorAddress: validatorAddress.toString(),
+          slot: slot.toString(),
+          dutyType,
+          blockIndexWithinCheckpoint,
+        });
+        return false;
+      }
+
+      await this.duties.set(key, {
+        ...existing,
+        status: DutyStatus.SIGNED,
+        signature,
+        completedAtMs: this.dateProvider.now(),
+      });
+
+      return true;
+    });
+  }
+
+  /**
+   * Delete a duty record.
+   * Only succeeds if the lockToken matches.
+   */
+  public deleteDuty(
+    rollupAddress: EthAddress,
+    validatorAddress: EthAddress,
+    slot: SlotNumber,
+    dutyType: DutyType,
+    lockToken: string,
+    blockIndexWithinCheckpoint: number,
+  ): Promise<boolean> {
+    const key = dutyKey(
+      rollupAddress.toString(),
+      validatorAddress.toString(),
+      slot.toString(),
+      dutyType,
+      blockIndexWithinCheckpoint,
+    );
+
+    return this.store.transactionAsync(async () => {
+      const existing = await this.duties.getAsync(key);
+      if (!existing || existing.lockToken !== lockToken) {
+        this.log.warn('Failed to delete duty: invalid token or duty not found', {
+          rollupAddress: rollupAddress.toString(),
+          validatorAddress: validatorAddress.toString(),
+          slot: slot.toString(),
+          dutyType,
+          blockIndexWithinCheckpoint,
+        });
+        return false;
+      }
+
+      await this.duties.delete(key);
+      return true;
+    });
+  }
+
+  /**
+   * Cleanup own stuck duties (SIGNING status older than maxAgeMs).
+   */
+  public cleanupOwnStuckDuties(nodeId: string, maxAgeMs: number): Promise<number> {
+    const cutoffMs = this.dateProvider.now() - maxAgeMs;
+
+    return this.store.transactionAsync(async () => {
+      const keysToDelete: string[] = [];
+      for await (const [key, record] of this.duties.entriesAsync()) {
+        if (record.nodeId === nodeId && record.status === DutyStatus.SIGNING && record.startedAtMs < cutoffMs) {
+          keysToDelete.push(key);
+        }
+      }
+      for (const key of keysToDelete) {
+        await this.duties.delete(key);
+      }
+      return keysToDelete.length;
+    });
+  }
+
+  /**
+   * Cleanup duties with outdated rollup address.
+   *
+   * This is always a no-op for the LMDB implementation: the underlying store is created via
+   * DatabaseVersionManager (in factory.ts), which already resets the entire data directory at
+   * startup whenever the rollup address changes.
+   */
+  public cleanupOutdatedRollupDuties(_currentRollupAddress: EthAddress): Promise<number> {
+    return Promise.resolve(0);
+  }
+
+  /**
+   * Cleanup old signed duties older than maxAgeMs.
+   */
+  public cleanupOldDuties(maxAgeMs: number): Promise<number> {
+    const cutoffMs = this.dateProvider.now() - maxAgeMs;
+
+    return this.store.transactionAsync(async () => {
+      const keysToDelete: string[] = [];
+      for await (const [key, record] of this.duties.entriesAsync()) {
+        if (
+          record.status === DutyStatus.SIGNED &&
+          record.completedAtMs !== undefined &&
+          record.completedAtMs < cutoffMs
+        ) {
+          keysToDelete.push(key);
+        }
+      }
+      for (const key of keysToDelete) {
+        await this.duties.delete(key);
+      }
+      return keysToDelete.length;
+    });
+  }
+
+  /**
+   * Close the underlying LMDB store.
+   */
+  public async close(): Promise<void> {
+    await this.store.close();
+    this.log.debug('LMDB slashing protection database closed');
+  }
+}

package/src/db/migrations/1_initial-schema.ts

@@ -1,21 +1,52 @@
 /**
  * Initial schema for validator HA slashing protection
  *
- * This migration imports SQL from the schema.ts file to ensure a single source of truth.
+ * Note: this migration contains a fixed snapshot of the schema at the time it was created.
+ * It must NOT import from schema.ts, which evolves over time and would cause this migration
+ * to produce different results on fresh runs vs. re-runs.
  */
 import type { MigrationBuilder } from 'node-pg-migrate';
 
-import { DROP_SCHEMA_VERSION_TABLE, DROP_VALIDATOR_DUTIES_TABLE, SCHEMA_SETUP, SCHEMA_VERSION } from '../schema.js';
+import { DROP_SCHEMA_VERSION_TABLE, DROP_VALIDATOR_DUTIES_TABLE } from '../schema.js';
+
+// Snapshot of the initial schema — does NOT include checkpoint_number (added in migration 2).
+const INITIAL_SCHEMA_SETUP = [
+  `CREATE TABLE IF NOT EXISTS schema_version (
+    version INTEGER PRIMARY KEY,
+    applied_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP
+  );`,
+  `CREATE TABLE IF NOT EXISTS validator_duties (
+    rollup_address VARCHAR(42) NOT NULL,
+    validator_address VARCHAR(42) NOT NULL,
+    slot BIGINT NOT NULL,
+    block_number BIGINT NOT NULL,
+    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')),
+    message_hash VARCHAR(66) NOT NULL,
+    signature VARCHAR(132),
+    node_id VARCHAR(255) NOT NULL,
+    lock_token VARCHAR(64) NOT NULL,
+    started_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+    completed_at TIMESTAMP,
+    error_message TEXT,
+
+    PRIMARY KEY (rollup_address, validator_address, slot, duty_type, block_index_within_checkpoint),
+    CHECK (completed_at IS NULL OR completed_at >= started_at)
+  );`,
+  `CREATE INDEX IF NOT EXISTS idx_validator_duties_status ON validator_duties(status, started_at);`,
+  `CREATE INDEX IF NOT EXISTS idx_validator_duties_node ON validator_duties(node_id, started_at);`,
+] as const;
 
 export function up(pgm: MigrationBuilder): void {
-  for (const statement of SCHEMA_SETUP) {
+  for (const statement of INITIAL_SCHEMA_SETUP) {
     pgm.sql(statement);
   }
 
   // Insert initial schema version
   pgm.sql(`
     INSERT INTO schema_version (version)
-    VALUES (${SCHEMA_VERSION})
+    VALUES (1)
     ON CONFLICT (version) DO NOTHING;
   `);
 }

package/src/db/migrations/2_add-checkpoint-number.ts

@@ -0,0 +1,19 @@
+/**
+ * Add checkpoint_number column to validator_duties table
+ */
+import type { MigrationBuilder } from 'node-pg-migrate';
+
+export function up(pgm: MigrationBuilder): void {
+  pgm.addColumn('validator_duties', {
+    // eslint-disable-next-line camelcase
+    checkpoint_number: { type: 'bigint', notNull: true, default: 0 },
+  });
+
+  pgm.sql(`UPDATE schema_version SET version = 2 WHERE version = 1`);
+}
+
+export function down(pgm: MigrationBuilder): void {
+  pgm.dropColumn('validator_duties', 'checkpoint_number');
+
+  pgm.sql(`UPDATE schema_version SET version = 1 WHERE version = 2`);
+}

package/src/db/postgres.ts

@@ -1,7 +1,7 @@
 /**
  * PostgreSQL implementation of SlashingProtectionDatabase
  */
-import { BlockNumber, SlotNumber } from '@aztec/foundation/branded-types';
+import { SlotNumber } from '@aztec/foundation/branded-types';
 import { randomBytes } from '@aztec/foundation/crypto/random';
 import { EthAddress } from '@aztec/foundation/eth-address';
 import { type Logger, createLogger } from '@aztec/foundation/log';
@@ -11,6 +11,8 @@ import type { QueryResult, QueryResultRow } from 'pg';
 
 import type { SlashingProtectionDatabase, TryInsertOrGetResult } from '../types.js';
 import {
+  CLEANUP_OLD_DUTIES,
+  CLEANUP_OUTDATED_ROLLUP_DUTIES,
   CLEANUP_OWN_STUCK_DUTIES,
   DELETE_DUTY,
   INSERT_OR_GET_DUTY,
@@ -18,7 +20,7 @@ import {
   UPDATE_DUTY_SIGNED,
 } from './schema.js';
 import type { CheckAndRecordParams, DutyRow, DutyType, InsertOrGetRow, ValidatorDutyRecord } from './types.js';
-import { getBlockIndexFromDutyIdentifier } from './types.js';
+import { getBlockIndexFromDutyIdentifier, recordFromFields } from './types.js';
 
 /**
  * Minimal pool interface for database operations.
@@ -101,9 +103,11 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
     const result = await retry<QueryResult<InsertOrGetRow>>(
       async () => {
         const queryResult: QueryResult<InsertOrGetRow> = await this.pool.query(INSERT_OR_GET_DUTY, [
+          params.rollupAddress.toString(),
           params.validatorAddress.toString(),
           params.slot.toString(),
           params.blockNumber.toString(),
+          params.checkpointNumber.toString(),
           blockIndexWithinCheckpoint,
           params.dutyType,
           params.messageHash,
@@ -148,6 +152,7 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
    * @returns true if the update succeeded, false if token didn't match or duty not found
    */
   async updateDutySigned(
+    rollupAddress: EthAddress,
     validatorAddress: EthAddress,
     slot: SlotNumber,
     dutyType: DutyType,
@@ -157,6 +162,7 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
   ): Promise<boolean> {
     const result = await this.pool.query(UPDATE_DUTY_SIGNED, [
       signature,
+      rollupAddress.toString(),
       validatorAddress.toString(),
       slot.toString(),
       dutyType,
@@ -166,6 +172,7 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
 
     if (result.rowCount === 0) {
       this.log.warn('Failed to update duty to signed status: invalid token or duty not found', {
+        rollupAddress: rollupAddress.toString(),
         validatorAddress: validatorAddress.toString(),
         slot: slot.toString(),
         dutyType,
@@ -184,6 +191,7 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
    * @returns true if the delete succeeded, false if token didn't match or duty not found
    */
   async deleteDuty(
+    rollupAddress: EthAddress,
     validatorAddress: EthAddress,
     slot: SlotNumber,
     dutyType: DutyType,
@@ -191,6 +199,7 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
     blockIndexWithinCheckpoint: number,
   ): Promise<boolean> {
     const result = await this.pool.query(DELETE_DUTY, [
+      rollupAddress.toString(),
       validatorAddress.toString(),
       slot.toString(),
       dutyType,
@@ -200,6 +209,7 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
 
     if (result.rowCount === 0) {
       this.log.warn('Failed to delete duty: invalid token or duty not found', {
+        rollupAddress: rollupAddress.toString(),
         validatorAddress: validatorAddress.toString(),
         slot: slot.toString(),
         dutyType,
@@ -211,13 +221,17 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
   }
 
   /**
-   * Convert a database row to a ValidatorDutyRecord
+   * Convert a database row to a ValidatorDutyRecord.
+   * Maps snake_case column names to StoredDutyRecord (camelCase, ms timestamps),
+   * then delegates to the shared recordFromFields() converter.
    */
   private rowToRecord(row: DutyRow): ValidatorDutyRecord {
-    return {
-      validatorAddress: EthAddress.fromString(row.validator_address),
-      slot: SlotNumber.fromString(row.slot),
-      blockNumber: BlockNumber.fromString(row.block_number),
+    return recordFromFields({
+      rollupAddress: row.rollup_address,
+      validatorAddress: row.validator_address,
+      slot: row.slot,
+      blockNumber: row.block_number,
+      checkpointNumber: row.checkpoint_number,
       blockIndexWithinCheckpoint: row.block_index_within_checkpoint,
       dutyType: row.duty_type,
       status: row.status,
@@ -225,10 +239,10 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
       signature: row.signature ?? undefined,
       nodeId: row.node_id,
       lockToken: row.lock_token,
-      startedAt: row.started_at,
-      completedAt: row.completed_at ?? undefined,
+      startedAtMs: row.started_at.getTime(),
+      completedAtMs: row.completed_at?.getTime(),
       errorMessage: row.error_message ?? undefined,
-    };
+    });
   }
 
   /**
@@ -244,8 +258,29 @@ export class PostgresSlashingProtectionDatabase implements SlashingProtectionDat
    * @returns the number of duties cleaned up
    */
   async cleanupOwnStuckDuties(nodeId: string, maxAgeMs: number): Promise<number> {
-    const cutoff = new Date(Date.now() - maxAgeMs);
-    const result = await this.pool.query(CLEANUP_OWN_STUCK_DUTIES, [nodeId, cutoff]);
+    const result = await this.pool.query(CLEANUP_OWN_STUCK_DUTIES, [nodeId, maxAgeMs]);
+    return result.rowCount ?? 0;
+  }
+
+  /**
+   * 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.
+   * @returns the number of duties cleaned up
+   */
+  async cleanupOutdatedRollupDuties(currentRollupAddress: EthAddress): Promise<number> {
+    const result = await this.pool.query(CLEANUP_OUTDATED_ROLLUP_DUTIES, [currentRollupAddress.toString()]);
+    return result.rowCount ?? 0;
+  }
+
+  /**
+   * Cleanup old signed duties.
+   * Removes only signed duties older than the specified age.
+   * Does not remove 'signing' duties as they may be in progress.
+   * @returns the number of duties cleaned up
+   */
+  async cleanupOldDuties(maxAgeMs: number): Promise<number> {
+    const result = await this.pool.query(CLEANUP_OLD_DUTIES, [maxAgeMs]);
     return result.rowCount ?? 0;
   }
 }