@aztec/validator-ha-signer 0.0.1-commit.fffb133c → 0.0.6-commit.a2d1860fe9

package/src/db/schema.ts

@@ -16,12 +16,13 @@ export const SCHEMA_VERSION = 1;
  */
 export const CREATE_VALIDATOR_DUTIES_TABLE = `
 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', '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,
@@ -30,7 +31,7 @@ CREATE TABLE IF NOT EXISTS validator_duties (
   completed_at TIMESTAMP,
   error_message TEXT,
 
-  PRIMARY KEY (validator_address, slot, duty_type, block_index_within_checkpoint),
+  PRIMARY KEY (rollup_address, validator_address, slot, duty_type, block_index_within_checkpoint),
   CHECK (completed_at IS NULL OR completed_at >= started_at)
 );
 `;
@@ -101,6 +102,7 @@ SELECT version FROM schema_version ORDER BY version DESC LIMIT 1;
 export const INSERT_OR_GET_DUTY = `
 WITH inserted AS (
   INSERT INTO validator_duties (
+    rollup_address,
     validator_address,
     slot,
     block_number,
@@ -111,9 +113,10 @@ WITH inserted AS (
     node_id,
     lock_token,
     started_at
-  ) VALUES ($1, $2, $3, $4, $5, 'signing', $6, $7, $8, CURRENT_TIMESTAMP)
-  ON CONFLICT (validator_address, slot, duty_type, block_index_within_checkpoint) DO NOTHING
+  ) VALUES ($1, $2, $3, $4, $5, $6, 'signing', $7, $8, $9, 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,
@@ -132,6 +135,7 @@ WITH inserted AS (
 SELECT * FROM inserted
 UNION ALL
 SELECT
+  rollup_address,
   validator_address,
   slot,
   block_number,
@@ -147,10 +151,11 @@ SELECT
   error_message,
   FALSE as is_new
 FROM validator_duties
-WHERE validator_address = $1
-  AND slot = $2
-  AND duty_type = $5
-  AND block_index_within_checkpoint = $4
+WHERE rollup_address = $1
+  AND validator_address = $2
+  AND slot = $3
+  AND duty_type = $6
+  AND block_index_within_checkpoint = $5
   AND NOT EXISTS (SELECT 1 FROM inserted);
 `;
 
@@ -162,12 +167,13 @@ UPDATE validator_duties
 SET status = 'signed',
     signature = $1,
     completed_at = CURRENT_TIMESTAMP
-WHERE validator_address = $2
-  AND slot = $3
-  AND duty_type = $4
-  AND block_index_within_checkpoint = $5
+WHERE rollup_address = $2
+  AND validator_address = $3
+  AND slot = $4
+  AND duty_type = $5
+  AND block_index_within_checkpoint = $6
   AND status = 'signing'
-  AND lock_token = $6;
+  AND lock_token = $7;
 `;
 
 /**
@@ -176,12 +182,13 @@ WHERE validator_address = $2
  */
 export const DELETE_DUTY = `
 DELETE FROM validator_duties
-WHERE validator_address = $1
-  AND slot = $2
-  AND duty_type = $3
-  AND block_index_within_checkpoint = $4
+WHERE rollup_address = $1
+  AND validator_address = $2
+  AND slot = $3
+  AND duty_type = $4
+  AND block_index_within_checkpoint = $5
   AND status = 'signing'
-  AND lock_token = $5;
+  AND lock_token = $6;
 `;
 
 /**
@@ -196,23 +203,34 @@ 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;
 `;
 
 /**
@@ -231,6 +249,7 @@ export const DROP_SCHEMA_VERSION_TABLE = `DROP TABLE IF EXISTS schema_version;`;
  */
 export const GET_STUCK_DUTIES = `
 SELECT
+  rollup_address,
   validator_address,
   slot,
   block_number,

package/src/db/types.ts

@@ -6,6 +6,7 @@ import type { Signature } from '@aztec/foundation/eth-signature';
  * Row type from PostgreSQL query
  */
 export interface DutyRow {
+  rollup_address: string;
   validator_address: string;
   slot: string;
   block_number: string;
@@ -54,6 +55,8 @@ export enum DutyStatus {
  * Record of a validator duty in the database
  */
 export interface ValidatorDutyRecord {
+  /** Ethereum address of the rollup contract */
+  rollupAddress: EthAddress;
   /** Ethereum address of the validator */
   validatorAddress: EthAddress;
   /** Slot number for this duty */
@@ -78,7 +81,7 @@ 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;
 }
 
@@ -87,6 +90,7 @@ export interface ValidatorDutyRecord {
  * blockIndexWithinCheckpoint is REQUIRED and must be >= 0.
  */
 export interface BlockProposalDutyIdentifier {
+  rollupAddress: EthAddress;
   validatorAddress: EthAddress;
   slot: SlotNumber;
   /** Block index within checkpoint (0, 1, 2...). Required for block proposals. */
@@ -99,6 +103,7 @@ export interface BlockProposalDutyIdentifier {
  * blockIndexWithinCheckpoint is not applicable (internally stored as -1).
  */
 export interface OtherDutyIdentifier {
+  rollupAddress: EthAddress;
   validatorAddress: EthAddress;
   slot: SlotNumber;
   dutyType:

package/src/slashing_protection_service.ts

@@ -40,6 +40,7 @@ export class SlashingProtectionService {
   private readonly maxStuckDutiesAgeMs: number;
 
   private cleanupRunningPromise: RunningPromise;
+  private lastOldDutiesCleanupAtMs?: number;
 
   constructor(
     private readonly db: SlashingProtectionDatabase,
@@ -51,11 +52,7 @@ export class SlashingProtectionService {
     // Default to 144s (2x 72s Aztec slot duration) if not explicitly configured
     this.maxStuckDutiesAgeMs = config.maxStuckDutiesAgeMs ?? 144_000;
 
-    this.cleanupRunningPromise = new RunningPromise(
-      this.cleanupStuckDuties.bind(this),
-      this.log,
-      this.maxStuckDutiesAgeMs,
-    );
+    this.cleanupRunningPromise = new RunningPromise(this.cleanup.bind(this), this.log, this.maxStuckDutiesAgeMs);
   }
 
   /**
@@ -67,7 +64,6 @@ export class SlashingProtectionService {
    * 2. If insert succeeds, we acquired the lock - return the lockToken
    * 3. If a record exists, handle based on status:
    *    - SIGNED: Throw appropriate error (already signed or slashing protection)
-   *    - FAILED: Delete the failed record
    *    - SIGNING: Wait and poll until status changes, then handle result
    *
    * @returns The lockToken that must be used for recordSuccess/deleteDuty
@@ -149,10 +145,11 @@ export class SlashingProtectionService {
    * @returns true if the update succeeded, false if token didn't match
    */
   async recordSuccess(params: RecordSuccessParams): Promise<boolean> {
-    const { validatorAddress, slot, dutyType, signature, nodeId, lockToken } = params;
+    const { rollupAddress, validatorAddress, slot, dutyType, signature, nodeId, lockToken } = params;
     const blockIndexWithinCheckpoint = getBlockIndexFromDutyIdentifier(params);
 
     const success = await this.db.updateDutySigned(
+      rollupAddress,
       validatorAddress,
       slot,
       dutyType,
@@ -184,10 +181,17 @@ export class SlashingProtectionService {
    * @returns true if the delete succeeded, false if token didn't match
    */
   async deleteDuty(params: DeleteDutyParams): Promise<boolean> {
-    const { validatorAddress, slot, dutyType, lockToken } = params;
+    const { rollupAddress, validatorAddress, slot, dutyType, lockToken } = params;
     const blockIndexWithinCheckpoint = getBlockIndexFromDutyIdentifier(params);
 
-    const success = await this.db.deleteDuty(validatorAddress, slot, dutyType, lockToken, blockIndexWithinCheckpoint);
+    const success = await this.db.deleteDuty(
+      rollupAddress,
+      validatorAddress,
+      slot,
+      dutyType,
+      lockToken,
+      blockIndexWithinCheckpoint,
+    );
 
     if (success) {
       this.log.info(`Deleted duty ${dutyType} at slot ${slot} to allow retry`, {
@@ -213,7 +217,19 @@ export class SlashingProtectionService {
    * Start running tasks.
    * Cleanup runs immediately on start to recover from any previous crashes.
    */
-  start() {
+  /**
+   * Start the background cleanup task.
+   * Also performs one-time cleanup of duties with outdated rollup addresses.
+   */
+  async start() {
+    // One-time cleanup at startup: remove duties from previous rollup versions
+    const numOutdatedRollupDuties = await this.db.cleanupOutdatedRollupDuties(this.config.l1Contracts.rollupAddress);
+    if (numOutdatedRollupDuties > 0) {
+      this.log.info(`Cleaned up ${numOutdatedRollupDuties} duties with outdated rollup address at startup`, {
+        currentRollupAddress: this.config.l1Contracts.rollupAddress.toString(),
+      });
+    }
+
     this.cleanupRunningPromise.start();
     this.log.info('Slashing protection service started', { nodeId: this.config.nodeId });
   }
@@ -236,15 +252,36 @@ export class SlashingProtectionService {
   }
 
   /**
-   * Cleanup own stuck duties
+   * Periodic cleanup of stuck duties and optionally old signed duties.
+   * Runs in the background via RunningPromise.
    */
-  private async cleanupStuckDuties() {
-    const numDuties = await this.db.cleanupOwnStuckDuties(this.config.nodeId, this.maxStuckDutiesAgeMs);
-    if (numDuties > 0) {
-      this.log.info(`Cleaned up ${numDuties} stuck duties`, {
+  private async cleanup() {
+    // 1. Clean up stuck duties (our own node's duties that got stuck in 'signing' status)
+    const numStuckDuties = await this.db.cleanupOwnStuckDuties(this.config.nodeId, this.maxStuckDutiesAgeMs);
+    if (numStuckDuties > 0) {
+      this.log.verbose(`Cleaned up ${numStuckDuties} stuck duties`, {
         nodeId: this.config.nodeId,
         maxStuckDutiesAgeMs: this.maxStuckDutiesAgeMs,
       });
     }
+
+    // 2. Clean up old signed duties if configured
+    // we shouldn't run this as often as stuck duty cleanup.
+    if (this.config.cleanupOldDutiesAfterHours !== undefined) {
+      const maxAgeMs = this.config.cleanupOldDutiesAfterHours * 60 * 60 * 1000;
+      const nowMs = Date.now();
+      const shouldRun =
+        this.lastOldDutiesCleanupAtMs === undefined || nowMs - this.lastOldDutiesCleanupAtMs >= maxAgeMs;
+      if (shouldRun) {
+        const numOldDuties = await this.db.cleanupOldDuties(maxAgeMs);
+        this.lastOldDutiesCleanupAtMs = nowMs;
+        if (numOldDuties > 0) {
+          this.log.verbose(`Cleaned up ${numOldDuties} old signed duties`, {
+            cleanupOldDutiesAfterHours: this.config.cleanupOldDutiesAfterHours,
+            maxAgeMs,
+          });
+        }
+      }
+    }
   }
 }

package/src/types.ts

@@ -14,6 +14,7 @@ import {
   type CheckAndRecordParams,
   type DeleteDutyParams,
   type DutyIdentifier,
+  type DutyRow,
   DutyType,
   type OtherDutyIdentifier,
   type RecordSuccessParams,
@@ -25,6 +26,7 @@ export type {
   CheckAndRecordParams,
   DeleteDutyParams,
   DutyIdentifier,
+  DutyRow,
   OtherDutyIdentifier,
   RecordSuccessParams,
   ValidatorDutyRecord,
@@ -170,6 +172,7 @@ export interface SlashingProtectionDatabase {
    * @returns true if the update succeeded, false if token didn't match or duty not found
    */
   updateDutySigned(
+    rollupAddress: EthAddress,
     validatorAddress: EthAddress,
     slot: SlotNumber,
     dutyType: DutyType,
@@ -186,6 +189,7 @@ export interface SlashingProtectionDatabase {
    * @returns true if the delete succeeded, false if token didn't match or duty not found
    */
   deleteDuty(
+    rollupAddress: EthAddress,
     validatorAddress: EthAddress,
     slot: SlotNumber,
     dutyType: DutyType,
@@ -199,6 +203,21 @@ export interface SlashingProtectionDatabase {
    */
   cleanupOwnStuckDuties(nodeId: string, maxAgeMs: number): Promise<number>;
 
+  /**
+   * 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
+   */
+  cleanupOutdatedRollupDuties(currentRollupAddress: EthAddress): Promise<number>;
+
+  /**
+   * Cleanup old signed duties.
+   * Removes only signed duties older than the specified age.
+   * @returns the number of duties cleaned up
+   */
+  cleanupOldDuties(maxAgeMs: number): Promise<number>;
+
   /**
    * Close the database connection.
    * Should be called during graceful shutdown.

package/src/validator_ha_signer.ts

@@ -6,7 +6,7 @@
  * node will sign for a given duty (slot + duty type).
  */
 import type { Buffer32 } from '@aztec/foundation/buffer';
-import type { EthAddress } from '@aztec/foundation/eth-address';
+import { EthAddress } from '@aztec/foundation/eth-address';
 import type { Signature } from '@aztec/foundation/eth-signature';
 import { type Logger, createLogger } from '@aztec/foundation/log';
 
@@ -41,6 +41,7 @@ import {
 export class ValidatorHASigner {
   private readonly log: Logger;
   private readonly slashingProtection: SlashingProtectionService;
+  private readonly rollupAddress: EthAddress;
 
   constructor(
     db: SlashingProtectionDatabase,
@@ -56,9 +57,11 @@ export class ValidatorHASigner {
     if (!config.nodeId || config.nodeId === '') {
       throw new Error('NODE_ID is required for high-availability setups');
     }
+    this.rollupAddress = config.l1Contracts.rollupAddress;
     this.slashingProtection = new SlashingProtectionService(db, config);
     this.log.info('Validator HA Signer initialized with slashing protection', {
       nodeId: config.nodeId,
+      rollupAddress: this.rollupAddress.toString(),
     });
   }
 
@@ -88,6 +91,7 @@ export class ValidatorHASigner {
     let dutyIdentifier: DutyIdentifier;
     if (context.dutyType === DutyType.BLOCK_PROPOSAL) {
       dutyIdentifier = {
+        rollupAddress: this.rollupAddress,
         validatorAddress,
         slot: context.slot,
         blockIndexWithinCheckpoint: context.blockIndexWithinCheckpoint,
@@ -95,6 +99,7 @@ export class ValidatorHASigner {
       };
     } else {
       dutyIdentifier = {
+        rollupAddress: this.rollupAddress,
         validatorAddress,
         slot: context.slot,
         dutyType: context.dutyType,
@@ -142,8 +147,8 @@ export class ValidatorHASigner {
    * Start the HA signer background tasks (cleanup of stuck duties).
    * Should be called after construction and before signing operations.
    */
-  start() {
-    this.slashingProtection.start();
+  async start() {
+    await this.slashingProtection.start();
   }
 
   /**