@aztec/prover-client 0.0.1-commit.ee80a48 → 0.0.1-commit.ef17749e1

package/src/orchestrator/epoch-proving-state.ts

@@ -76,7 +76,7 @@ export class EpochProvingState {
     public readonly epochNumber: EpochNumber,
     public readonly totalNumCheckpoints: number,
     private readonly finalBlobBatchingChallenges: FinalBlobBatchingChallenges,
-    private onCheckpointBlobAccumulatorSet: (checkpoint: CheckpointProvingState) => void,
+    private onCheckpointBlobAccumulatorSet: (checkpoint: CheckpointProvingState) => Promise<void>,
     private completionCallback: (result: ProvingResult) => void,
     private rejectionCallback: (reason: string) => void,
   ) {
@@ -254,9 +254,11 @@ export class EpochProvingState {
       }
       outHashes.push(outHash);
 
-      // Get or create hints for the next checkpoint.
-      hint = checkpoint.getOutHashHintForNextCheckpoint() ?? (await computeOutHashHint(outHashes));
-      checkpoint.setOutHashHintForNextCheckpoint(hint);
+      // If this is NOT the last checkpoint, get or create the hint for the next checkpoint.
+      if (i !== this.totalNumCheckpoints - 1) {
+        hint = checkpoint.getOutHashHintForNextCheckpoint() ?? (await computeOutHashHint(outHashes));
+        checkpoint.setOutHashHintForNextCheckpoint(hint);
+      }
     }
   }
 

package/src/orchestrator/orchestrator.ts

@@ -71,11 +71,6 @@ import { EpochProvingState, type ProvingResult, type TreeSnapshots } from './epo
 import { ProvingOrchestratorMetrics } from './orchestrator_metrics.js';
 import { TxProvingState } from './tx-proving-state.js';
 
-type WorldStateFork = {
-  fork: MerkleTreeWriteOperations;
-  cleanupPromise: Promise<void> | undefined;
-};
-
 /**
  * Implements an event driven proving scheduler to build the recursive proof tree. The idea being:
  * 1. Transactions are provided to the scheduler post simulation.
@@ -97,7 +92,7 @@ export class ProvingOrchestrator implements EpochProver {
   private provingPromise: Promise<ProvingResult> | undefined = undefined;
   private metrics: ProvingOrchestratorMetrics;
   // eslint-disable-next-line aztec-custom/no-non-primitive-in-collections
-  private dbs: Map<BlockNumber, WorldStateFork> = new Map();
+  private dbs: Map<BlockNumber, MerkleTreeWriteOperations> = new Map();
   private logger: Logger;
 
   constructor(
@@ -182,7 +177,7 @@ export class ProvingOrchestrator implements EpochProver {
     const db = await this.dbProvider.fork(lastBlockNumber);
 
     const firstBlockNumber = BlockNumber(lastBlockNumber + 1);
-    this.dbs.set(firstBlockNumber, { fork: db, cleanupPromise: undefined });
+    this.dbs.set(firstBlockNumber, db);
 
     // Get archive sibling path before any block in this checkpoint lands.
     const lastArchiveSiblingPath = await getLastSiblingPath(MerkleTreeId.ARCHIVE, db);
@@ -240,9 +235,9 @@ export class ProvingOrchestrator implements EpochProver {
     if (!this.dbs.has(blockNumber)) {
       // Fork world state at the end of the immediately previous block
       const db = await this.dbProvider.fork(BlockNumber(blockNumber - 1));
-      this.dbs.set(blockNumber, { fork: db, cleanupPromise: undefined });
+      this.dbs.set(blockNumber, db);
     }
-    const db = this.dbs.get(blockNumber)!.fork;
+    const db = this.getDbForBlock(blockNumber);
 
     // Get archive snapshot and sibling path before any txs in this block lands.
     const lastArchiveTreeSnapshot = await getTreeSnapshot(MerkleTreeId.ARCHIVE, db);
@@ -317,7 +312,7 @@ export class ProvingOrchestrator implements EpochProver {
 
     this.logger.info(`Adding ${txs.length} transactions to block ${blockNumber}`);
 
-    const db = this.dbs.get(blockNumber)!.fork;
+    const db = this.getDbForBlock(blockNumber);
     const lastArchive = provingState.lastArchiveTreeSnapshot;
     const newL1ToL2MessageTreeSnapshot = provingState.newL1ToL2MessageTreeSnapshot;
     const spongeBlobState = provingState.getStartSpongeBlob().clone();
@@ -445,14 +440,20 @@ export class ProvingOrchestrator implements EpochProver {
       throw new Error('Block header mismatch');
     }
 
-    // Get db for this block
-    const db = this.dbs.get(provingState.blockNumber)!.fork;
+    // Get db for this block and remove from map — no other code should use it after this point.
+    const db = this.getDbForBlock(provingState.blockNumber);
+    this.dbs.delete(provingState.blockNumber);
 
-    // Update the archive tree, so we're ready to start processing the next block:
-    this.logger.verbose(
-      `Updating archive tree with block ${provingState.blockNumber} header ${(await header.hash()).toString()}`,
-    );
-    await db.updateArchive(header);
+    // Update the archive tree, capture the snapshot, and close the fork deterministically.
+    try {
+      this.logger.verbose(
+        `Updating archive tree with block ${provingState.blockNumber} header ${(await header.hash()).toString()}`,
+      );
+      await db.updateArchive(header);
+      provingState.setBuiltArchive(await getTreeSnapshot(MerkleTreeId.ARCHIVE, db));
+    } finally {
+      await db.close();
+    }
 
     await this.verifyBuiltBlockAgainstSyncedState(provingState);
 
@@ -472,6 +473,13 @@ export class ProvingOrchestrator implements EpochProver {
       this.logger.debug('Block root rollup proof not built yet, skipping header check.');
       return;
     }
+
+    const newArchive = provingState.getBuiltArchive();
+    if (!newArchive) {
+      this.logger.debug('Archive snapshot not yet captured, skipping header check.');
+      return;
+    }
+
     const header = await buildHeaderFromCircuitOutputs(output);
 
     if (!(await header.hash()).equals(await builtBlockHeader.hash())) {
@@ -480,11 +488,7 @@ export class ProvingOrchestrator implements EpochProver {
       return;
     }
 
-    // Get db for this block
     const blockNumber = provingState.blockNumber;
-    const db = this.dbs.get(blockNumber)!.fork;
-
-    const newArchive = await getTreeSnapshot(MerkleTreeId.ARCHIVE, db);
     const syncedArchive = await getTreeSnapshot(MerkleTreeId.ARCHIVE, this.dbProvider.getSnapshot(blockNumber));
     if (!syncedArchive.equals(newArchive)) {
       this.logger.error(
@@ -502,12 +506,6 @@ export class ProvingOrchestrator implements EpochProver {
       provingState.reject(`New archive mismatch.`);
       return;
     }
-
-    // TODO(palla/prover): This closes the fork only on the happy path. If this epoch orchestrator
-    // is aborted and never reaches this point, it will leak the fork. We need to add a global cleanup,
-    // but have to make sure it only runs once all operations are completed, otherwise some function here
-    // will attempt to access the fork after it was closed.
-    void this.cleanupDBFork(blockNumber);
   }
 
   /**
@@ -523,6 +521,19 @@ export class ProvingOrchestrator implements EpochProver {
     }
 
     this.provingState?.cancel();
+
+    for (const [blockNumber, db] of this.dbs.entries()) {
+      void db.close().catch(err => this.logger.error(`Error closing db for block ${blockNumber}`, err));
+    }
+    this.dbs.clear();
+  }
+
+  private getDbForBlock(blockNumber: BlockNumber): MerkleTreeWriteOperations {
+    const db = this.dbs.get(blockNumber);
+    if (!db) {
+      throw new Error(`World state fork for block ${blockNumber} not found.`);
+    }
+    return db;
   }
 
   /**
@@ -554,24 +565,6 @@ export class ProvingOrchestrator implements EpochProver {
     return epochProofResult;
   }
 
-  private async cleanupDBFork(blockNumber: BlockNumber): Promise<void> {
-    this.logger.debug(`Cleaning up world state fork for ${blockNumber}`);
-    const fork = this.dbs.get(blockNumber);
-    if (!fork) {
-      return;
-    }
-
-    try {
-      if (!fork.cleanupPromise) {
-        fork.cleanupPromise = fork.fork.close();
-      }
-      await fork.cleanupPromise;
-      this.dbs.delete(blockNumber);
-    } catch (err) {
-      this.logger.error(`Error closing db for block ${blockNumber}`, err);
-    }
-  }
-
   /**
    * Enqueue a job to be scheduled
    * @param provingState - The proving state object being operated on
@@ -894,17 +887,15 @@ export class ProvingOrchestrator implements EpochProver {
         const leafLocation = provingState.setBlockRootRollupProof(result);
         const checkpointProvingState = provingState.parentCheckpoint;
 
-        // If the proofs were slower than the block header building, then we need to try validating the block header hashes here.
+        // Verification is called from both here and setBlockCompleted. Whichever runs last
+        // will be the first to see all three pieces (header, proof output, archive) and run the checks.
         await this.verifyBuiltBlockAgainstSyncedState(provingState);
 
         if (checkpointProvingState.totalNumBlocks === 1) {
-          this.checkAndEnqueueCheckpointRootRollup(checkpointProvingState);
+          await this.checkAndEnqueueCheckpointRootRollup(checkpointProvingState);
         } else {
-          this.checkAndEnqueueNextBlockMergeRollup(checkpointProvingState, leafLocation);
+          await this.checkAndEnqueueNextBlockMergeRollup(checkpointProvingState, leafLocation);
         }
-
-        // We are finished with the block at this point, ensure the fork is cleaned up
-        void this.cleanupDBFork(provingState.blockNumber);
       },
     );
   }
@@ -1009,14 +1000,14 @@ export class ProvingOrchestrator implements EpochProver {
         },
         signal => this.prover.getBlockMergeRollupProof(inputs, signal, provingState.epochNumber),
       ),
-      result => {
+      async result => {
         provingState.setBlockMergeRollupProof(location, result);
-        this.checkAndEnqueueNextBlockMergeRollup(provingState, location);
+        await this.checkAndEnqueueNextBlockMergeRollup(provingState, location);
       },
     );
   }
 
-  private enqueueCheckpointRootRollup(provingState: CheckpointProvingState) {
+  private async enqueueCheckpointRootRollup(provingState: CheckpointProvingState) {
     if (!provingState.verifyState()) {
       this.logger.debug('Not running checkpoint root rollup. State no longer valid.');
       return;
@@ -1031,7 +1022,7 @@ export class ProvingOrchestrator implements EpochProver {
 
     this.logger.debug(`Enqueuing ${rollupType} for checkpoint ${provingState.index}.`);
 
-    const inputs = provingState.getCheckpointRootRollupInputs();
+    const inputs = await provingState.getCheckpointRootRollupInputs();
 
     this.deferredProving(
       provingState,
@@ -1191,25 +1182,28 @@ export class ProvingOrchestrator implements EpochProver {
     this.enqueueBlockRootRollup(provingState);
   }
 
-  private checkAndEnqueueNextBlockMergeRollup(provingState: CheckpointProvingState, currentLocation: TreeNodeLocation) {
+  private async checkAndEnqueueNextBlockMergeRollup(
+    provingState: CheckpointProvingState,
+    currentLocation: TreeNodeLocation,
+  ) {
     if (!provingState.isReadyForBlockMerge(currentLocation)) {
       return;
     }
 
     const parentLocation = provingState.getParentLocation(currentLocation);
     if (parentLocation.level === 0) {
-      this.checkAndEnqueueCheckpointRootRollup(provingState);
+      await this.checkAndEnqueueCheckpointRootRollup(provingState);
     } else {
       this.enqueueBlockMergeRollup(provingState, parentLocation);
     }
   }
 
-  private checkAndEnqueueCheckpointRootRollup(provingState: CheckpointProvingState) {
+  private async checkAndEnqueueCheckpointRootRollup(provingState: CheckpointProvingState) {
     if (!provingState.isReadyForCheckpointRoot()) {
       return;
     }
 
-    this.enqueueCheckpointRootRollup(provingState);
+    await this.enqueueCheckpointRootRollup(provingState);
   }
 
   private checkAndEnqueueNextCheckpointMergeRollup(provingState: EpochProvingState, currentLocation: TreeNodeLocation) {

package/src/prover-client/prover-client.ts

@@ -29,20 +29,16 @@ export class ProverClient implements EpochProverManager {
   private running = false;
   private agents: ProvingAgent[] = [];
 
-  private proofStore: ProofStore;
-  private failedProofStore: ProofStore | undefined;
-
   private constructor(
     private config: ProverClientConfig,
     private worldState: ForkMerkleTreeOperations & ReadonlyWorldStateAccess,
     private orchestratorClient: ProvingJobProducer,
+    private proofStore: ProofStore,
+    private failedProofStore: ProofStore | undefined,
     private agentClient?: ProvingJobConsumer,
     private telemetry: TelemetryClient = getTelemetryClient(),
     private log: Logger = createLogger('prover-client:tx-prover'),
-  ) {
-    this.proofStore = new InlineProofStore();
-    this.failedProofStore = this.config.failedProofStore ? createProofStore(this.config.failedProofStore) : undefined;
-  }
+  ) {}
 
   public createEpochProver(): EpochProver {
     const bindings = this.log.getBindings();
@@ -118,7 +114,9 @@ export class ProverClient implements EpochProverManager {
     broker: ProvingJobBroker,
     telemetry: TelemetryClient = getTelemetryClient(),
   ) {
-    const prover = new ProverClient(config, worldState, broker, broker, telemetry);
+    const proofStore = await createProofStore(config.proofStore);
+    const failedProofStore = config.failedProofStore ? await createProofStore(config.failedProofStore) : undefined;
+    const prover = new ProverClient(config, worldState, broker, proofStore, failedProofStore, broker, telemetry);
     await prover.start();
     return prover;
   }

package/src/proving_broker/config.ts

@@ -6,8 +6,8 @@ import {
   numberConfigHelper,
 } from '@aztec/foundation/config';
 import { pickConfigMappings } from '@aztec/foundation/config';
-import { type DataStoreConfig, dataConfigMappings } from '@aztec/kv-store/config';
 import { type ChainConfig, chainConfigMappings } from '@aztec/stdlib/config';
+import { type DataStoreConfig, dataConfigMappings } from '@aztec/stdlib/kv-store';
 import { ProvingRequestType } from '@aztec/stdlib/proofs';
 
 import { z } from 'zod';
@@ -31,6 +31,8 @@ export const ProverBrokerConfig = z.object({
   proverBrokerBatchIntervalMs: z.number().int().nonnegative(),
   /** The maximum number of epochs to keep results for */
   proverBrokerMaxEpochsToKeepResultsFor: z.number().int().nonnegative(),
+  /** Enable debug replay mode for replaying proving jobs from stored inputs */
+  proverBrokerDebugReplayEnabled: z.boolean(),
 });
 
 export type ProverBrokerConfig = z.infer<typeof ProverBrokerConfig> &
@@ -74,6 +76,11 @@ export const proverBrokerConfigMappings: ConfigMappingsType<ProverBrokerConfig>
     parseEnv: (val: string | undefined) => (val ? +val : undefined),
     description: "The size of the prover broker's database. Will override the dataStoreMapSizeKb if set.",
   },
+  proverBrokerDebugReplayEnabled: {
+    env: 'PROVER_BROKER_DEBUG_REPLAY_ENABLED',
+    description: 'Enable debug replay mode for replaying proving jobs from stored inputs',
+    ...booleanConfigHelper(false),
+  },
   ...dataConfigMappings,
   ...l1ReaderConfigMappings,
   ...pickConfigMappings(chainConfigMappings, ['rollupVersion']),
@@ -102,6 +109,8 @@ export const ProverAgentConfig = z.object({
   proverTestVerificationDelayMs: z.number().optional(),
   /** Whether to abort pending proving jobs when the orchestrator is cancelled */
   cancelJobsOnStop: z.boolean(),
+  /** Where to store proving results. Must be accessible to both prover node and agents. If not set will inline-encode the parameters */
+  proofStore: z.string().optional(),
 });
 
 export type ProverAgentConfig = z.infer<typeof ProverAgentConfig>;
@@ -162,4 +171,8 @@ export const proverAgentConfigMappings: ConfigMappingsType<ProverAgentConfig> =
       'When false (default), jobs remain in the broker queue and can be reused on restart/reorg.',
     ...booleanConfigHelper(false),
   },
+  proofStore: {
+    env: 'PROVER_PROOF_STORE',
+    description: 'Optional proof input store for the prover',
+  },
 };

package/src/proving_broker/proof_store/factory.ts

@@ -1,42 +1,20 @@
 import { createLogger } from '@aztec/foundation/log';
+import { createFileStore } from '@aztec/stdlib/file-store';
 
-import { GoogleCloudStorageProofStore } from './gcs_proof_store.js';
+import { FileStoreProofStore } from './file_store_proof_store.js';
 import { InlineProofStore } from './inline_proof_store.js';
 import type { ProofStore } from './proof_store.js';
 
-export function createProofStore(config: string | undefined, logger = createLogger('prover-client:proof-store')) {
-  if (config === undefined) {
+export async function createProofStore(
+  config: string | undefined,
+  logger = createLogger('prover-client:proof-store'),
+): Promise<ProofStore> {
+  if (!config) {
     logger.info('Creating inline proof store');
     return new InlineProofStore();
-  } else if (config.startsWith('gs://')) {
-    try {
-      const url = new URL(config);
-      const bucket = url.host;
-      const path = url.pathname.replace(/^\/+/, '');
-      logger.info(`Creating google cloud proof store at ${bucket}`, { bucket, path });
-      return new GoogleCloudStorageProofStore(bucket, path);
-    } catch {
-      throw new Error(
-        `Invalid google cloud proof store definition: '${config}'. Supported values are 'gs://bucket-name/path/to/store'.`,
-      );
-    }
-  } else {
-    throw new Error(`Unknown proof store config: '${config}'. Supported values are 'gs://bucket-name/path/to/store'.`);
   }
-}
 
-export function createProofStoreForUri(
-  uri: string,
-  logger = createLogger('prover-client:proof-store'),
-): Pick<ProofStore, 'getProofInput' | 'getProofOutput'> {
-  if (uri.startsWith('data://')) {
-    return createProofStore(undefined, logger);
-  } else if (uri.startsWith('gs://')) {
-    const url = new URL(uri);
-    const basePath = url.pathname.replace(/^\/+/, '').split('/').slice(0, -3);
-    url.pathname = basePath.join('/');
-    return createProofStore(uri, logger);
-  } else {
-    throw new Error(`Unknown proof store config: '${uri}'. Supported protocols are 'data://' and 'gs://'.`);
-  }
+  const fileStore = await createFileStore(config, logger);
+  logger.info(`Creating file store proof store at ${config}`);
+  return new FileStoreProofStore(fileStore);
 }

package/src/proving_broker/proof_store/file_store_proof_store.ts

@@ -0,0 +1,78 @@
+import { jsonParseWithSchema, jsonStringify } from '@aztec/foundation/json-rpc';
+import type { FileStore } from '@aztec/stdlib/file-store';
+import {
+  type ProofUri,
+  type ProvingJobId,
+  type ProvingJobInputs,
+  type ProvingJobInputsMap,
+  ProvingJobResult,
+  type ProvingJobResultsMap,
+  getProvingJobInputClassFor,
+} from '@aztec/stdlib/interfaces/server';
+import { ProvingRequestType } from '@aztec/stdlib/proofs';
+
+import type { ProofStore } from './proof_store.js';
+
+const INPUTS_PATH = 'inputs';
+const OUTPUTS_PATH = 'outputs';
+
+/**
+ * A proof store implementation backed by a generic FileStore.
+ * Supports any storage backend (GCS, S3, local filesystem) via the FileStore abstraction.
+ */
+export class FileStoreProofStore implements ProofStore {
+  constructor(private readonly fileStore: FileStore) {}
+
+  async saveProofInput<T extends ProvingRequestType>(
+    id: ProvingJobId,
+    type: T,
+    inputs: ProvingJobInputsMap[T],
+  ): Promise<ProofUri> {
+    const path = `${INPUTS_PATH}/${ProvingRequestType[type]}/${id}`;
+    const uri = await this.fileStore.save(path, inputs.toBuffer());
+    return uri as ProofUri;
+  }
+
+  async saveProofOutput<T extends ProvingRequestType>(
+    id: ProvingJobId,
+    type: T,
+    result: ProvingJobResultsMap[T],
+  ): Promise<ProofUri> {
+    const jobResult = { type, result } as ProvingJobResult;
+    const json = jsonStringify(jobResult);
+    const path = `${OUTPUTS_PATH}/${ProvingRequestType[type]}/${id}.json`;
+    const uri = await this.fileStore.save(path, Buffer.from(json, 'utf-8'));
+    return uri as ProofUri;
+  }
+
+  async getProofInput(uri: ProofUri): Promise<ProvingJobInputs> {
+    try {
+      const buffer = await this.fileStore.read(uri);
+      const type = this.extractTypeFromUri(uri);
+      const inputs = getProvingJobInputClassFor(type).fromBuffer(buffer);
+      return { inputs, type } as ProvingJobInputs;
+    } catch (err) {
+      throw new Error(`Error getting proof input at ${uri}: ${err}`);
+    }
+  }
+
+  async getProofOutput(uri: ProofUri): Promise<ProvingJobResult> {
+    try {
+      const buffer = await this.fileStore.read(uri);
+      return jsonParseWithSchema(buffer.toString('utf-8'), ProvingJobResult);
+    } catch (err) {
+      throw new Error(`Error getting proof output at ${uri}: ${err}`);
+    }
+  }
+
+  private extractTypeFromUri(uri: string): ProvingRequestType {
+    const url = new URL(uri);
+    const pathParts = url.pathname.split('/').filter(Boolean);
+    const typeString = pathParts.at(-2);
+    const type = typeString ? ProvingRequestType[typeString as keyof typeof ProvingRequestType] : undefined;
+    if (type === undefined) {
+      throw new Error(`Unrecognized proof type ${typeString} in URI ${uri}`);
+    }
+    return type;
+  }
+}

package/src/proving_broker/proof_store/index.ts

@@ -1,4 +1,4 @@
 export * from './proof_store.js';
 export * from './inline_proof_store.js';
 export * from './factory.js';
-export * from './gcs_proof_store.js';
+export * from './file_store_proof_store.js';

package/src/proving_broker/proving_broker.ts

@@ -7,6 +7,7 @@ import {
   type GetProvingJobResponse,
   type ProofUri,
   type ProvingJob,
+  type ProvingJobBrokerDebug,
   type ProvingJobConsumer,
   type ProvingJobFilter,
   type ProvingJobId,
@@ -36,7 +37,7 @@ type EnqueuedProvingJob = Pick<ProvingJob, 'id' | 'epochNumber'>;
  * A broker that manages proof requests and distributes them to workers based on their priority.
  * It takes a backend that is responsible for storing and retrieving proof requests and results.
  */
-export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Traceable {
+export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, ProvingJobBrokerDebug, Traceable {
   private queues: ProvingQueues = {
     [ProvingRequestType.PUBLIC_VM]: new PriorityMemoryQueue<EnqueuedProvingJob>(provingJobComparator),
     [ProvingRequestType.PUBLIC_CHONK_VERIFIER]: new PriorityMemoryQueue<EnqueuedProvingJob>(provingJobComparator),
@@ -114,6 +115,8 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
 
   private started = false;
 
+  private debugReplayEnabled: boolean;
+
   public constructor(
     private database: ProvingBrokerDatabase,
     {
@@ -121,6 +124,7 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
       proverBrokerPollIntervalMs,
       proverBrokerJobMaxRetries,
       proverBrokerMaxEpochsToKeepResultsFor,
+      proverBrokerDebugReplayEnabled,
     }: Required<
       Pick<
         ProverBrokerConfig,
@@ -128,6 +132,7 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
         | 'proverBrokerPollIntervalMs'
         | 'proverBrokerJobMaxRetries'
         | 'proverBrokerMaxEpochsToKeepResultsFor'
+        | 'proverBrokerDebugReplayEnabled'
       >
     > = defaultProverBrokerConfig,
     client: TelemetryClient = getTelemetryClient(),
@@ -139,6 +144,7 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
     this.jobTimeoutMs = proverBrokerJobTimeoutMs!;
     this.maxRetries = proverBrokerJobMaxRetries!;
     this.maxEpochsToKeepResultsFor = proverBrokerMaxEpochsToKeepResultsFor!;
+    this.debugReplayEnabled = proverBrokerDebugReplayEnabled ?? false;
   }
 
   private measureQueueDepth: MonitorCallback = (type: ProvingRequestType) => {
@@ -241,6 +247,29 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
     return Promise.resolve(this.#reportProvingJobProgress(id, startedAt, filter));
   }
 
+  public async replayProvingJob(
+    jobId: ProvingJobId,
+    type: ProvingRequestType,
+    epochNumber: EpochNumber,
+    inputsUri: ProofUri,
+  ): Promise<ProvingJobStatus> {
+    if (!this.debugReplayEnabled) {
+      throw new Error('Debug replay not enabled. Set PROVER_BROKER_DEBUG_REPLAY_ENABLED=true');
+    }
+
+    this.logger.info(`Replaying proving job`, { provingJobId: jobId, epochNumber, inputsUri });
+
+    // Clear existing state and enqueue
+    this.cleanUpProvingJobState([jobId]);
+
+    const job: ProvingJob = { id: jobId, type, epochNumber, inputsUri };
+    this.jobsCache.set(jobId, job);
+    await this.database.addProvingJob(job);
+    this.enqueueJobInternal(job);
+
+    return { status: 'in-queue' };
+  }
+
   async #enqueueProvingJob(job: ProvingJob): Promise<ProvingJobStatus> {
     // We return the job status at the start of this call
     const jobStatus = this.#getProvingJobStatus(job.id);
@@ -285,7 +314,7 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
     // notify listeners of the cancellation
     if (!this.resultsCache.has(id)) {
       this.logger.info(`Cancelling job id=${id}`, { provingJobId: id });
-      await this.#reportProvingJobError(id, 'Aborted', false);
+      await this.#reportProvingJobError(id, 'Aborted', false, undefined, true);
     }
   }
 
@@ -366,6 +395,7 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
     err: string,
     retry = false,
     filter?: ProvingJobFilter,
+    aborted = false,
   ): Promise<GetProvingJobResponse | undefined> {
     const info = this.inProgress.get(id);
     const item = this.jobsCache.get(id);
@@ -426,7 +456,11 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
     this.promises.get(id)!.resolve(result);
     this.completedJobNotifications.push(id);
 
-    this.instrumentation.incRejectedJobs(item.type);
+    if (aborted) {
+      this.instrumentation.incAbortedJobs(item.type);
+    } else {
+      this.instrumentation.incRejectedJobs(item.type);
+    }
     if (info) {
       const duration = this.msTimeSource() - info.startedAt;
       this.instrumentation.recordJobDuration(item.type, duration);

package/src/proving_broker/proving_broker_instrumentation.ts

@@ -18,6 +18,7 @@ export class ProvingBrokerInstrumentation {
   private activeJobs: ObservableGauge;
   private resolvedJobs: UpDownCounter;
   private rejectedJobs: UpDownCounter;
+  private abortedJobs: UpDownCounter;
   private timedOutJobs: UpDownCounter;
   private cachedJobs: UpDownCounter;
   private totalJobs: UpDownCounter;
@@ -39,6 +40,8 @@ export class ProvingBrokerInstrumentation {
 
     this.rejectedJobs = createUpDownCounterWithDefault(meter, Metrics.PROVING_QUEUE_REJECTED_JOBS, provingJobAttrs);
 
+    this.abortedJobs = createUpDownCounterWithDefault(meter, Metrics.PROVING_QUEUE_ABORTED_JOBS, provingJobAttrs);
+
     this.retriedJobs = createUpDownCounterWithDefault(meter, Metrics.PROVING_QUEUE_RETRIED_JOBS, provingJobAttrs);
 
     this.timedOutJobs = createUpDownCounterWithDefault(meter, Metrics.PROVING_QUEUE_TIMED_OUT_JOBS, provingJobAttrs);
@@ -72,6 +75,12 @@ export class ProvingBrokerInstrumentation {
     });
   }
 
+  incAbortedJobs(proofType: ProvingRequestType) {
+    this.abortedJobs.add(1, {
+      [Attributes.PROVING_JOB_TYPE]: ProvingRequestType[proofType],
+    });
+  }
+
   incRetriedJobs(proofType: ProvingRequestType) {
     this.retriedJobs.add(1, {
       [Attributes.PROVING_JOB_TYPE]: ProvingRequestType[proofType],

package/src/proving_broker/rpc.ts

@@ -1,9 +1,11 @@
+import { EpochNumberSchema } from '@aztec/foundation/branded-types';
 import { createSafeJsonRpcClient } from '@aztec/foundation/json-rpc/client';
 import {
   type GetProvingJobResponse,
   ProofUri,
   ProvingJob,
   type ProvingJobBroker,
+  type ProvingJobBrokerDebug,
   type ProvingJobConsumer,
   ProvingJobId,
   type ProvingJobProducer,
@@ -53,6 +55,18 @@ export const ProvingJobBrokerSchema: ApiSchemaFor<ProvingJobBroker> = {
   ...ProvingJobProducerSchema,
 };
 
+export const ProvingJobBrokerDebugSchema: ApiSchemaFor<ProvingJobBrokerDebug> = {
+  replayProvingJob: z
+    .function()
+    .args(ProvingJobId, z.nativeEnum(ProvingRequestType), EpochNumberSchema, ProofUri)
+    .returns(ProvingJobStatus),
+};
+
+export const ProvingJobBrokerSchemaWithDebug: ApiSchemaFor<ProvingJobBroker & ProvingJobBrokerDebug> = {
+  ...ProvingJobBrokerSchema,
+  ...ProvingJobBrokerDebugSchema,
+};
+
 export function createProvingJobBrokerClient(
   url: string,
   versions: Partial<ComponentsVersions>,