@aztec/prover-client 0.0.1-commit.ee80a48 → 0.0.1-commit.f103f88

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);
@@ -598,10 +632,26 @@ export class ProvingBroker implements ProvingJobProducer, ProvingJobConsumer, Tr
       const now = this.msTimeSource();
       const msSinceLastUpdate = now - metadata.lastUpdatedAt;
       if (msSinceLastUpdate >= this.jobTimeoutMs) {
-        this.logger.warn(`Proving job id=${id} timed out. Adding it back to the queue.`, { provingJobId: id });
         this.inProgress.delete(id);
-        this.enqueueJobInternal(item);
         this.instrumentation.incTimedOutJobs(item.type);
+
+        const retries = this.retries.get(id) ?? 0;
+        if (retries + 1 < this.maxRetries && !this.isJobStale(item)) {
+          this.logger.warn(`Proving job id=${id} timed out. Re-enqueueing (retry ${retries + 1}/${this.maxRetries}).`, {
+            provingJobId: id,
+          });
+          this.retries.set(id, retries + 1);
+          this.enqueueJobInternal(item);
+        } else {
+          this.logger.error(`Proving job id=${id} timed out after ${retries + 1} attempts. Marking as failed.`, {
+            provingJobId: id,
+          });
+          const result: ProvingJobSettledResult = { status: 'rejected', reason: 'Timed out' };
+          this.resultsCache.set(id, result);
+          this.promises.get(id)?.resolve(result);
+          this.completedJobNotifications.push(id);
+          this.instrumentation.incRejectedJobs(item.type);
+        }
       }
     }
   }

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,
@@ -16,6 +18,15 @@ import { makeTracedFetch } from '@aztec/telemetry-client';
 
 import { z } from 'zod';
 
+/** Indefinite backoff for broker communication: 1, 1, 1, 2, 4, 4, 4, ... seconds. */
+export function* proverBrokerBackoff() {
+  const v = [1, 1, 1, 2, 4];
+  let i = 0;
+  while (true) {
+    yield v[Math.min(i++, v.length - 1)];
+  }
+}
+
 const ProvingJobFilterSchema = z.object({
   allowList: z.array(z.nativeEnum(ProvingRequestType)),
 });
@@ -53,10 +64,22 @@ 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>,
-  fetch = makeTracedFetch([1, 2, 3], false),
+  fetch = makeTracedFetch(proverBrokerBackoff, false),
 ): ProvingJobBroker {
   return createSafeJsonRpcClient(url, ProvingJobBrokerSchema, {
     namespaceMethods: 'proverBroker',
@@ -68,7 +91,7 @@ export function createProvingJobBrokerClient(
 export function createProvingJobProducerClient(
   url: string,
   versions: Partial<ComponentsVersions>,
-  fetch = makeTracedFetch([1, 2, 3], false),
+  fetch = makeTracedFetch(proverBrokerBackoff, false),
 ): ProvingJobProducer {
   return createSafeJsonRpcClient(url, ProvingJobProducerSchema, {
     namespaceMethods: 'provingJobProducer',
@@ -80,7 +103,7 @@ export function createProvingJobProducerClient(
 export function createProvingJobConsumerClient(
   url: string,
   versions: Partial<ComponentsVersions>,
-  fetch = makeTracedFetch([1, 2, 3], false),
+  fetch = makeTracedFetch(proverBrokerBackoff, false),
 ): ProvingJobConsumer {
   return createSafeJsonRpcClient(url, ProvingJobConsumerSchema, {
     namespaceMethods: 'provingJobConsumer',

package/dest/proving_broker/proof_store/gcs_proof_store.d.ts

@@ -1,14 +0,0 @@
-import { type ProofUri, type ProvingJobId, type ProvingJobInputs, type ProvingJobInputsMap, type ProvingJobResult, type ProvingJobResultsMap } from '@aztec/stdlib/interfaces/server';
-import { ProvingRequestType } from '@aztec/stdlib/proofs';
-import type { ProofStore } from './proof_store.js';
-export declare class GoogleCloudStorageProofStore implements ProofStore {
-    private readonly bucketName;
-    private readonly path;
-    private readonly storage;
-    constructor(bucketName: string, path: string);
-    saveProofInput<T extends ProvingRequestType>(id: ProvingJobId, type: T, inputs: ProvingJobInputsMap[T]): Promise<ProofUri>;
-    saveProofOutput<T extends ProvingRequestType>(_id: ProvingJobId, _type: T, _result: ProvingJobResultsMap[T]): Promise<ProofUri>;
-    getProofInput(uri: ProofUri): Promise<ProvingJobInputs>;
-    getProofOutput(_uri: ProofUri): Promise<ProvingJobResult>;
-}
-//# sourceMappingURL=data:application/json;base64,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

package/dest/proving_broker/proof_store/gcs_proof_store.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"gcs_proof_store.d.ts","sourceRoot":"","sources":["../../../src/proving_broker/proof_store/gcs_proof_store.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,QAAQ,EACb,KAAK,YAAY,EACjB,KAAK,gBAAgB,EACrB,KAAK,mBAAmB,EACxB,KAAK,gBAAgB,EACrB,KAAK,oBAAoB,EAE1B,MAAM,iCAAiC,CAAC;AACzC,OAAO,EAAE,kBAAkB,EAAE,MAAM,sBAAsB,CAAC;AAK1D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAKnD,qBAAa,4BAA6B,YAAW,UAAU;IAI3D,OAAO,CAAC,QAAQ,CAAC,UAAU;IAC3B,OAAO,CAAC,QAAQ,CAAC,IAAI;IAJvB,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAU;IAElC,YACmB,UAAU,EAAE,MAAM,EAClB,IAAI,EAAE,MAAM,EAG9B;IAEY,cAAc,CAAC,CAAC,SAAS,kBAAkB,EACtD,EAAE,EAAE,YAAY,EAChB,IAAI,EAAE,CAAC,EACP,MAAM,EAAE,mBAAmB,CAAC,CAAC,CAAC,GAC7B,OAAO,CAAC,QAAQ,CAAC,CAKnB;IAED,eAAe,CAAC,CAAC,SAAS,kBAAkB,EAC1C,GAAG,EAAE,YAAY,EACjB,KAAK,EAAE,CAAC,EACR,OAAO,EAAE,oBAAoB,CAAC,CAAC,CAAC,GAC/B,OAAO,CAAC,QAAQ,CAAC,CAEnB;IAEY,aAAa,CAAC,GAAG,EAAE,QAAQ,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAsBnE;IAED,cAAc,CAAC,IAAI,EAAE,QAAQ,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAExD;CACF"}

package/dest/proving_broker/proof_store/gcs_proof_store.js

@@ -1,52 +0,0 @@
-import { getProvingJobInputClassFor } from '@aztec/stdlib/interfaces/server';
-import { ProvingRequestType } from '@aztec/stdlib/proofs';
-import { Storage } from '@google-cloud/storage';
-import { join } from 'path';
-const INPUTS_PATH = 'inputs';
-// REFACTOR(#13067): Use the stdlib/file-store instead of referencing google-cloud-storage directly.
-export class GoogleCloudStorageProofStore {
-    bucketName;
-    path;
-    storage;
-    constructor(bucketName, path){
-        this.bucketName = bucketName;
-        this.path = path;
-        this.storage = new Storage();
-    }
-    async saveProofInput(id, type, inputs) {
-        const path = join(this.path, INPUTS_PATH, ProvingRequestType[type], id);
-        const file = this.storage.bucket(this.bucketName).file(path);
-        await file.save(inputs.toBuffer());
-        return file.cloudStorageURI.toString();
-    }
-    saveProofOutput(_id, _type, _result) {
-        throw new Error('Not implemented');
-    }
-    async getProofInput(uri) {
-        try {
-            const url = new URL(uri);
-            const bucket = this.storage.bucket(url.host);
-            const path = url.pathname.replace(/^\/+/, '');
-            const file = bucket.file(path);
-            if (!await file.exists()) {
-                throw new Error(`File at ${uri} does not exist`);
-            }
-            const typeString = path.split('/').at(-2);
-            const type = typeString ? ProvingRequestType[typeString] : undefined;
-            if (type === undefined) {
-                throw new Error(`Unrecognized proof type ${type} in path ${path}`);
-            }
-            const contents = await file.download();
-            const inputs = getProvingJobInputClassFor(type).fromBuffer(contents[0]);
-            return {
-                inputs,
-                type
-            };
-        } catch (err) {
-            throw new Error(`Error getting proof input at ${uri}: ${err}`);
-        }
-    }
-    getProofOutput(_uri) {
-        throw new Error('Not implemented');
-    }
-}

package/src/proving_broker/proof_store/gcs_proof_store.ts

@@ -1,76 +0,0 @@
-import {
-  type ProofUri,
-  type ProvingJobId,
-  type ProvingJobInputs,
-  type ProvingJobInputsMap,
-  type ProvingJobResult,
-  type ProvingJobResultsMap,
-  getProvingJobInputClassFor,
-} from '@aztec/stdlib/interfaces/server';
-import { ProvingRequestType } from '@aztec/stdlib/proofs';
-
-import { Storage } from '@google-cloud/storage';
-import { join } from 'path';
-
-import type { ProofStore } from './proof_store.js';
-
-const INPUTS_PATH = 'inputs';
-
-// REFACTOR(#13067): Use the stdlib/file-store instead of referencing google-cloud-storage directly.
-export class GoogleCloudStorageProofStore implements ProofStore {
-  private readonly storage: Storage;
-
-  constructor(
-    private readonly bucketName: string,
-    private readonly path: string,
-  ) {
-    this.storage = new Storage();
-  }
-
-  public async saveProofInput<T extends ProvingRequestType>(
-    id: ProvingJobId,
-    type: T,
-    inputs: ProvingJobInputsMap[T],
-  ): Promise<ProofUri> {
-    const path = join(this.path, INPUTS_PATH, ProvingRequestType[type], id);
-    const file = this.storage.bucket(this.bucketName).file(path);
-    await file.save(inputs.toBuffer());
-    return file.cloudStorageURI.toString() as ProofUri;
-  }
-
-  saveProofOutput<T extends ProvingRequestType>(
-    _id: ProvingJobId,
-    _type: T,
-    _result: ProvingJobResultsMap[T],
-  ): Promise<ProofUri> {
-    throw new Error('Not implemented');
-  }
-
-  public async getProofInput(uri: ProofUri): Promise<ProvingJobInputs> {
-    try {
-      const url = new URL(uri);
-      const bucket = this.storage.bucket(url.host);
-      const path = url.pathname.replace(/^\/+/, '');
-      const file = bucket.file(path);
-      if (!(await file.exists())) {
-        throw new Error(`File at ${uri} does not exist`);
-      }
-
-      const typeString = path.split('/').at(-2);
-      const type = typeString ? ProvingRequestType[typeString as keyof typeof ProvingRequestType] : undefined;
-      if (type === undefined) {
-        throw new Error(`Unrecognized proof type ${type} in path ${path}`);
-      }
-
-      const contents = await file.download();
-      const inputs = getProvingJobInputClassFor(type).fromBuffer(contents[0]);
-      return { inputs, type } as ProvingJobInputs;
-    } catch (err) {
-      throw new Error(`Error getting proof input at ${uri}: ${err}`);
-    }
-  }
-
-  getProofOutput(_uri: ProofUri): Promise<ProvingJobResult> {
-    throw new Error('Not implemented');
-  }
-}