@kb-labs/workflow-engine 2.107.0 → 2.110.0

package/dist/index.d.ts

@@ -1,10 +1,10 @@
 import { WorkflowSpec, RunTrigger, IdempotencyKey, ConcurrencyGroup, WorkflowRun, JobRun, StepRun, RetryPolicy, ArtifactMergeConfig } from '@kb-labs/workflow-contracts';
 import { ILogger, ICache, IEventBus, IAnalytics, ISnapshotManager, Unsubscribe, IJobScheduler, JobDefinition, JobHandle, CronExpression, JobFilter } from '@kb-labs/core-platform';
 import { JobPriority, WorkflowEventName } from '@kb-labs/workflow-constants';
+import { ClassifiedFailure, IExecutionBackend } from '@kb-labs/core-contracts';
 import { ArtifactClient } from '@kb-labs/workflow-artifacts';
 import { IEntityRegistry } from '@kb-labs/core-registry';
 import { PlatformServices, JobHandlerDecl, WorkflowHandlerDecl, CronDecl, PluginContextDescriptor } from '@kb-labs/plugin-contracts';
-import { IExecutionBackend } from '@kb-labs/core-contracts';
 
 /**
  * @deprecated Use ILogger from @kb-labs/core-platform instead.
@@ -229,7 +229,13 @@ declare class WorkflowEngine {
      * Mark job as failed and optionally schedule retry.
      * Implements exponential/linear backoff retry logic.
      */
-    markJobFailed(runId: string, jobId: string, error: Error, shouldRetry?: boolean): Promise<void>;
+    markJobFailed(runId: string, jobId: string, error: Error, failure?: ClassifiedFailure, policyOverride?: false | {
+        max: number;
+        backoff?: 'exp' | 'lin';
+        initialIntervalMs?: number;
+        maxIntervalMs?: number;
+        on?: string[];
+    }): Promise<void>;
     /**
      * Recursively cancel jobs blocked on a failed/cancelled upstream job, so the
      * DAG does not leave dependents stuck in 'queued'. Cancellation cascades:
@@ -300,14 +306,6 @@ declare class WorkflowEngine {
      * Re-queues jobs that were interrupted during previous shutdown.
      */
     resumeInterruptedJobs(): Promise<void>;
-    /**
-     * Determine if job should be retried based on retry policy.
-     */
-    private shouldRetryJob;
-    /**
-     * Calculate backoff delay using exponential or linear strategy.
-     */
-    private calculateBackoff;
     /**
      * Move permanently failed job to Dead Letter Queue.
      */

package/dist/index.js

@@ -4,6 +4,7 @@ import { parse, stringify } from 'yaml';
 import { WorkflowSpecSchema, evaluateExpression } from '@kb-labs/workflow-contracts';
 import { randomUUID } from 'crypto';
 import { WORKFLOW_REDIS_CHANNEL, EVENT_NAMES, IDEMPOTENCY_TTL_ENV, CONCURRENCY_TTL_ENV } from '@kb-labs/workflow-constants';
+import { classifyFailure, decideRetry } from '@kb-labs/core-retry';
 import { createFileSystemArtifactClient } from '@kb-labs/workflow-artifacts';
 import { existsSync } from 'fs';
 import { nanoid } from 'nanoid';
@@ -794,7 +795,7 @@ var WorkflowEngine = class {
    * Mark job as failed and optionally schedule retry.
    * Implements exponential/linear backoff retry logic.
    */
-  async markJobFailed(runId, jobId, error, shouldRetry2 = true) {
+  async markJobFailed(runId, jobId, error, failure, policyOverride) {
     const run = await this.stateStore.getRun(runId);
     if (!run) {
       this.logger.warn("Cannot mark job as failed: run not found", { runId, jobId });
@@ -820,13 +821,26 @@ var WorkflowEngine = class {
       jobId,
       attempt: (job.attempt || 0) + 1
     });
+    const classified = failure ?? classifyFailure(error, { source: "execution", phase: "response" });
+    const retryConfig = policyOverride !== void 0 && typeof policyOverride !== "boolean" ? {
+      maxAttempts: policyOverride.max + 1,
+      retryOn: policyOverride.on ?? ["command", "network", "timeout", "rate_limit", "server", "infrastructure"],
+      initialDelayMs: policyOverride.initialIntervalMs ?? 1e3,
+      backoff: policyOverride.backoff === "lin" ? "linear" : "exponential",
+      maxDelayMs: policyOverride.maxIntervalMs ?? 3e4,
+      respectRetryAfter: true,
+      requireIdempotencyForUnsafeFailures: true
+    } : void 0;
+    const retryDecision = policyOverride === false ? { retry: false, delayMs: 0 } : retryConfig ? decideRetry({ failure: classified, attempt: job.attempt || 0, policy: retryConfig }) : decideRetry({ failure: classified, attempt: job.attempt || 0 });
     this.analytics?.track("workflow.job.failed", {
       runId,
       jobId,
       jobName: job.jobName,
       attempt: (job.attempt || 0) + 1,
       errorMessage: error.message,
-      willRetry: shouldRetry2 && this.shouldRetryJob(job)
+      willRetry: retryDecision.retry,
+      failureKind: classified.kind,
+      failureCode: classified.code
     }).catch(() => {
     });
     await this.events.publish({
@@ -835,8 +849,8 @@ var WorkflowEngine = class {
       jobId,
       payload: { jobName: job.jobName, error: error.message, attempt: (job.attempt || 0) + 1 }
     });
-    if (shouldRetry2 && this.shouldRetryJob(job)) {
-      const backoffMs = this.calculateBackoff(job.attempt || 0, job.retries);
+    if (retryDecision.retry) {
+      const backoffMs = retryDecision.delayMs;
       this.logger.info("Scheduling job retry", {
         runId,
         jobId,
@@ -1295,31 +1309,6 @@ var WorkflowEngine = class {
       this.logger.info("Resumed interrupted jobs", { count: resumedCount });
     }
   }
-  /**
-   * Determine if job should be retried based on retry policy.
-   */
-  shouldRetryJob(job) {
-    const retryPolicy = job.retries || { max: 3};
-    const attempt = job.attempt || 0;
-    return attempt < retryPolicy.max;
-  }
-  /**
-   * Calculate backoff delay using exponential or linear strategy.
-   */
-  calculateBackoff(attempt, policy) {
-    const config = {
-      backoff: policy?.backoff || "exp",
-      initialIntervalMs: policy?.initialIntervalMs || 1e3,
-      maxIntervalMs: policy?.maxIntervalMs || 6e4
-    };
-    let backoffMs;
-    if (config.backoff === "exp") {
-      backoffMs = config.initialIntervalMs * Math.pow(2, attempt);
-    } else {
-      backoffMs = config.initialIntervalMs * (attempt + 1);
-    }
-    return Math.min(backoffMs, config.maxIntervalMs);
-  }
   /**
    * Move permanently failed job to Dead Letter Queue.
    */