@kb-labs/workflow-engine 2.96.0 → 2.98.0

package/dist/index.d.ts

@@ -320,6 +320,18 @@ declare class WorkflowEngine {
     private deadLetterJob;
     updateRun(runId: string, mutator: (run: WorkflowRun) => WorkflowRun | void): Promise<WorkflowRun | null>;
     finalizeRun(runId: string, status: WorkflowRun['status'], context?: Partial<RunContext>): Promise<WorkflowRun | null>;
+    /**
+     * Persist a snapshot for a run that just reached a terminal state, so it
+     * can be replayed from any step later (`replayRun`, used by both
+     * `workflow runs restart --from-step` and `rerun --failed-only`). Only
+     * 'success'/'failed' carry meaningful step output data worth snapshotting.
+     *
+     * Shared by finalizeRun() and checkRunCompletion() — checkRunCompletion is
+     * the run-completion path actually driven by the worker in production, so
+     * without this call here, no snapshot is ever created outside of tests
+     * that call finalizeRun() directly (issue #263).
+     */
+    private snapshotTerminalRun;
     nextJob(): Promise<JobQueueEntry | null>;
     rescheduleJob(entry: JobQueueEntry, delayMs: number): Promise<void>;
     publishRunEvent(type: WorkflowEventName, run: WorkflowRun): Promise<void>;

package/dist/index.js

@@ -1038,7 +1038,7 @@ var WorkflowEngine = class {
       return;
     }
     if (allSuccess) {
-      await this.stateStore.updateRun(runId, (draft) => {
+      const updated = await this.stateStore.updateRun(runId, (draft) => {
         draft.status = "success";
         draft.finishedAt = (/* @__PURE__ */ new Date()).toISOString();
         return draft;
@@ -1056,9 +1056,12 @@ var WorkflowEngine = class {
         runId,
         payload: { status: "success", name: run.name }
       });
+      if (updated) {
+        await this.snapshotTerminalRun(updated);
+      }
     } else if (anyFailed) {
       const failedJob = run.jobs.find((j) => j.status === "failed");
-      await this.stateStore.updateRun(runId, (draft) => {
+      const updated = await this.stateStore.updateRun(runId, (draft) => {
         draft.status = "failed";
         draft.finishedAt = (/* @__PURE__ */ new Date()).toISOString();
         if (failedJob?.error) {
@@ -1086,6 +1089,9 @@ var WorkflowEngine = class {
         runId,
         payload: { status: "failed", name: run.name }
       });
+      if (updated) {
+        await this.snapshotTerminalRun(updated);
+      }
     }
   }
   /**
@@ -1370,21 +1376,40 @@ var WorkflowEngine = class {
         status === "failed" ? EVENT_NAMES.run.failed : status === "cancelled" ? EVENT_NAMES.run.cancelled : EVENT_NAMES.run.finished,
         updated
       );
-      if (status === "success" || status === "failed") {
-        const stepOutputs = {};
-        for (const job of updated.jobs) {
-          for (const step of job.steps) {
-            if (step.outputs && Object.keys(step.outputs).length > 0) {
-              stepOutputs[step.id] = step.outputs;
-            }
-          }
+      await this.snapshotTerminalRun(updated);
+    }
+    return updated;
+  }
+  /**
+   * Persist a snapshot for a run that just reached a terminal state, so it
+   * can be replayed from any step later (`replayRun`, used by both
+   * `workflow runs restart --from-step` and `rerun --failed-only`). Only
+   * 'success'/'failed' carry meaningful step output data worth snapshotting.
+   *
+   * Shared by finalizeRun() and checkRunCompletion() — checkRunCompletion is
+   * the run-completion path actually driven by the worker in production, so
+   * without this call here, no snapshot is ever created outside of tests
+   * that call finalizeRun() directly (issue #263).
+   */
+  async snapshotTerminalRun(run) {
+    if (run.status !== "success" && run.status !== "failed") {
+      return;
+    }
+    const stepOutputs = {};
+    for (const job of run.jobs) {
+      for (const step of job.steps) {
+        if (step.outputs && Object.keys(step.outputs).length > 0) {
+          stepOutputs[step.id] = step.outputs;
         }
-        await this.snapshotStorage.createSnapshot(updated, stepOutputs, updated.env ?? {}).catch((err) => {
-          this.logger.warn("Failed to create run snapshot", { runId, error: err instanceof Error ? err.message : String(err) });
-        });
       }
     }
-    return updated;
+    await this.snapshotStorage.createSnapshot(run, stepOutputs, run.env ?? {}).catch((err) => {
+      this.logger.error(
+        "Failed to create run snapshot \u2014 restart --from-step and rerun --failed-only will not work for this run",
+        err instanceof Error ? err : new Error(String(err)),
+        { runId: run.id, status: run.status }
+      );
+    });
   }
   async nextJob() {
     return this.scheduler.dequeueJob();