fix(run-engine): don't mislabel DB errors as UnclassifiableWaitpointId in completeWaitpoint

internal-packages/run-engine/src/engine/tests/completeWaitpointErrorClassification.test.ts

@@ -0,0 +1,82 @@
+// completeWaitpoint's store-selection guard must only turn a genuine id-classification
+// failure into UnclassifiableWaitpointId. forWaitpointCompletion also probes the DB to
+// resolve the owning store, so a transient database/infra error can surface from the same
+// call — and those must bubble up UNCHANGED (keeping their original type, retryability, and
+// error grouping) rather than being mislabelled as an unclassifiable id.
+//
+// This is a hermetic unit test: the error is thrown on the very first line of
+// completeWaitpoint (runStore.forWaitpointCompletion), before any snapshot/enqueue work,
+// so we can drive it with a minimal SystemResources and a fake runStore — no DB, no Redis.
+import { UnclassifiableRunId } from "@trigger.dev/core/v3/isomorphic";
+import { expect } from "vitest";
+import { UnclassifiableWaitpointId } from "../errors.js";
+import type { SystemResources } from "../systems/systems.js";
+import { WaitpointSystem } from "../systems/waitpointSystem.js";
+
+function createWaitpointSystem(forWaitpointCompletion: () => Promise<never>) {
+  const runStore = { forWaitpointCompletion };
+
+  const resources = {
+    runStore,
+    logger: {
+      error: vi.fn(),
+      warn: vi.fn(),
+      info: vi.fn(),
+      debug: vi.fn(),
+    },
+  } as unknown as SystemResources;
+
+  return new WaitpointSystem({
+    resources,
+    // Never reached on the store-resolution error path.
+    executionSnapshotSystem: {} as any,
+    enqueueSystem: {} as any,
+  });
+}
+
+describe("completeWaitpoint store-resolution error classification", () => {
+  it("rethrows a transient database error unchanged (never wraps it as UnclassifiableWaitpointId)", async () => {
+    const dbError = new Error("Can't reach database server at db:5432");
+    const waitpointSystem = createWaitpointSystem(() => Promise.reject(dbError));
+
+    // The original error bubbles up as-is...
+    await expect(waitpointSystem.completeWaitpoint({ id: "waitpoint_transient" })).rejects.toBe(
+      dbError
+    );
+    // ...and is NOT relabelled as a classification failure.
+    await expect(
+      waitpointSystem.completeWaitpoint({ id: "waitpoint_transient" })
+    ).rejects.not.toBeInstanceOf(UnclassifiableWaitpointId);
+  });
+
+  it("wraps a classification failure from a foreign module instance (matched by name, not instanceof)", async () => {
+    const waitpointId = "waitpoint_foreign_instance";
+    const foreignError = new Error(`Unclassifiable run-ops id: ${waitpointId}`);
+    foreignError.name = "UnclassifiableRunId";
+    const waitpointSystem = createWaitpointSystem(() => Promise.reject(foreignError));
+
+    const caught = (await waitpointSystem
+      .completeWaitpoint({ id: waitpointId })
+      .catch((error: unknown) => error)) as UnclassifiableWaitpointId;
+    expect(caught).toBeInstanceOf(UnclassifiableWaitpointId);
+    expect(caught.waitpointId).toBe(waitpointId);
+    expect(caught.cause).toBe(foreignError);
+  });
+
+  it("wraps a genuine UnclassifiableRunId as UnclassifiableWaitpointId with the original as cause", async () => {
+    const waitpointId = "waitpoint_unclassifiable";
+    const classificationError = new UnclassifiableRunId(waitpointId);
+    const waitpointSystem = createWaitpointSystem(() => Promise.reject(classificationError));
+
+    await expect(waitpointSystem.completeWaitpoint({ id: waitpointId })).rejects.toBeInstanceOf(
+      UnclassifiableWaitpointId
+    );
+
+    const caught = (await waitpointSystem
+      .completeWaitpoint({ id: waitpointId })
+      .catch((error: unknown) => error)) as UnclassifiableWaitpointId;
+    expect(caught).toBeInstanceOf(UnclassifiableWaitpointId);
+    expect(caught.waitpointId).toBe(waitpointId);
+    expect(caught.cause).toBe(classificationError);
+  });
+});

internal-packages/run-engine/src/engine/waitpointCoordinator/legacyPostgresCoordinator.ts

@@ -1,6 +1,10 @@
 import type { RunStore } from "@internal/run-store";
 import { tryCatch } from "@trigger.dev/core/v3";
-import { mintWaitpointIdFor, mintWaitpointIdForShard } from "@trigger.dev/core/v3/isomorphic";
+import {
+  mintWaitpointIdFor,
+  mintWaitpointIdForShard,
+  UnclassifiableRunId,
+} from "@trigger.dev/core/v3/isomorphic";
 import type { Logger } from "@trigger.dev/core/logger";
 import type { PrismaClient, Waitpoint } from "@trigger.dev/database";
 import { boundedIn, Prisma } from "@trigger.dev/database";
@@ -118,11 +122,28 @@ export class LegacyPostgresWaitpointCoordinator implements WaitpointCoordinator
     try {
       store = await this.runStore.forWaitpointCompletion(waitpointId, { routeKind: "MANUAL" });
     } catch (error) {
-      this.logger.error("completeWaitpoint: unclassifiable waitpointId", {
+      // Only a genuine id-classification failure should become UnclassifiableWaitpointId.
+      // forWaitpointCompletion also probes the DB to resolve the owning store, so a transient
+      // database/infra error (e.g. can't reach the database) can surface here too. Those MUST
+      // bubble up unchanged so they keep their original type, retryability, and error grouping
+      // instead of being mislabelled as an unclassifiable id.
+      const isClassificationFailure =
+        error instanceof UnclassifiableRunId ||
+        (error instanceof Error && error.name === "UnclassifiableRunId");
+
+      if (isClassificationFailure) {
+        this.logger.error("completeWaitpoint: unclassifiable waitpointId", {
+          waitpointId,
+          error,
+        });
+        throw new UnclassifiableWaitpointId(waitpointId, { cause: error });
+      }
+
+      this.logger.error("completeWaitpoint: error resolving waitpoint store", {
         waitpointId,
         error,
       });
-      throw new UnclassifiableWaitpointId(waitpointId, { cause: error });
+      throw error;
     }
 
     // 1. Complete the Waitpoint (if not completed)