fix(scheduler): skip legacy keyless entries in reconciliation (#403)

* fix(scheduler): skip legacy keyless entries in reconciliation

Pre-BullMQ-5 repeat zset members (MD5-style, no scheduler hash — ~95 in
prod) make getJobSchedulers() yield undefined/keyless entries. The first
v0.34.11 prod boot TypeError'd on one, aborting the whole reconciliation
pass (caught by the outer guard, so boot was unaffected, but reconciliation
was inert). Skip such entries — they are exactly the "foreign, never touch"
case — and keep processing real entries after them.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* style: prettier

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: therealbrad

testplanit/scheduler.reconcile.test.ts

@@ -177,4 +177,47 @@ describe("reconcileStaleSchedulers", () => {
       )
     ).resolves.toBeUndefined();
   });
+
+  it("skips undefined/keyless entries from legacy repeat-zset members", async () => {
+    // Pre-BullMQ-5 repeat members (MD5-style, no scheduler hash) make
+    // getJobSchedulers() yield undefined or keyless entries. Prod incident
+    // 2026-06-05: one undefined entry TypeError'd the whole reconciliation
+    // pass. Stale entries AFTER the bad ones must still be processed.
+    const removed: string[] = [];
+    const q = {
+      name: "forecast-updates",
+      getJobSchedulers: vi.fn(async () => [
+        undefined,
+        null,
+        { name: JOB }, // keyless
+        { key: null, name: JOB },
+        { key: `${JOB}-gone`, name: JOB }, // real stale entry after the junk
+      ]),
+      removeJobScheduler: vi.fn(async (id: string) => {
+        removed.push(id);
+      }),
+    };
+    await expect(
+      reconcileStaleSchedulers(
+        [{ queue: q as never, jobNames: [JOB] }],
+        new Set<string>()
+      )
+    ).resolves.toBeUndefined();
+    expect(removed).toEqual([`${JOB}-gone`]);
+  });
+
+  it("tolerates getJobSchedulers resolving to null", async () => {
+    const q = {
+      name: "forecast-updates",
+      getJobSchedulers: vi.fn(async () => null),
+      removeJobScheduler: vi.fn(),
+    };
+    await expect(
+      reconcileStaleSchedulers(
+        [{ queue: q as never, jobNames: [JOB] }],
+        new Set<string>()
+      )
+    ).resolves.toBeUndefined();
+    expect(q.removeJobScheduler).not.toHaveBeenCalled();
+  });
 });

testplanit/scheduler.ts

@@ -56,7 +56,9 @@ export async function reconcileStaleSchedulers(
         start?: number,
         end?: number,
         asc?: boolean
-      ) => Promise<Array<{ key: string; name: string }>>;
+      ) => Promise<
+        Array<{ key?: string | null; name?: string } | undefined | null>
+      >;
       removeJobScheduler: (id: string) => Promise<unknown>;
       name: string;
     } | null;
@@ -78,8 +80,14 @@ export async function reconcileStaleSchedulers(
       continue;
     }
 
-    for (const scheduler of schedulers) {
-      const schedulerId = scheduler.key;
+    for (const scheduler of schedulers ?? []) {
+      // Legacy (pre-BullMQ-5) repeat zset members have no scheduler hash, so
+      // getJobSchedulers() yields undefined/keyless entries for them — seen
+      // in prod 2026-06-05 (95 MD5-style members), where one such entry
+      // TypeError'd the whole reconciliation pass. Skip them; they are
+      // exactly the "foreign — never touch" case.
+      const schedulerId = scheduler?.key;
+      if (typeof schedulerId !== "string" || schedulerId.length === 0) continue;
       // Match the longest job name first so e.g. a hypothetical
       // "send-daily-digest-summary" job is not mistaken for
       // "send-daily-digest" with tenant "summary".