From 16fba65cb6ad1d0ecfebbfbbd8f3b24eab7d1541 Mon Sep 17 00:00:00 2001
From: Alix-007 <li.long15@xydigit.com>
Date: Sat, 20 Jun 2026 03:35:42 +0800
Subject: [PATCH] fix(cron): honor configured retry.backoffMs for recurring
 error backoff floor (#93051)

Merged via squash.

Prepared head SHA: c8026d0aef2e6acfac1a752fecab0a302fb14e34
Co-authored-by: Alix-007 <267018309+Alix-007@users.noreply.github.com>
Co-authored-by: steipete <58493+steipete@users.noreply.github.com>
Reviewed-by: @steipete
---
 .../timer.backoff-config-readback.test.ts     | 80 +++++++++++++++++++
 src/cron/service/timer.ts                     |  5 +-
 2 files changed, 84 insertions(+), 1 deletion(-)
 create mode 100644 src/cron/service/timer.backoff-config-readback.test.ts

diff --git a/src/cron/service/timer.backoff-config-readback.test.ts b/src/cron/service/timer.backoff-config-readback.test.ts
new file mode 100644
index 000000000000..ab443f038b77
--- /dev/null
+++ b/src/cron/service/timer.backoff-config-readback.test.ts
@@ -0,0 +1,80 @@
+// Verifies the configured retry.backoffMs floor for a recurring job survives a
+// real cron service run and is persisted to the SQLite-backed store, not just
+// computed in memory by applyJobResult.
+import path from "node:path";
+import { describe, expect, it, vi } from "vitest";
+import { resetTaskRegistryForTests } from "../../tasks/task-registry.js";
+import { withEnvAsync } from "../../test-utils/env.js";
+import { setupCronServiceSuite, writeCronStoreSnapshot } from "../service.test-harness.js";
+import { loadCronStore } from "../store.js";
+import type { CronJob } from "../types.js";
+import { run } from "./ops.js";
+import { createCronServiceState } from "./state.js";
+
+const { logger, makeStorePath } = setupCronServiceSuite({
+  prefix: "cron-backoff-config-readback",
+});
+
+function createDueRecurringJob(now: number): CronJob {
+  return {
+    id: "recurring-backoff-readback",
+    name: "recurring backoff readback",
+    enabled: true,
+    createdAtMs: now - 60_000,
+    updatedAtMs: now - 60_000,
+    schedule: { kind: "every", everyMs: 1_000, anchorMs: now - 60_000 },
+    sessionTarget: "isolated",
+    wakeMode: "next-heartbeat",
+    payload: { kind: "agentTurn", message: "ping" },
+    sessionKey: "agent:main:main",
+    state: { nextRunAtMs: now - 1 },
+  };
+}
+
+describe("recurring error backoff floor persistence", () => {
+  it("persists the configured retry.backoffMs floor across a real run and SQLite readback", async () => {
+    const now = Date.parse("2026-03-02T12:00:00.000Z");
+    const { storePath } = await makeStorePath();
+    const stateRoot = path.dirname(path.dirname(storePath));
+    const job = createDueRecurringJob(now);
+
+    let persistedJob: CronJob | undefined;
+    resetTaskRegistryForTests();
+    try {
+      await withEnvAsync({ OPENCLAW_STATE_DIR: stateRoot }, async () => {
+        await writeCronStoreSnapshot({ storePath, jobs: [job] });
+
+        const state = createCronServiceState({
+          storePath,
+          cronEnabled: true,
+          log: logger,
+          nowMs: () => now,
+          enqueueSystemEvent: vi.fn(),
+          requestHeartbeat: vi.fn(),
+          // Permanent (non-retryable) error -> recurring safety-net backoff
+          // floor, the branch that must honor the configured backoffMs.
+          runIsolatedAgentJob: vi.fn(async () => {
+            throw new Error("permanent: bad request");
+          }),
+          cronConfig: { retry: { backoffMs: [300_000] } },
+        });
+
+        // mode "due" (not "force") keeps preserveSchedule false, so the error
+        // path computes the safety-net backoff floor rather than preserving the
+        // recurring anchor.
+        await expect(run(state, job.id, "due")).resolves.toEqual({ ok: true, ran: true });
+
+        const persisted = (await loadCronStore(storePath)) as { jobs: CronJob[] };
+        persistedJob = persisted.jobs.find((entry) => entry.id === job.id);
+      });
+    } finally {
+      resetTaskRegistryForTests();
+    }
+
+    // The floor read back from the SQLite-backed store must be endedAt(=now) +
+    // the configured backoffMs[0], not the hardcoded 30000 default.
+    expect(persistedJob?.state.nextRunAtMs).toBe(now + 300_000);
+    expect(persistedJob?.state.lastStatus).toBe("error");
+    expect(persistedJob?.state.consecutiveErrors).toBe(1);
+  });
+});
diff --git a/src/cron/service/timer.ts b/src/cron/service/timer.ts
index ddbd2c1ce7b2..0ba13ac4195d 100644
--- a/src/cron/service/timer.ts
+++ b/src/cron/service/timer.ts
@@ -893,7 +893,10 @@ export function applyJobResult(
         }
       }
       // Apply exponential backoff for errored jobs to prevent retry storms.
-      const backoff = errorBackoffMs(job.state.consecutiveErrors ?? 1);
+      const backoff = errorBackoffMs(
+        job.state.consecutiveErrors ?? 1,
+        state.deps.cronConfig?.retry?.backoffMs ?? DEFAULT_ERROR_BACKOFF_SCHEDULE_MS,
+      );
       normalNext = computeNormalNext();
       const backoffNext = result.endedAt + backoff;
       // Use whichever is later: the natural next run or the backoff delay.