fix(telegram): prevent update offset skipping queued updates (#23284)

frankekn · obviyus · web-flow · commit e33d7fcd13d9 · 2026-02-22T15:50:33.000+05:30
Merged via /review-pr -> /prepare-pr -> /merge-pr. Prepared head SHA: 92efaf9 Co-authored-by: frankekn <4488090+frankekn@users.noreply.github.com> Co-authored-by: obviyus <22031114+obviyus@users.noreply.github.com> Reviewed-by: @obviyus
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -31,6 +31,7 @@ Docs: https://docs.openclaw.ai
 - Security/Hooks transforms: enforce symlink-safe containment for webhook transform module paths (including `hooks.transformsDir` and `hooks.mappings[].transform.module`) by resolving existing-path ancestors via realpath before import, while preserving in-root symlink support; add regression coverage for both escape and allow cases. This ships in the next npm release. Thanks @aether-ai-agent for reporting.
 - Telegram/WSL2: disable `autoSelectFamily` by default on WSL2 and memoize WSL2 detection in Telegram network decision logic to avoid repeated sync `/proc/version` probes on fetch/send paths. (#21916) Thanks @MizukiMachine.
 - Telegram/Streaming: preserve archived draft preview mapping after flush and clean superseded reasoning preview bubbles so multi-message preview finals no longer cross-edit or orphan stale messages under send/rotation races. (#23202) Thanks @obviyus.
+- Telegram/Polling: persist a safe update-offset watermark bounded by pending updates so crash/restart cannot skip queued lower `update_id` updates after out-of-order completion. (#23284) thanks @frankekn.
 - Slack/Slash commands: preserve the Bolt app receiver when registering external select options handlers so monitor startup does not crash on runtimes that require bound `app.options` calls. (#23209) Thanks @0xgaia.
 - Slack/Telegram slash sessions: await session metadata persistence before dispatch so first-turn native slash runs do not race session-origin metadata updates. (#23065) thanks @hydro13.
 - Agents/Ollama: preserve unsafe integer tool-call arguments as exact strings during NDJSON parsing, preventing large numeric IDs from being rounded before tool execution. (#23170) Thanks @BestJoester.
diff --git a/src/telegram/bot.create-telegram-bot.test.ts b/src/telegram/bot.create-telegram-bot.test.ts
@@ -445,6 +445,83 @@ describe("createTelegramBot", () => {
     });
     expect(replySpy).toHaveBeenCalledTimes(1);
   });
+
+  it("does not persist update offset past pending updates", async () => {
+    // For this test we need sequentialize(...) to behave like a normal middleware and call next().
+    sequentializeSpy.mockImplementationOnce(
+      () => async (_ctx: unknown, next: () => Promise<void>) => {
+        await next();
+      },
+    );
+
+    const onUpdateId = vi.fn();
+    loadConfig.mockReturnValue({
+      channels: { telegram: { dmPolicy: "open", allowFrom: ["*"] } },
+    });
+
+    createTelegramBot({
+      token: "tok",
+      updateOffset: {
+        lastUpdateId: 100,
+        onUpdateId,
+      },
+    });
+
+    type Middleware = (
+      ctx: Record<string, unknown>,
+      next: () => Promise<void>,
+    ) => Promise<void> | void;
+
+    const middlewares = middlewareUseSpy.mock.calls
+      .map((call) => call[0])
+      .filter((fn): fn is Middleware => typeof fn === "function");
+
+    const runMiddlewareChain = async (
+      ctx: Record<string, unknown>,
+      finalNext: () => Promise<void>,
+    ) => {
+      let idx = -1;
+      const dispatch = async (i: number): Promise<void> => {
+        if (i <= idx) {
+          throw new Error("middleware dispatch called multiple times");
+        }
+        idx = i;
+        const fn = middlewares[i];
+        if (!fn) {
+          await finalNext();
+          return;
+        }
+        await fn(ctx, async () => dispatch(i + 1));
+      };
+      await dispatch(0);
+    };
+
+    let releaseUpdate101: (() => void) | undefined;
+    const update101Gate = new Promise<void>((resolve) => {
+      releaseUpdate101 = resolve;
+    });
+
+    // Start processing update 101 but keep it pending (simulates an update queued behind sequentialize()).
+    const p101 = runMiddlewareChain({ update: { update_id: 101 } }, async () => update101Gate);
+    // Let update 101 enter the chain and mark itself pending before 102 completes.
+    await Promise.resolve();
+
+    // Complete update 102 while 101 is still pending. The persisted watermark must not jump to 102.
+    await runMiddlewareChain({ update: { update_id: 102 } }, async () => {});
+
+    const persistedValues = onUpdateId.mock.calls.map((call) => Number(call[0]));
+    const maxPersisted = persistedValues.length > 0 ? Math.max(...persistedValues) : -Infinity;
+    expect(maxPersisted).toBeLessThan(101);
+
+    releaseUpdate101?.();
+    await p101;
+
+    // Once the pending update finishes, the watermark can safely catch up.
+    const persistedAfterDrain = onUpdateId.mock.calls.map((call) => Number(call[0]));
+    const maxPersistedAfterDrain =
+      persistedAfterDrain.length > 0 ? Math.max(...persistedAfterDrain) : -Infinity;
+    expect(maxPersistedAfterDrain).toBe(102);
+  });
   it("allows distinct callback_query ids without update_id", async () => {
     loadConfig.mockReturnValue({
       channels: {
diff --git a/src/telegram/bot.ts b/src/telegram/bot.ts
@@ -148,34 +148,53 @@ export function createTelegramBot(opts: TelegramBotOptions) {
 
   const bot = new Bot(opts.token, client ? { client } : undefined);
   bot.api.config.use(apiThrottler());
-  bot.use(sequentialize(getTelegramSequentialKey));
   // Catch all errors from bot middleware to prevent unhandled rejections
   bot.catch((err) => {
     runtime.error?.(danger(`telegram bot error: ${formatUncaughtError(err)}`));
   });
 
   const recentUpdates = createTelegramUpdateDedupe();
-  let lastUpdateId =
+  const initialUpdateId =
     typeof opts.updateOffset?.lastUpdateId === "number" ? opts.updateOffset.lastUpdateId : null;
 
-  const recordUpdateId = (ctx: TelegramUpdateKeyContext) => {
-    const updateId = resolveTelegramUpdateId(ctx);
-    if (typeof updateId !== "number") {
+  // Track update_ids that have entered the middleware pipeline but have not completed yet.
+  // This includes updates that are "queued" behind sequentialize(...) for a chat/topic key.
+  // We only persist a watermark that is strictly less than the smallest pending update_id,
+  // so we never write an offset that would skip an update still waiting to run.
+  const pendingUpdateIds = new Set<number>();
+  let highestCompletedUpdateId: number | null = initialUpdateId;
+  let highestPersistedUpdateId: number | null = initialUpdateId;
+  const maybePersistSafeWatermark = () => {
+    if (typeof opts.updateOffset?.onUpdateId !== "function") {
       return;
     }
-    if (lastUpdateId !== null && updateId <= lastUpdateId) {
+    if (highestCompletedUpdateId === null) {
       return;
     }
-    lastUpdateId = updateId;
-    void opts.updateOffset?.onUpdateId?.(updateId);
+    let safe = highestCompletedUpdateId;
+    if (pendingUpdateIds.size > 0) {
+      let minPending: number | null = null;
+      for (const id of pendingUpdateIds) {
+        if (minPending === null || id < minPending) {
+          minPending = id;
+        }
+      }
+      if (minPending !== null) {
+        safe = Math.min(safe, minPending - 1);
+      }
+    }
+    if (highestPersistedUpdateId !== null && safe <= highestPersistedUpdateId) {
+      return;
+    }
+    highestPersistedUpdateId = safe;
+    void opts.updateOffset.onUpdateId(safe);
   };
 
   const shouldSkipUpdate = (ctx: TelegramUpdateKeyContext) => {
     const updateId = resolveTelegramUpdateId(ctx);
-    if (typeof updateId === "number" && lastUpdateId !== null) {
-      if (updateId <= lastUpdateId) {
-        return true;
-      }
+    const skipCutoff = highestPersistedUpdateId ?? initialUpdateId;
+    if (typeof updateId === "number" && skipCutoff !== null && updateId <= skipCutoff) {
+      return true;
     }
     const key = buildTelegramUpdateKey(ctx);
     const skipped = recentUpdates.check(key);
@@ -185,6 +204,26 @@ export function createTelegramBot(opts: TelegramBotOptions) {
     return skipped;
   };
 
+  bot.use(async (ctx, next) => {
+    const updateId = resolveTelegramUpdateId(ctx);
+    if (typeof updateId === "number") {
+      pendingUpdateIds.add(updateId);
+    }
+    try {
+      await next();
+    } finally {
+      if (typeof updateId === "number") {
+        pendingUpdateIds.delete(updateId);
+        if (highestCompletedUpdateId === null || updateId > highestCompletedUpdateId) {
+          highestCompletedUpdateId = updateId;
+        }
+        maybePersistSafeWatermark();
+      }
+    }
+  });
+
+  bot.use(sequentialize(getTelegramSequentialKey));
+
   const rawUpdateLogger = createSubsystemLogger("gateway/channels/telegram/raw-update");
   const MAX_RAW_UPDATE_CHARS = 8000;
   const MAX_RAW_UPDATE_STRING = 500;
@@ -223,7 +262,6 @@ export function createTelegramBot(opts: TelegramBotOptions) {
       }
     }
     await next();
-    recordUpdateId(ctx);
   });
 
   const historyLimit = Math.max(
