offline-transactions: a mutation function that never settles stalls the outbox, and a reconnect cannot recover it

Open
#2,441 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
30/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Active
Tech stack
node.js, typescript
Domain
backend

Research direction

The executor's runMutationFn in the offline-transactions package awaits the mutation function with no bound, and the strict FIFO only arms the next retry once the head settles. Start by reading that method and the online handler that calls executeAll(), then run the attached repro.mjs in hang mode to see the stall. Done would be an opt-in per-attempt timeout (or an AbortSignal in OfflineMutationFnParams) that turns a hung attempt into a retriable failure, with a test covering reconnect recovery.

Written by the indexing model from the issue text.

Description

  • I've validated the bug against the latest version of DB packages (@tanstack/offline-transactions 1.2.1 with @tanstack/db 1.0.0-rc.1.5; also 1.1.4 with 0.15.2)

Describe the bug
TransactionExecutor.runMutationFn awaits the mutation function with no bound, and passes it no signal. The strict FIFO only arms its next retry timer once the head settles. So when one attempt never settles, nothing after it runs, for as long as the page lives:

  • every queued transaction behind the head stays queued, and its commit() stays pending;
  • a reconnect cannot help: the online handler calls executeAll(), which returns the in-flight executionPromise while isExecuting is true;
  • the reconnect also arms deferDataRefresh on every collection with that same promise, so collection refreshes stay deferred too.

A mutation function that never settles is ordinary in the browser. fetch has no default timeout, and a connection the network accepts and never answers (a captive portal, a proxy, a server that dies mid-request, a laptop resuming from sleep) leaves its promise pending. The app sees no error, so neither the retry policy nor onlineDetector gets a chance to act.

To Reproduce
Plain Node, no browser needed (script below):

  1. Start an executor with an in-memory StorageAdapter, a leader that always wins, and an onlineDetector the script can trigger. Register one mutation function whose first call returns new Promise(() => {}) and whose later calls resolve.
  2. Commit transaction A, then transaction B.
  3. After 2.5 s: the mutation function has been called once (A). A and B are both pending, and getPendingCount() is 2.
  4. Call onlineDetector.notifyOnline(). After another 2.5 s nothing has changed: still one call, both commits pending, and the deferDataRefresh promise it armed has not settled.

Output on 1.2.1:

[   14ms] mutationFn called for A
[ 2567ms] after 2.5s: {"calls":["A"],"settle":{"A":"pending","B":"pending"},"pending":2}
[ 5070ms] 2.5s after notifyOnline(): {"calls":["A"],"settle":{"A":"pending","B":"pending"},"pending":2}
[ 5070ms] refresh barrier armed: true, settled: false

repro.mjs

// Repro: a mutation function that never settles stalls the offline outbox.
// Usage: node repro.mjs <env dir with node_modules> <mode>
//   mode "hang"     – the head's first attempt never settles
//   mode "deadline" – same, wrapped in a 500 ms race that rejects retriably
//   mode "abort"    – same, but the race rejects with an AbortError DOMException
import { pathToFileURL } from "node:url";
import path from "node:path";

const [envDir, mode = "hang"] = process.argv.slice(2);
const load = (p) =>
  import(pathToFileURL(path.join(envDir, "node_modules", p)).href);
const { createCollection } = await load("@tanstack/db/dist/esm/index.js");
const { startOfflineExecutor } = await load(
  "@tanstack/offline-transactions/dist/esm/index.js",
);

const t0 = Date.now();
const log = (...a) => console.log(`[${String(Date.now() - t0).padStart(5)}ms]`, ...a);
const sleep = (ms) => new Promise((r) => setTimeout(r, ms));

class MemoryStorage {
  store = new Map();
  async get(k) { return this.store.get(k) ?? null; }
  async set(k, v) { this.store.set(k, v); }
  async delete(k) { this.store.delete(k); }
  async keys() { return [...this.store.keys()]; }
  async clear() { this.store.clear(); }
}
class Leader {
  listeners = new Set();
  async requestLeadership() { return true; }
  releaseLeadership() {}
  isLeader() { return true; }
  onLeadershipChange(cb) { this.listeners.add(cb); return () => this.listeners.delete(cb); }
}
class Online {
  listeners = new Set();
  subscribe(cb) { this.listeners.add(cb); return () => this.listeners.delete(cb); }
  notifyOnline() { for (const cb of this.listeners) cb(); }
  isOnline() { return true; }
  dispose() {}
}

const collection = createCollection({
  id: "items",
  getKey: (item) => item.id,
  startSync: true,
  sync: { sync: ({ markReady }) => { markReady(); } },
});

const calls = [];
let first = true;
const persist = async ({ transaction }) => {
  const key = transaction.mutations[0].key;
  calls.push(key);
  log(`mutationFn called for ${key}`);
  if (first) {
    first = false;
    return new Promise(() => {}); // a request the network accepted and never answered
  }
};

const withDeadline = (fn, ms, makeError) => async (params) => {
  let timer;
  try {
    return await Promise.race([
      fn(params),
      new Promise((_, reject) => {
        timer = setTimeout(() => reject(makeError()), ms);
      }),
    ]);
  } finally {
    clearTimeout(timer);
  }
};

const mutationFn =
  mode === "deadline"
    ? withDeadline(persist, 500, () => new Error("persist attempt timed out"))
    : mode === "abort"
      ? withDeadline(persist, 500, () => new DOMException("timed out", "AbortError"))
      : persist;

const online = new Online();
const executor = startOfflineExecutor({
  collections: { items: collection },
  mutationFns: { persist: mutationFn },
  storage: new MemoryStorage(),
  leaderElection: new Leader(),
  onlineDetector: online,
  jitter: false,
});
await executor.waitForInit();
log(`offline enabled: ${executor.isOfflineEnabled}`);

const settle = {};
const write = (id) => {
  const tx = executor.createOfflineTransaction({ mutationFnName: "persist", autoCommit: false });
  tx.mutate(() => collection.insert({ id }));
  settle[id] = "pending";
  tx.commit().then(
    () => { settle[id] = "persisted"; log(`${id} persisted`); },
    (e) => { settle[id] = `rejected: ${e?.name}: ${e?.message}`; log(`${id} rejected`, e?.name, e?.message); },
  );
};

write("A");
await sleep(50);
write("B");

let refreshSettled = false;
const report = (label) =>
  log(label, JSON.stringify({ calls, settle, pending: executor.getPendingCount(), rows: [...collection.keys()] }));

await sleep(2500);
report("after 2.5s:");

// A reconnect resets retry delays and drains; it also arms the refresh barrier.
online.notifyOnline();
const barrier = collection.deferDataRefresh;
if (barrier) barrier.then(() => { refreshSettled = true; }, () => { refreshSettled = true; });
await sleep(2500);
report("2.5s after notifyOnline():");
log(`refresh barrier armed: ${barrier != null}, settled: ${refreshSettled}`);
process.exit(0);

Expected behavior
An opt-in bound on one attempt, so an app can choose to recover without wrapping every mutation function itself. Either shape would do:

  1. A per-attempt timeout on OfflineConfig (for example mutationFnTimeoutMs), after which the executor treats the attempt as a retriable failure and moves on to its backoff.
  2. An AbortSignal in OfflineMutationFnParams, which the executor aborts on its own timeout (if one is configured) and on dispose(), so a fetch can be cancelled.

I don't think we need/want a default. #1625 explains why a long mutation function is intended when it waits for a sync echo, and an opt-in keeps that contract.

Screenshots
N/A

Desktop (please complete the following information):

Additional context

My workaround (see below) wraps each mutation function in a race against a timer, and hands the attempt an AbortSignal the request carries into fetch. On expiry it aborts the signal and rejects with a retriable error, and the executor's backoff then re-sends the write under its idempotency key. Upstream support would let me drop it.

const persistWithinDeadline = async (deadlineMs, persist) => {
  const controller = new AbortController()
  // Not an AbortError: DefaultRetryPolicy would treat that as permanent.
  const expiry = new Error(`persist attempt did not settle within ${deadlineMs}ms`)
  const expired = new Promise((_, reject) => {
    controller.signal.addEventListener('abort', () => reject(expiry), { once: true })
  })
  const timer = setTimeout(() => controller.abort(expiry), deadlineMs)
  try {
    await Promise.race([persist(controller.signal), expired])
  } finally {
    clearTimeout(timer)
  }
}

Related, not duplicates: #1602 (confirm writes off the serial drain), #1625 (mutation log RFC), #2184 (a preload waiting on a persisting transaction).

Dominant language
TypeScript
Stars
4k
Forks
286
Avg merge
1d 1h
Merged PRs (30d)
104

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from TanStack/db

All issues in TanStack/db

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.